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_phys(struct device *dev, phys_addr_t phys,
516 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 (unlikely(attrs & DMA_ATTR_MMIO))
525 return ret;
526
527 if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl))))
528 goto out_fail;
529 ret = iommu_map_phys(dev, tbl, phys, size, dma_get_mask(dev),
530 direction, attrs);
531 if (unlikely(ret == DMA_MAPPING_ERROR))
532 goto out_deallocate;
533 return ret;
534
535out_deallocate:
536 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl)));
537out_fail:
538 atomic_inc(&viodev->cmo.allocs_failed);
539 return DMA_MAPPING_ERROR;
540}
541
542static void vio_dma_iommu_unmap_phys(struct device *dev, dma_addr_t dma_handle,
543 size_t size,
544 enum dma_data_direction direction,
545 unsigned long attrs)
546{
547 struct vio_dev *viodev = to_vio_dev(dev);
548 struct iommu_table *tbl = get_iommu_table_base(dev);
549
550 iommu_unmap_phys(tbl, dma_handle, size, direction, attrs);
551 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl)));
552}
553
554static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist,
555 int nelems, enum dma_data_direction direction,
556 unsigned long attrs)
557{
558 struct vio_dev *viodev = to_vio_dev(dev);
559 struct iommu_table *tbl = get_iommu_table_base(dev);
560 struct scatterlist *sgl;
561 int ret, count;
562 size_t alloc_size = 0;
563
564 for_each_sg(sglist, sgl, nelems, count)
565 alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE(tbl));
566
567 ret = vio_cmo_alloc(viodev, alloc_size);
568 if (ret)
569 goto out_fail;
570 ret = ppc_iommu_map_sg(dev, tbl, sglist, nelems, dma_get_mask(dev),
571 direction, attrs);
572 if (unlikely(!ret))
573 goto out_deallocate;
574
575 for_each_sg(sglist, sgl, ret, count)
576 alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE(tbl));
577 if (alloc_size)
578 vio_cmo_dealloc(viodev, alloc_size);
579 return ret;
580
581out_deallocate:
582 vio_cmo_dealloc(viodev, alloc_size);
583out_fail:
584 atomic_inc(&viodev->cmo.allocs_failed);
585 return ret;
586}
587
588static void vio_dma_iommu_unmap_sg(struct device *dev,
589 struct scatterlist *sglist, int nelems,
590 enum dma_data_direction direction,
591 unsigned long attrs)
592{
593 struct vio_dev *viodev = to_vio_dev(dev);
594 struct iommu_table *tbl = get_iommu_table_base(dev);
595 struct scatterlist *sgl;
596 size_t alloc_size = 0;
597 int count;
598
599 for_each_sg(sglist, sgl, nelems, count)
600 alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE(tbl));
601
602 ppc_iommu_unmap_sg(tbl, sglist, nelems, direction, attrs);
603 vio_cmo_dealloc(viodev, alloc_size);
604}
605
606static const struct dma_map_ops vio_dma_mapping_ops = {
607 .alloc = vio_dma_iommu_alloc_coherent,
608 .free = vio_dma_iommu_free_coherent,
609 .map_sg = vio_dma_iommu_map_sg,
610 .unmap_sg = vio_dma_iommu_unmap_sg,
611 .map_phys = vio_dma_iommu_map_phys,
612 .unmap_phys = vio_dma_iommu_unmap_phys,
613 .dma_supported = dma_iommu_dma_supported,
614 .get_required_mask = dma_iommu_get_required_mask,
615 .mmap = dma_common_mmap,
616 .get_sgtable = dma_common_get_sgtable,
617 .alloc_pages_op = dma_common_alloc_pages,
618 .free_pages = dma_common_free_pages,
619};
620
621/**
622 * vio_cmo_set_dev_desired - Set desired entitlement for a device
623 *
624 * @viodev: struct vio_dev for device to alter
625 * @desired: new desired entitlement level in bytes
626 *
627 * For use by devices to request a change to their entitlement at runtime or
628 * through sysfs. The desired entitlement level is changed and a balancing
629 * of system resources is scheduled to run in the future.
630 */
631void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired)
632{
633 unsigned long flags;
634 struct vio_cmo_dev_entry *dev_ent;
635 int found = 0;
636
637 if (!firmware_has_feature(FW_FEATURE_CMO))
638 return;
639
640 spin_lock_irqsave(&vio_cmo.lock, flags);
641 if (desired < VIO_CMO_MIN_ENT)
642 desired = VIO_CMO_MIN_ENT;
643
644 /*
645 * Changes will not be made for devices not in the device list.
646 * If it is not in the device list, then no driver is loaded
647 * for the device and it can not receive entitlement.
648 */
649 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
650 if (viodev == dev_ent->viodev) {
651 found = 1;
652 break;
653 }
654 if (!found) {
655 spin_unlock_irqrestore(&vio_cmo.lock, flags);
656 return;
657 }
658
659 /* Increase/decrease in desired device entitlement */
660 if (desired >= viodev->cmo.desired) {
661 /* Just bump the bus and device values prior to a balance*/
662 vio_cmo.desired += desired - viodev->cmo.desired;
663 viodev->cmo.desired = desired;
664 } else {
665 /* Decrease bus and device values for desired entitlement */
666 vio_cmo.desired -= viodev->cmo.desired - desired;
667 viodev->cmo.desired = desired;
668 /*
669 * If less entitlement is desired than current entitlement, move
670 * any reserve memory in the change region to the excess pool.
671 */
672 if (viodev->cmo.entitled > desired) {
673 vio_cmo.reserve.size -= viodev->cmo.entitled - desired;
674 vio_cmo.excess.size += viodev->cmo.entitled - desired;
675 /*
676 * If entitlement moving from the reserve pool to the
677 * excess pool is currently unused, add to the excess
678 * free counter.
679 */
680 if (viodev->cmo.allocated < viodev->cmo.entitled)
681 vio_cmo.excess.free += viodev->cmo.entitled -
682 max(viodev->cmo.allocated, desired);
683 viodev->cmo.entitled = desired;
684 }
685 }
686 schedule_delayed_work(&vio_cmo.balance_q, 0);
687 spin_unlock_irqrestore(&vio_cmo.lock, flags);
688}
689
690/**
691 * vio_cmo_bus_probe - Handle CMO specific bus probe activities
692 *
693 * @viodev - Pointer to struct vio_dev for device
694 *
695 * Determine the devices IO memory entitlement needs, attempting
696 * to satisfy the system minimum entitlement at first and scheduling
697 * a balance operation to take care of the rest at a later time.
698 *
699 * Returns: 0 on success, -EINVAL when device doesn't support CMO, and
700 * -ENOMEM when entitlement is not available for device or
701 * device entry.
702 *
703 */
704static int vio_cmo_bus_probe(struct vio_dev *viodev)
705{
706 struct vio_cmo_dev_entry *dev_ent;
707 struct device *dev = &viodev->dev;
708 struct iommu_table *tbl;
709 struct vio_driver *viodrv = to_vio_driver(dev->driver);
710 unsigned long flags;
711 size_t size;
712 bool dma_capable = false;
713
714 tbl = get_iommu_table_base(dev);
715
716 /* A device requires entitlement if it has a DMA window property */
717 switch (viodev->family) {
718 case VDEVICE:
719 if (of_get_property(viodev->dev.of_node,
720 "ibm,my-dma-window", NULL))
721 dma_capable = true;
722 break;
723 case PFO:
724 dma_capable = false;
725 break;
726 default:
727 dev_warn(dev, "unknown device family: %d\n", viodev->family);
728 BUG();
729 break;
730 }
731
732 /* Configure entitlement for the device. */
733 if (dma_capable) {
734 /* Check that the driver is CMO enabled and get desired DMA */
735 if (!viodrv->get_desired_dma) {
736 dev_err(dev, "%s: device driver does not support CMO\n",
737 __func__);
738 return -EINVAL;
739 }
740
741 viodev->cmo.desired =
742 IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev), tbl);
743 if (viodev->cmo.desired < VIO_CMO_MIN_ENT)
744 viodev->cmo.desired = VIO_CMO_MIN_ENT;
745 size = VIO_CMO_MIN_ENT;
746
747 dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry),
748 GFP_KERNEL);
749 if (!dev_ent)
750 return -ENOMEM;
751
752 dev_ent->viodev = viodev;
753 spin_lock_irqsave(&vio_cmo.lock, flags);
754 list_add(&dev_ent->list, &vio_cmo.device_list);
755 } else {
756 viodev->cmo.desired = 0;
757 size = 0;
758 spin_lock_irqsave(&vio_cmo.lock, flags);
759 }
760
761 /*
762 * If the needs for vio_cmo.min have not changed since they
763 * were last set, the number of devices in the OF tree has
764 * been constant and the IO memory for this is already in
765 * the reserve pool.
766 */
767 if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) *
768 VIO_CMO_MIN_ENT)) {
769 /* Updated desired entitlement if device requires it */
770 if (size)
771 vio_cmo.desired += (viodev->cmo.desired -
772 VIO_CMO_MIN_ENT);
773 } else {
774 size_t tmp;
775
776 tmp = vio_cmo.spare + vio_cmo.excess.free;
777 if (tmp < size) {
778 dev_err(dev, "%s: insufficient free "
779 "entitlement to add device. "
780 "Need %lu, have %lu\n", __func__,
781 size, (vio_cmo.spare + tmp));
782 spin_unlock_irqrestore(&vio_cmo.lock, flags);
783 return -ENOMEM;
784 }
785
786 /* Use excess pool first to fulfill request */
787 tmp = min(size, vio_cmo.excess.free);
788 vio_cmo.excess.free -= tmp;
789 vio_cmo.excess.size -= tmp;
790 vio_cmo.reserve.size += tmp;
791
792 /* Use spare if excess pool was insufficient */
793 vio_cmo.spare -= size - tmp;
794
795 /* Update bus accounting */
796 vio_cmo.min += size;
797 vio_cmo.desired += viodev->cmo.desired;
798 }
799 spin_unlock_irqrestore(&vio_cmo.lock, flags);
800 return 0;
801}
802
803/**
804 * vio_cmo_bus_remove - Handle CMO specific bus removal activities
805 *
806 * @viodev - Pointer to struct vio_dev for device
807 *
808 * Remove the device from the cmo device list. The minimum entitlement
809 * will be reserved for the device as long as it is in the system. The
810 * rest of the entitlement the device had been allocated will be returned
811 * to the system.
812 */
813static void vio_cmo_bus_remove(struct vio_dev *viodev)
814{
815 struct vio_cmo_dev_entry *dev_ent;
816 unsigned long flags;
817 size_t tmp;
818
819 spin_lock_irqsave(&vio_cmo.lock, flags);
820 if (viodev->cmo.allocated) {
821 dev_err(&viodev->dev, "%s: device had %lu bytes of IO "
822 "allocated after remove operation.\n",
823 __func__, viodev->cmo.allocated);
824 BUG();
825 }
826
827 /*
828 * Remove the device from the device list being maintained for
829 * CMO enabled devices.
830 */
831 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
832 if (viodev == dev_ent->viodev) {
833 list_del(&dev_ent->list);
834 kfree(dev_ent);
835 break;
836 }
837
838 /*
839 * Devices may not require any entitlement and they do not need
840 * to be processed. Otherwise, return the device's entitlement
841 * back to the pools.
842 */
843 if (viodev->cmo.entitled) {
844 /*
845 * This device has not yet left the OF tree, it's
846 * minimum entitlement remains in vio_cmo.min and
847 * vio_cmo.desired
848 */
849 vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT);
850
851 /*
852 * Save min allocation for device in reserve as long
853 * as it exists in OF tree as determined by later
854 * balance operation
855 */
856 viodev->cmo.entitled -= VIO_CMO_MIN_ENT;
857
858 /* Replenish spare from freed reserve pool */
859 if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) {
860 tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT -
861 vio_cmo.spare));
862 vio_cmo.spare += tmp;
863 viodev->cmo.entitled -= tmp;
864 }
865
866 /* Remaining reserve goes to excess pool */
867 vio_cmo.excess.size += viodev->cmo.entitled;
868 vio_cmo.excess.free += viodev->cmo.entitled;
869 vio_cmo.reserve.size -= viodev->cmo.entitled;
870
871 /*
872 * Until the device is removed it will keep a
873 * minimum entitlement; this will guarantee that
874 * a module unload/load will result in a success.
875 */
876 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
877 viodev->cmo.desired = VIO_CMO_MIN_ENT;
878 atomic_set(&viodev->cmo.allocs_failed, 0);
879 }
880
881 spin_unlock_irqrestore(&vio_cmo.lock, flags);
882}
883
884static void vio_cmo_set_dma_ops(struct vio_dev *viodev)
885{
886 set_dma_ops(&viodev->dev, &vio_dma_mapping_ops);
887}
888
889/**
890 * vio_cmo_bus_init - CMO entitlement initialization at bus init time
891 *
892 * Set up the reserve and excess entitlement pools based on available
893 * system entitlement and the number of devices in the OF tree that
894 * require entitlement in the reserve pool.
895 */
896static void vio_cmo_bus_init(void)
897{
898 struct hvcall_mpp_data mpp_data;
899 int err;
900
901 memset(&vio_cmo, 0, sizeof(struct vio_cmo));
902 spin_lock_init(&vio_cmo.lock);
903 INIT_LIST_HEAD(&vio_cmo.device_list);
904 INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance);
905
906 /* Get current system entitlement */
907 err = h_get_mpp(&mpp_data);
908
909 /*
910 * On failure, continue with entitlement set to 0, will panic()
911 * later when spare is reserved.
912 */
913 if (err != H_SUCCESS) {
914 printk(KERN_ERR "%s: unable to determine system IO "\
915 "entitlement. (%d)\n", __func__, err);
916 vio_cmo.entitled = 0;
917 } else {
918 vio_cmo.entitled = mpp_data.entitled_mem;
919 }
920
921 /* Set reservation and check against entitlement */
922 vio_cmo.spare = VIO_CMO_MIN_ENT;
923 vio_cmo.reserve.size = vio_cmo.spare;
924 vio_cmo.reserve.size += (vio_cmo_num_OF_devs() *
925 VIO_CMO_MIN_ENT);
926 if (vio_cmo.reserve.size > vio_cmo.entitled) {
927 printk(KERN_ERR "%s: insufficient system entitlement\n",
928 __func__);
929 panic("%s: Insufficient system entitlement", __func__);
930 }
931
932 /* Set the remaining accounting variables */
933 vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size;
934 vio_cmo.excess.free = vio_cmo.excess.size;
935 vio_cmo.min = vio_cmo.reserve.size;
936 vio_cmo.desired = vio_cmo.reserve.size;
937}
938
939/* sysfs device functions and data structures for CMO */
940
941#define viodev_cmo_rd_attr(name) \
942static ssize_t cmo_##name##_show(struct device *dev, \
943 struct device_attribute *attr, \
944 char *buf) \
945{ \
946 return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name); \
947}
948
949static ssize_t cmo_allocs_failed_show(struct device *dev,
950 struct device_attribute *attr, char *buf)
951{
952 struct vio_dev *viodev = to_vio_dev(dev);
953 return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed));
954}
955
956static ssize_t cmo_allocs_failed_store(struct device *dev,
957 struct device_attribute *attr, const char *buf, size_t count)
958{
959 struct vio_dev *viodev = to_vio_dev(dev);
960 atomic_set(&viodev->cmo.allocs_failed, 0);
961 return count;
962}
963
964static ssize_t cmo_desired_store(struct device *dev,
965 struct device_attribute *attr, const char *buf, size_t count)
966{
967 struct vio_dev *viodev = to_vio_dev(dev);
968 size_t new_desired;
969 int ret;
970
971 ret = kstrtoul(buf, 10, &new_desired);
972 if (ret)
973 return ret;
974
975 vio_cmo_set_dev_desired(viodev, new_desired);
976 return count;
977}
978
979viodev_cmo_rd_attr(desired);
980viodev_cmo_rd_attr(entitled);
981viodev_cmo_rd_attr(allocated);
982
983static ssize_t name_show(struct device *, struct device_attribute *, char *);
984static ssize_t devspec_show(struct device *, struct device_attribute *, char *);
985static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
986 char *buf);
987
988static struct device_attribute dev_attr_name;
989static struct device_attribute dev_attr_devspec;
990static struct device_attribute dev_attr_modalias;
991
992static DEVICE_ATTR_RO(cmo_entitled);
993static DEVICE_ATTR_RO(cmo_allocated);
994static DEVICE_ATTR_RW(cmo_desired);
995static DEVICE_ATTR_RW(cmo_allocs_failed);
996
997/* sysfs bus functions and data structures for CMO */
998
999#define viobus_cmo_rd_attr(name) \
1000static ssize_t cmo_bus_##name##_show(const struct bus_type *bt, char *buf) \
1001{ \
1002 return sprintf(buf, "%lu\n", vio_cmo.name); \
1003} \
1004static struct bus_attribute bus_attr_cmo_bus_##name = \
1005 __ATTR(cmo_##name, S_IRUGO, cmo_bus_##name##_show, NULL)
1006
1007#define viobus_cmo_pool_rd_attr(name, var) \
1008static ssize_t \
1009cmo_##name##_##var##_show(const struct bus_type *bt, char *buf) \
1010{ \
1011 return sprintf(buf, "%lu\n", vio_cmo.name.var); \
1012} \
1013static BUS_ATTR_RO(cmo_##name##_##var)
1014
1015viobus_cmo_rd_attr(entitled);
1016viobus_cmo_rd_attr(spare);
1017viobus_cmo_rd_attr(min);
1018viobus_cmo_rd_attr(desired);
1019viobus_cmo_rd_attr(curr);
1020viobus_cmo_pool_rd_attr(reserve, size);
1021viobus_cmo_pool_rd_attr(excess, size);
1022viobus_cmo_pool_rd_attr(excess, free);
1023
1024static ssize_t cmo_high_show(const struct bus_type *bt, char *buf)
1025{
1026 return sprintf(buf, "%lu\n", vio_cmo.high);
1027}
1028
1029static ssize_t cmo_high_store(const struct bus_type *bt, const char *buf,
1030 size_t count)
1031{
1032 unsigned long flags;
1033
1034 spin_lock_irqsave(&vio_cmo.lock, flags);
1035 vio_cmo.high = vio_cmo.curr;
1036 spin_unlock_irqrestore(&vio_cmo.lock, flags);
1037
1038 return count;
1039}
1040static BUS_ATTR_RW(cmo_high);
1041
1042static struct attribute *vio_bus_attrs[] = {
1043 &bus_attr_cmo_bus_entitled.attr,
1044 &bus_attr_cmo_bus_spare.attr,
1045 &bus_attr_cmo_bus_min.attr,
1046 &bus_attr_cmo_bus_desired.attr,
1047 &bus_attr_cmo_bus_curr.attr,
1048 &bus_attr_cmo_high.attr,
1049 &bus_attr_cmo_reserve_size.attr,
1050 &bus_attr_cmo_excess_size.attr,
1051 &bus_attr_cmo_excess_free.attr,
1052 NULL,
1053};
1054ATTRIBUTE_GROUPS(vio_bus);
1055
1056static void __init vio_cmo_sysfs_init(void) { }
1057#else /* CONFIG_PPC_SMLPAR */
1058int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; }
1059void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {}
1060static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; }
1061static void vio_cmo_bus_remove(struct vio_dev *viodev) {}
1062static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {}
1063static void vio_cmo_bus_init(void) {}
1064static void __init vio_cmo_sysfs_init(void) { }
1065#endif /* CONFIG_PPC_SMLPAR */
1066EXPORT_SYMBOL(vio_cmo_entitlement_update);
1067EXPORT_SYMBOL(vio_cmo_set_dev_desired);
1068
1069
1070/*
1071 * Platform Facilities Option (PFO) support
1072 */
1073
1074/**
1075 * vio_h_cop_sync - Perform a synchronous PFO co-processor operation
1076 *
1077 * @vdev - Pointer to a struct vio_dev for device
1078 * @op - Pointer to a struct vio_pfo_op for the operation parameters
1079 *
1080 * Calls the hypervisor to synchronously perform the PFO operation
1081 * described in @op. In the case of a busy response from the hypervisor,
1082 * the operation will be re-submitted indefinitely unless a non-zero timeout
1083 * is specified or an error occurs. The timeout places a limit on when to
1084 * stop re-submitting a operation, the total time can be exceeded if an
1085 * operation is in progress.
1086 *
1087 * If op->hcall_ret is not NULL, this will be set to the return from the
1088 * last h_cop_op call or it will be 0 if an error not involving the h_call
1089 * was encountered.
1090 *
1091 * Returns:
1092 * 0 on success,
1093 * -EINVAL if the h_call fails due to an invalid parameter,
1094 * -E2BIG if the h_call can not be performed synchronously,
1095 * -EBUSY if a timeout is specified and has elapsed,
1096 * -EACCES if the memory area for data/status has been rescinded, or
1097 * -EPERM if a hardware fault has been indicated
1098 */
1099int vio_h_cop_sync(struct vio_dev *vdev, struct vio_pfo_op *op)
1100{
1101 struct device *dev = &vdev->dev;
1102 unsigned long deadline = 0;
1103 long hret = 0;
1104 int ret = 0;
1105
1106 if (op->timeout)
1107 deadline = jiffies + msecs_to_jiffies(m: op->timeout);
1108
1109 while (true) {
1110 hret = plpar_hcall_norets(H_COP, op->flags,
1111 vdev->resource_id,
1112 op->in, op->inlen, op->out,
1113 op->outlen, op->csbcpb);
1114
1115 if (hret == H_SUCCESS ||
1116 (hret != H_NOT_ENOUGH_RESOURCES &&
1117 hret != H_BUSY && hret != H_RESOURCE) ||
1118 (op->timeout && time_after(deadline, jiffies)))
1119 break;
1120
1121 dev_dbg(dev, "%s: hcall ret(%ld), retrying.\n", __func__, hret);
1122 }
1123
1124 switch (hret) {
1125 case H_SUCCESS:
1126 ret = 0;
1127 break;
1128 case H_OP_MODE:
1129 case H_TOO_BIG:
1130 ret = -E2BIG;
1131 break;
1132 case H_RESCINDED:
1133 ret = -EACCES;
1134 break;
1135 case H_HARDWARE:
1136 ret = -EPERM;
1137 break;
1138 case H_NOT_ENOUGH_RESOURCES:
1139 case H_RESOURCE:
1140 case H_BUSY:
1141 ret = -EBUSY;
1142 break;
1143 default:
1144 ret = -EINVAL;
1145 break;
1146 }
1147
1148 if (ret)
1149 dev_dbg(dev, "%s: Sync h_cop_op failure (ret:%d) (hret:%ld)\n",
1150 __func__, ret, hret);
1151
1152 op->hcall_err = hret;
1153 return ret;
1154}
1155EXPORT_SYMBOL(vio_h_cop_sync);
1156
1157static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
1158{
1159 const __be32 *dma_window;
1160 struct iommu_table *tbl;
1161 unsigned long offset, size;
1162
1163 dma_window = of_get_property(node: dev->dev.of_node,
1164 name: "ibm,my-dma-window", NULL);
1165 if (!dma_window)
1166 return NULL;
1167
1168 tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
1169 if (tbl == NULL)
1170 return NULL;
1171
1172 kref_init(kref: &tbl->it_kref);
1173
1174 of_parse_dma_window(dev->dev.of_node, dma_window,
1175 &tbl->it_index, &offset, &size);
1176
1177 /* TCE table size - measured in tce entries */
1178 tbl->it_page_shift = IOMMU_PAGE_SHIFT_4K;
1179 tbl->it_size = size >> tbl->it_page_shift;
1180 /* offset for VIO should always be 0 */
1181 tbl->it_offset = offset >> tbl->it_page_shift;
1182 tbl->it_busno = 0;
1183 tbl->it_type = TCE_VB;
1184 tbl->it_blocksize = 16;
1185
1186 if (firmware_has_feature(FW_FEATURE_LPAR))
1187 tbl->it_ops = &iommu_table_lpar_multi_ops;
1188 else
1189 tbl->it_ops = &iommu_table_pseries_ops;
1190
1191 return iommu_init_table(tbl, -1, 0, 0);
1192}
1193
1194/**
1195 * vio_match_device: - Tell if a VIO device has a matching
1196 * VIO device id structure.
1197 * @ids: array of VIO device id structures to search in
1198 * @dev: the VIO device structure to match against
1199 *
1200 * Used by a driver to check whether a VIO device present in the
1201 * system is in its list of supported devices. Returns the matching
1202 * vio_device_id structure or NULL if there is no match.
1203 */
1204static const struct vio_device_id *vio_match_device(
1205 const struct vio_device_id *ids, const struct vio_dev *dev)
1206{
1207 while (ids->type[0] != '\0') {
1208 if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) &&
1209 of_device_is_compatible(device: dev->dev.of_node,
1210 ids->compat))
1211 return ids;
1212 ids++;
1213 }
1214 return NULL;
1215}
1216
1217/*
1218 * Convert from struct device to struct vio_dev and pass to driver.
1219 * dev->driver has already been set by generic code because vio_bus_match
1220 * succeeded.
1221 */
1222static int vio_bus_probe(struct device *dev)
1223{
1224 struct vio_dev *viodev = to_vio_dev(dev);
1225 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1226 const struct vio_device_id *id;
1227 int error = -ENODEV;
1228
1229 if (!viodrv->probe)
1230 return error;
1231
1232 id = vio_match_device(ids: viodrv->id_table, dev: viodev);
1233 if (id) {
1234 memset(&viodev->cmo, 0, sizeof(viodev->cmo));
1235 if (firmware_has_feature(FW_FEATURE_CMO)) {
1236 error = vio_cmo_bus_probe(viodev);
1237 if (error)
1238 return error;
1239 }
1240 error = viodrv->probe(viodev, id);
1241 if (error && firmware_has_feature(FW_FEATURE_CMO))
1242 vio_cmo_bus_remove(viodev);
1243 }
1244
1245 return error;
1246}
1247
1248/* convert from struct device to struct vio_dev and pass to driver. */
1249static void vio_bus_remove(struct device *dev)
1250{
1251 struct vio_dev *viodev = to_vio_dev(dev);
1252 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1253 struct device *devptr;
1254
1255 /*
1256 * Hold a reference to the device after the remove function is called
1257 * to allow for CMO accounting cleanup for the device.
1258 */
1259 devptr = get_device(dev);
1260
1261 if (viodrv->remove)
1262 viodrv->remove(viodev);
1263
1264 if (firmware_has_feature(FW_FEATURE_CMO))
1265 vio_cmo_bus_remove(viodev);
1266
1267 put_device(dev: devptr);
1268}
1269
1270static void vio_bus_shutdown(struct device *dev)
1271{
1272 struct vio_dev *viodev = to_vio_dev(dev);
1273 struct vio_driver *viodrv;
1274
1275 if (dev->driver) {
1276 viodrv = to_vio_driver(dev->driver);
1277 if (viodrv->shutdown)
1278 viodrv->shutdown(viodev);
1279 else if (kexec_in_progress)
1280 vio_bus_remove(dev);
1281 }
1282}
1283
1284/**
1285 * vio_register_driver: - Register a new vio driver
1286 * @viodrv: The vio_driver structure to be registered.
1287 */
1288int __vio_register_driver(struct vio_driver *viodrv, struct module *owner,
1289 const char *mod_name)
1290{
1291 // vio_bus_type is only initialised for pseries
1292 if (!machine_is(pseries))
1293 return -ENODEV;
1294
1295 pr_debug("%s: driver %s registering\n", __func__, viodrv->name);
1296
1297 /* fill in 'struct driver' fields */
1298 viodrv->driver.name = viodrv->name;
1299 viodrv->driver.pm = viodrv->pm;
1300 viodrv->driver.bus = &vio_bus_type;
1301 viodrv->driver.owner = owner;
1302 viodrv->driver.mod_name = mod_name;
1303
1304 return driver_register(drv: &viodrv->driver);
1305}
1306EXPORT_SYMBOL(__vio_register_driver);
1307
1308/**
1309 * vio_unregister_driver - Remove registration of vio driver.
1310 * @viodrv: The vio_driver struct to be removed form registration
1311 */
1312void vio_unregister_driver(struct vio_driver *viodrv)
1313{
1314 driver_unregister(drv: &viodrv->driver);
1315}
1316EXPORT_SYMBOL(vio_unregister_driver);
1317
1318/* vio_dev refcount hit 0 */
1319static void vio_dev_release(struct device *dev)
1320{
1321 struct iommu_table *tbl = get_iommu_table_base(dev);
1322
1323 if (tbl)
1324 iommu_tce_table_put(tbl);
1325 of_node_put(node: dev->of_node);
1326 kfree(objp: to_vio_dev(dev));
1327}
1328
1329/**
1330 * vio_register_device_node: - Register a new vio device.
1331 * @of_node: The OF node for this device.
1332 *
1333 * Creates and initializes a vio_dev structure from the data in
1334 * of_node and adds it to the list of virtual devices.
1335 * Returns a pointer to the created vio_dev or NULL if node has
1336 * NULL device_type or compatible fields.
1337 */
1338struct vio_dev *vio_register_device_node(struct device_node *of_node)
1339{
1340 struct vio_dev *viodev;
1341 struct device_node *parent_node;
1342 const __be32 *prop;
1343 enum vio_dev_family family;
1344
1345 /*
1346 * Determine if this node is a under the /vdevice node or under the
1347 * /ibm,platform-facilities node. This decides the device's family.
1348 */
1349 parent_node = of_get_parent(node: of_node);
1350 if (parent_node) {
1351 if (of_node_is_type(np: parent_node, type: "ibm,platform-facilities"))
1352 family = PFO;
1353 else if (of_node_is_type(np: parent_node, type: "vdevice"))
1354 family = VDEVICE;
1355 else {
1356 pr_warn("%s: parent(%pOF) of %pOFn not recognized.\n",
1357 __func__,
1358 parent_node,
1359 of_node);
1360 of_node_put(node: parent_node);
1361 return NULL;
1362 }
1363 of_node_put(node: parent_node);
1364 } else {
1365 pr_warn("%s: could not determine the parent of node %pOFn.\n",
1366 __func__, of_node);
1367 return NULL;
1368 }
1369
1370 if (family == PFO) {
1371 if (of_property_read_bool(np: of_node, propname: "interrupt-controller")) {
1372 pr_debug("%s: Skipping the interrupt controller %pOFn.\n",
1373 __func__, of_node);
1374 return NULL;
1375 }
1376 }
1377
1378 /* allocate a vio_dev for this node */
1379 viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
1380 if (viodev == NULL) {
1381 pr_warn("%s: allocation failure for VIO device.\n", __func__);
1382 return NULL;
1383 }
1384
1385 /* we need the 'device_type' property, in order to match with drivers */
1386 viodev->family = family;
1387 if (viodev->family == VDEVICE) {
1388 unsigned int unit_address;
1389
1390 viodev->type = of_node_get_device_type(np: of_node);
1391 if (!viodev->type) {
1392 pr_warn("%s: node %pOFn is missing the 'device_type' "
1393 "property.\n", __func__, of_node);
1394 goto out;
1395 }
1396
1397 prop = of_get_property(node: of_node, name: "reg", NULL);
1398 if (prop == NULL) {
1399 pr_warn("%s: node %pOFn missing 'reg'\n",
1400 __func__, of_node);
1401 goto out;
1402 }
1403 unit_address = of_read_number(cell: prop, size: 1);
1404 dev_set_name(dev: &viodev->dev, name: "%x", unit_address);
1405 viodev->irq = irq_of_parse_and_map(node: of_node, index: 0);
1406 viodev->unit_address = unit_address;
1407 } else {
1408 /* PFO devices need their resource_id for submitting COP_OPs
1409 * This is an optional field for devices, but is required when
1410 * performing synchronous ops */
1411 prop = of_get_property(node: of_node, name: "ibm,resource-id", NULL);
1412 if (prop != NULL)
1413 viodev->resource_id = of_read_number(cell: prop, size: 1);
1414
1415 dev_set_name(dev: &viodev->dev, name: "%pOFn", of_node);
1416 viodev->type = dev_name(dev: &viodev->dev);
1417 viodev->irq = 0;
1418 }
1419
1420 viodev->name = of_node->name;
1421 viodev->dev.of_node = of_node_get(node: of_node);
1422
1423 set_dev_node(dev: &viodev->dev, node: of_node_to_nid(np: of_node));
1424
1425 /* init generic 'struct device' fields: */
1426 viodev->dev.parent = &vio_bus_device.dev;
1427 viodev->dev.bus = &vio_bus_type;
1428 viodev->dev.release = vio_dev_release;
1429
1430 if (of_property_present(np: viodev->dev.of_node, propname: "ibm,my-dma-window")) {
1431 if (firmware_has_feature(FW_FEATURE_CMO))
1432 vio_cmo_set_dma_ops(viodev);
1433 else
1434 set_dma_ops(dev: &viodev->dev, dma_ops: &dma_iommu_ops);
1435
1436 set_iommu_table_base(&viodev->dev,
1437 vio_build_iommu_table(dev: viodev));
1438
1439 /* needed to ensure proper operation of coherent allocations
1440 * later, in case driver doesn't set it explicitly */
1441 viodev->dev.coherent_dma_mask = DMA_BIT_MASK(64);
1442 viodev->dev.dma_mask = &viodev->dev.coherent_dma_mask;
1443 }
1444
1445 /* register with generic device framework */
1446 if (device_register(dev: &viodev->dev)) {
1447 printk(KERN_ERR "%s: failed to register device %s\n",
1448 __func__, dev_name(&viodev->dev));
1449 put_device(dev: &viodev->dev);
1450 return NULL;
1451 }
1452
1453 return viodev;
1454
1455out: /* Use this exit point for any return prior to device_register */
1456 kfree(objp: viodev);
1457
1458 return NULL;
1459}
1460EXPORT_SYMBOL(vio_register_device_node);
1461
1462/*
1463 * vio_bus_scan_for_devices - Scan OF and register each child device
1464 * @root_name - OF node name for the root of the subtree to search.
1465 * This must be non-NULL
1466 *
1467 * Starting from the root node provide, register the device node for
1468 * each child beneath the root.
1469 */
1470static void __init vio_bus_scan_register_devices(char *root_name)
1471{
1472 struct device_node *node_root, *node_child;
1473
1474 if (!root_name)
1475 return;
1476
1477 node_root = of_find_node_by_name(NULL, name: root_name);
1478 if (node_root) {
1479
1480 /*
1481 * Create struct vio_devices for each virtual device in
1482 * the device tree. Drivers will associate with them later.
1483 */
1484 node_child = of_get_next_child(node: node_root, NULL);
1485 while (node_child) {
1486 vio_register_device_node(node_child);
1487 node_child = of_get_next_child(node: node_root, prev: node_child);
1488 }
1489 of_node_put(node: node_root);
1490 }
1491}
1492
1493/**
1494 * vio_bus_init: - Initialize the virtual IO bus
1495 */
1496static int __init vio_bus_init(void)
1497{
1498 int err;
1499
1500 if (firmware_has_feature(FW_FEATURE_CMO))
1501 vio_cmo_sysfs_init();
1502
1503 err = bus_register(bus: &vio_bus_type);
1504 if (err) {
1505 printk(KERN_ERR "failed to register VIO bus\n");
1506 return err;
1507 }
1508
1509 /*
1510 * The fake parent of all vio devices, just to give us
1511 * a nice directory
1512 */
1513 err = device_register(dev: &vio_bus_device.dev);
1514 if (err) {
1515 printk(KERN_WARNING "%s: device_register returned %i\n",
1516 __func__, err);
1517 return err;
1518 }
1519
1520 if (firmware_has_feature(FW_FEATURE_CMO))
1521 vio_cmo_bus_init();
1522
1523 return 0;
1524}
1525machine_postcore_initcall(pseries, vio_bus_init);
1526
1527static int __init vio_device_init(void)
1528{
1529 vio_bus_scan_register_devices(root_name: "vdevice");
1530 vio_bus_scan_register_devices(root_name: "ibm,platform-facilities");
1531
1532 return 0;
1533}
1534machine_device_initcall(pseries, vio_device_init);
1535
1536static ssize_t name_show(struct device *dev,
1537 struct device_attribute *attr, char *buf)
1538{
1539 return sprintf(buf, fmt: "%s\n", to_vio_dev(dev)->name);
1540}
1541static DEVICE_ATTR_RO(name);
1542
1543static ssize_t devspec_show(struct device *dev,
1544 struct device_attribute *attr, char *buf)
1545{
1546 struct device_node *of_node = dev->of_node;
1547
1548 return sprintf(buf, fmt: "%pOF\n", of_node);
1549}
1550static DEVICE_ATTR_RO(devspec);
1551
1552static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1553 char *buf)
1554{
1555 const struct vio_dev *vio_dev = to_vio_dev(dev);
1556 struct device_node *dn;
1557 const char *cp;
1558
1559 dn = dev->of_node;
1560 if (!dn) {
1561 strcpy(p: buf, q: "\n");
1562 return strlen(buf);
1563 }
1564 cp = of_get_property(node: dn, name: "compatible", NULL);
1565 if (!cp) {
1566 strcpy(p: buf, q: "\n");
1567 return strlen(buf);
1568 }
1569
1570 return sprintf(buf, fmt: "vio:T%sS%s\n", vio_dev->type, cp);
1571}
1572static DEVICE_ATTR_RO(modalias);
1573
1574void vio_unregister_device(struct vio_dev *viodev)
1575{
1576 device_unregister(dev: &viodev->dev);
1577 if (viodev->family == VDEVICE)
1578 irq_dispose_mapping(virq: viodev->irq);
1579}
1580EXPORT_SYMBOL(vio_unregister_device);
1581
1582static int vio_bus_match(struct device *dev, const struct device_driver *drv)
1583{
1584 const struct vio_dev *vio_dev = to_vio_dev(dev);
1585 const struct vio_driver *vio_drv = to_vio_driver(drv);
1586 const struct vio_device_id *ids = vio_drv->id_table;
1587
1588 return (ids != NULL) && (vio_match_device(ids, dev: vio_dev) != NULL);
1589}
1590
1591static int vio_hotplug(const struct device *dev, struct kobj_uevent_env *env)
1592{
1593 const struct vio_dev *vio_dev = to_vio_dev(dev);
1594 const struct device_node *dn;
1595 const char *cp;
1596
1597 dn = dev->of_node;
1598 if (dn && (cp = of_get_property(node: dn, name: "compatible", NULL)))
1599 add_uevent_var(env, format: "MODALIAS=vio:T%sS%s", vio_dev->type, cp);
1600
1601 return 0;
1602}
1603
1604#ifdef CONFIG_PPC_SMLPAR
1605static struct attribute *vio_cmo_dev_attrs[] = {
1606 &dev_attr_name.attr,
1607 &dev_attr_devspec.attr,
1608 &dev_attr_modalias.attr,
1609 &dev_attr_cmo_entitled.attr,
1610 &dev_attr_cmo_allocated.attr,
1611 &dev_attr_cmo_desired.attr,
1612 &dev_attr_cmo_allocs_failed.attr,
1613 NULL,
1614};
1615ATTRIBUTE_GROUPS(vio_cmo_dev);
1616
1617const struct bus_type vio_bus_type = {
1618 .name = "vio",
1619 .dev_groups = vio_cmo_dev_groups,
1620 .bus_groups = vio_bus_groups,
1621 .uevent = vio_hotplug,
1622 .match = vio_bus_match,
1623 .probe = vio_bus_probe,
1624 .remove = vio_bus_remove,
1625 .shutdown = vio_bus_shutdown,
1626};
1627#else /* CONFIG_PPC_SMLPAR */
1628static struct attribute *vio_dev_attrs[] = {
1629 &dev_attr_name.attr,
1630 &dev_attr_devspec.attr,
1631 &dev_attr_modalias.attr,
1632 NULL,
1633};
1634ATTRIBUTE_GROUPS(vio_dev);
1635
1636const struct bus_type vio_bus_type = {
1637 .name = "vio",
1638 .dev_groups = vio_dev_groups,
1639 .uevent = vio_hotplug,
1640 .match = vio_bus_match,
1641 .probe = vio_bus_probe,
1642 .remove = vio_bus_remove,
1643 .shutdown = vio_bus_shutdown,
1644};
1645#endif /* CONFIG_PPC_SMLPAR */
1646
1647/**
1648 * vio_get_attribute: - get attribute for virtual device
1649 * @vdev: The vio device to get property.
1650 * @which: The property/attribute to be extracted.
1651 * @length: Pointer to length of returned data size (unused if NULL).
1652 *
1653 * Calls prom.c's of_get_property() to return the value of the
1654 * attribute specified by @which
1655*/
1656const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length)
1657{
1658 return of_get_property(node: vdev->dev.of_node, name: which, lenp: length);
1659}
1660EXPORT_SYMBOL(vio_get_attribute);
1661
1662/* vio_find_name() - internal because only vio.c knows how we formatted the
1663 * kobject name
1664 */
1665static struct vio_dev *vio_find_name(const char *name)
1666{
1667 struct device *found;
1668
1669 found = bus_find_device_by_name(bus: &vio_bus_type, NULL, name);
1670 if (!found)
1671 return NULL;
1672
1673 return to_vio_dev(found);
1674}
1675
1676/**
1677 * vio_find_node - find an already-registered vio_dev
1678 * @vnode: device_node of the virtual device we're looking for
1679 *
1680 * Takes a reference to the embedded struct device which needs to be dropped
1681 * after use.
1682 */
1683struct vio_dev *vio_find_node(struct device_node *vnode)
1684{
1685 char kobj_name[20];
1686 struct device_node *vnode_parent;
1687
1688 vnode_parent = of_get_parent(node: vnode);
1689 if (!vnode_parent)
1690 return NULL;
1691
1692 /* construct the kobject name from the device node */
1693 if (of_node_is_type(np: vnode_parent, type: "vdevice")) {
1694 const __be32 *prop;
1695
1696 prop = of_get_property(node: vnode, name: "reg", NULL);
1697 if (!prop)
1698 goto out;
1699 snprintf(buf: kobj_name, size: sizeof(kobj_name), fmt: "%x",
1700 (uint32_t)of_read_number(cell: prop, size: 1));
1701 } else if (of_node_is_type(np: vnode_parent, type: "ibm,platform-facilities"))
1702 snprintf(buf: kobj_name, size: sizeof(kobj_name), fmt: "%pOFn", vnode);
1703 else
1704 goto out;
1705
1706 of_node_put(node: vnode_parent);
1707 return vio_find_name(name: kobj_name);
1708out:
1709 of_node_put(node: vnode_parent);
1710 return NULL;
1711}
1712EXPORT_SYMBOL(vio_find_node);
1713
1714int vio_enable_interrupts(struct vio_dev *dev)
1715{
1716 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE);
1717 if (rc != H_SUCCESS)
1718 printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc);
1719 return rc;
1720}
1721EXPORT_SYMBOL(vio_enable_interrupts);
1722
1723int vio_disable_interrupts(struct vio_dev *dev)
1724{
1725 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE);
1726 if (rc != H_SUCCESS)
1727 printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc);
1728 return rc;
1729}
1730EXPORT_SYMBOL(vio_disable_interrupts);
1731
1732static int __init vio_init(void)
1733{
1734 dma_debug_add_bus(bus: &vio_bus_type);
1735 return 0;
1736}
1737machine_fs_initcall(pseries, vio_init);
1738

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