2 * IBM PowerPC Virtual I/O Infrastructure Support.
4 * Copyright (c) 2003,2008 IBM Corp.
5 * Dave Engebretsen engebret@us.ibm.com
6 * Santiago Leon santil@us.ibm.com
7 * Hollis Blanchard <hollisb@us.ibm.com>
9 * Robert Jennings <rcjenn@us.ibm.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
17 #include <linux/types.h>
18 #include <linux/device.h>
19 #include <linux/init.h>
20 #include <linux/console.h>
21 #include <linux/module.h>
23 #include <linux/dma-mapping.h>
24 #include <linux/kobject.h>
26 #include <asm/iommu.h>
30 #include <asm/firmware.h>
32 #include <asm/abs_addr.h>
34 #include <asm/hvcall.h>
35 #include <asm/iseries/vio.h>
36 #include <asm/iseries/hv_types.h>
37 #include <asm/iseries/hv_lp_config.h>
38 #include <asm/iseries/hv_call_xm.h>
39 #include <asm/iseries/iommu.h>
41 static struct bus_type vio_bus_type;
43 static struct vio_dev vio_bus_device = { /* fake "parent" device */
44 .name = vio_bus_device.dev.bus_id,
47 .dev.bus = &vio_bus_type,
50 #ifdef CONFIG_PPC_SMLPAR
52 * vio_cmo_pool - A pool of IO memory for CMO use
54 * @size: The size of the pool in bytes
55 * @free: The amount of free memory in the pool
62 /* How many ms to delay queued balance work */
63 #define VIO_CMO_BALANCE_DELAY 100
65 /* Portion out IO memory to CMO devices by this chunk size */
66 #define VIO_CMO_BALANCE_CHUNK 131072
69 * vio_cmo_dev_entry - A device that is CMO-enabled and requires entitlement
71 * @vio_dev: struct vio_dev pointer
72 * @list: pointer to other devices on bus that are being tracked
74 struct vio_cmo_dev_entry {
75 struct vio_dev *viodev;
76 struct list_head list;
80 * vio_cmo - VIO bus accounting structure for CMO entitlement
82 * @lock: spinlock for entire structure
83 * @balance_q: work queue for balancing system entitlement
84 * @device_list: list of CMO-enabled devices requiring entitlement
85 * @entitled: total system entitlement in bytes
86 * @reserve: pool of memory from which devices reserve entitlement, incl. spare
87 * @excess: pool of excess entitlement not needed for device reserves or spare
88 * @spare: IO memory for device hotplug functionality
89 * @min: minimum necessary for system operation
90 * @desired: desired memory for system operation
91 * @curr: bytes currently allocated
92 * @high: high water mark for IO data usage
96 struct delayed_work balance_q;
97 struct list_head device_list;
99 struct vio_cmo_pool reserve;
100 struct vio_cmo_pool excess;
109 * vio_cmo_OF_devices - Count the number of OF devices that have DMA windows
111 static int vio_cmo_num_OF_devs(void)
113 struct device_node *node_vroot;
117 * Count the number of vdevice entries with an
118 * ibm,my-dma-window OF property
120 node_vroot = of_find_node_by_name(NULL, "vdevice");
122 struct device_node *of_node;
123 struct property *prop;
125 for_each_child_of_node(node_vroot, of_node) {
126 prop = of_find_property(of_node, "ibm,my-dma-window",
132 of_node_put(node_vroot);
137 * vio_cmo_alloc - allocate IO memory for CMO-enable devices
139 * @viodev: VIO device requesting IO memory
140 * @size: size of allocation requested
142 * Allocations come from memory reserved for the devices and any excess
143 * IO memory available to all devices. The spare pool used to service
144 * hotplug must be equal to %VIO_CMO_MIN_ENT for the excess pool to be
148 * 0 for successful allocation and -ENOMEM for a failure
150 static inline int vio_cmo_alloc(struct vio_dev *viodev, size_t size)
153 size_t reserve_free = 0;
154 size_t excess_free = 0;
157 spin_lock_irqsave(&vio_cmo.lock, flags);
159 /* Determine the amount of free entitlement available in reserve */
160 if (viodev->cmo.entitled > viodev->cmo.allocated)
161 reserve_free = viodev->cmo.entitled - viodev->cmo.allocated;
163 /* If spare is not fulfilled, the excess pool can not be used. */
164 if (vio_cmo.spare >= VIO_CMO_MIN_ENT)
165 excess_free = vio_cmo.excess.free;
167 /* The request can be satisfied */
168 if ((reserve_free + excess_free) >= size) {
169 vio_cmo.curr += size;
170 if (vio_cmo.curr > vio_cmo.high)
171 vio_cmo.high = vio_cmo.curr;
172 viodev->cmo.allocated += size;
173 size -= min(reserve_free, size);
174 vio_cmo.excess.free -= size;
178 spin_unlock_irqrestore(&vio_cmo.lock, flags);
183 * vio_cmo_dealloc - deallocate IO memory from CMO-enable devices
184 * @viodev: VIO device freeing IO memory
185 * @size: size of deallocation
187 * IO memory is freed by the device back to the correct memory pools.
188 * The spare pool is replenished first from either memory pool, then
189 * the reserve pool is used to reduce device entitlement, the excess
190 * pool is used to increase the reserve pool toward the desired entitlement
191 * target, and then the remaining memory is returned to the pools.
194 static inline void vio_cmo_dealloc(struct vio_dev *viodev, size_t size)
197 size_t spare_needed = 0;
198 size_t excess_freed = 0;
199 size_t reserve_freed = size;
203 spin_lock_irqsave(&vio_cmo.lock, flags);
204 vio_cmo.curr -= size;
206 /* Amount of memory freed from the excess pool */
207 if (viodev->cmo.allocated > viodev->cmo.entitled) {
208 excess_freed = min(reserve_freed, (viodev->cmo.allocated -
209 viodev->cmo.entitled));
210 reserve_freed -= excess_freed;
213 /* Remove allocation from device */
214 viodev->cmo.allocated -= (reserve_freed + excess_freed);
216 /* Spare is a subset of the reserve pool, replenish it first. */
217 spare_needed = VIO_CMO_MIN_ENT - vio_cmo.spare;
220 * Replenish the spare in the reserve pool from the excess pool.
221 * This moves entitlement into the reserve pool.
223 if (spare_needed && excess_freed) {
224 tmp = min(excess_freed, spare_needed);
225 vio_cmo.excess.size -= tmp;
226 vio_cmo.reserve.size += tmp;
227 vio_cmo.spare += tmp;
234 * Replenish the spare in the reserve pool from the reserve pool.
235 * This removes entitlement from the device down to VIO_CMO_MIN_ENT,
236 * if needed, and gives it to the spare pool. The amount of used
237 * memory in this pool does not change.
239 if (spare_needed && reserve_freed) {
240 tmp = min(spare_needed, min(reserve_freed,
241 (viodev->cmo.entitled -
244 vio_cmo.spare += tmp;
245 viodev->cmo.entitled -= tmp;
246 reserve_freed -= tmp;
252 * Increase the reserve pool until the desired allocation is met.
253 * Move an allocation freed from the excess pool into the reserve
254 * pool and schedule a balance operation.
256 if (excess_freed && (vio_cmo.desired > vio_cmo.reserve.size)) {
257 tmp = min(excess_freed, (vio_cmo.desired - vio_cmo.reserve.size));
259 vio_cmo.excess.size -= tmp;
260 vio_cmo.reserve.size += tmp;
265 /* Return memory from the excess pool to that pool */
267 vio_cmo.excess.free += excess_freed;
270 schedule_delayed_work(&vio_cmo.balance_q, VIO_CMO_BALANCE_DELAY);
271 spin_unlock_irqrestore(&vio_cmo.lock, flags);
275 * vio_cmo_entitlement_update - Manage system entitlement changes
277 * @new_entitlement: new system entitlement to attempt to accommodate
279 * Increases in entitlement will be used to fulfill the spare entitlement
280 * and the rest is given to the excess pool. Decreases, if they are
281 * possible, come from the excess pool and from unused device entitlement
283 * Returns: 0 on success, -ENOMEM when change can not be made
285 int vio_cmo_entitlement_update(size_t new_entitlement)
287 struct vio_dev *viodev;
288 struct vio_cmo_dev_entry *dev_ent;
290 size_t avail, delta, tmp;
292 spin_lock_irqsave(&vio_cmo.lock, flags);
294 /* Entitlement increases */
295 if (new_entitlement > vio_cmo.entitled) {
296 delta = new_entitlement - vio_cmo.entitled;
298 /* Fulfill spare allocation */
299 if (vio_cmo.spare < VIO_CMO_MIN_ENT) {
300 tmp = min(delta, (VIO_CMO_MIN_ENT - vio_cmo.spare));
301 vio_cmo.spare += tmp;
302 vio_cmo.reserve.size += tmp;
306 /* Remaining new allocation goes to the excess pool */
307 vio_cmo.entitled += delta;
308 vio_cmo.excess.size += delta;
309 vio_cmo.excess.free += delta;
314 /* Entitlement decreases */
315 delta = vio_cmo.entitled - new_entitlement;
316 avail = vio_cmo.excess.free;
319 * Need to check how much unused entitlement each device can
320 * sacrifice to fulfill entitlement change.
322 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
326 viodev = dev_ent->viodev;
327 if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
328 (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
329 avail += viodev->cmo.entitled -
330 max_t(size_t, viodev->cmo.allocated,
334 if (delta <= avail) {
335 vio_cmo.entitled -= delta;
337 /* Take entitlement from the excess pool first */
338 tmp = min(vio_cmo.excess.free, delta);
339 vio_cmo.excess.size -= tmp;
340 vio_cmo.excess.free -= tmp;
344 * Remove all but VIO_CMO_MIN_ENT bytes from devices
345 * until entitlement change is served
347 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
351 viodev = dev_ent->viodev;
353 if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
354 (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
355 tmp = viodev->cmo.entitled -
356 max_t(size_t, viodev->cmo.allocated,
358 viodev->cmo.entitled -= min(tmp, delta);
359 delta -= min(tmp, delta);
362 spin_unlock_irqrestore(&vio_cmo.lock, flags);
367 schedule_delayed_work(&vio_cmo.balance_q, 0);
368 spin_unlock_irqrestore(&vio_cmo.lock, flags);
373 * vio_cmo_balance - Balance entitlement among devices
375 * @work: work queue structure for this operation
377 * Any system entitlement above the minimum needed for devices, or
378 * already allocated to devices, can be distributed to the devices.
379 * The list of devices is iterated through to recalculate the desired
380 * entitlement level and to determine how much entitlement above the
381 * minimum entitlement is allocated to devices.
383 * Small chunks of the available entitlement are given to devices until
384 * their requirements are fulfilled or there is no entitlement left to give.
385 * Upon completion sizes of the reserve and excess pools are calculated.
387 * The system minimum entitlement level is also recalculated here.
388 * Entitlement will be reserved for devices even after vio_bus_remove to
389 * accommodate reloading the driver. The OF tree is walked to count the
390 * number of devices present and this will remove entitlement for devices
391 * that have actually left the system after having vio_bus_remove called.
393 static void vio_cmo_balance(struct work_struct *work)
396 struct vio_dev *viodev;
397 struct vio_cmo_dev_entry *dev_ent;
399 size_t avail = 0, level, chunk, need;
400 int devcount = 0, fulfilled;
402 cmo = container_of(work, struct vio_cmo, balance_q.work);
404 spin_lock_irqsave(&vio_cmo.lock, flags);
406 /* Calculate minimum entitlement and fulfill spare */
407 cmo->min = vio_cmo_num_OF_devs() * VIO_CMO_MIN_ENT;
408 BUG_ON(cmo->min > cmo->entitled);
409 cmo->spare = min_t(size_t, VIO_CMO_MIN_ENT, (cmo->entitled - cmo->min));
410 cmo->min += cmo->spare;
411 cmo->desired = cmo->min;
414 * Determine how much entitlement is available and reset device
417 avail = cmo->entitled - cmo->spare;
418 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
419 viodev = dev_ent->viodev;
421 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
422 cmo->desired += (viodev->cmo.desired - VIO_CMO_MIN_ENT);
423 avail -= max_t(size_t, viodev->cmo.allocated, VIO_CMO_MIN_ENT);
427 * Having provided each device with the minimum entitlement, loop
428 * over the devices portioning out the remaining entitlement
429 * until there is nothing left.
431 level = VIO_CMO_MIN_ENT;
434 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
435 viodev = dev_ent->viodev;
437 if (viodev->cmo.desired <= level) {
443 * Give the device up to VIO_CMO_BALANCE_CHUNK
444 * bytes of entitlement, but do not exceed the
445 * desired level of entitlement for the device.
447 chunk = min_t(size_t, avail, VIO_CMO_BALANCE_CHUNK);
448 chunk = min(chunk, (viodev->cmo.desired -
449 viodev->cmo.entitled));
450 viodev->cmo.entitled += chunk;
453 * If the memory for this entitlement increase was
454 * already allocated to the device it does not come
455 * from the available pool being portioned out.
457 need = max(viodev->cmo.allocated, viodev->cmo.entitled)-
458 max(viodev->cmo.allocated, level);
462 if (fulfilled == devcount)
464 level += VIO_CMO_BALANCE_CHUNK;
467 /* Calculate new reserve and excess pool sizes */
468 cmo->reserve.size = cmo->min;
469 cmo->excess.free = 0;
470 cmo->excess.size = 0;
472 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
473 viodev = dev_ent->viodev;
474 /* Calculated reserve size above the minimum entitlement */
475 if (viodev->cmo.entitled)
476 cmo->reserve.size += (viodev->cmo.entitled -
478 /* Calculated used excess entitlement */
479 if (viodev->cmo.allocated > viodev->cmo.entitled)
480 need += viodev->cmo.allocated - viodev->cmo.entitled;
482 cmo->excess.size = cmo->entitled - cmo->reserve.size;
483 cmo->excess.free = cmo->excess.size - need;
485 cancel_delayed_work(container_of(work, struct delayed_work, work));
486 spin_unlock_irqrestore(&vio_cmo.lock, flags);
489 static void *vio_dma_iommu_alloc_coherent(struct device *dev, size_t size,
490 dma_addr_t *dma_handle, gfp_t flag)
492 struct vio_dev *viodev = to_vio_dev(dev);
495 if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE))) {
496 atomic_inc(&viodev->cmo.allocs_failed);
500 ret = dma_iommu_ops.alloc_coherent(dev, size, dma_handle, flag);
501 if (unlikely(ret == NULL)) {
502 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
503 atomic_inc(&viodev->cmo.allocs_failed);
509 static void vio_dma_iommu_free_coherent(struct device *dev, size_t size,
510 void *vaddr, dma_addr_t dma_handle)
512 struct vio_dev *viodev = to_vio_dev(dev);
514 dma_iommu_ops.free_coherent(dev, size, vaddr, dma_handle);
516 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
519 static dma_addr_t vio_dma_iommu_map_page(struct device *dev, struct page *page,
520 unsigned long offset, size_t size,
521 enum dma_data_direction direction,
522 struct dma_attrs *attrs)
524 struct vio_dev *viodev = to_vio_dev(dev);
525 dma_addr_t ret = DMA_ERROR_CODE;
527 if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE))) {
528 atomic_inc(&viodev->cmo.allocs_failed);
532 ret = dma_iommu_ops.map_page(dev, page, offset, size, direction, attrs);
533 if (unlikely(dma_mapping_error(dev, ret))) {
534 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
535 atomic_inc(&viodev->cmo.allocs_failed);
541 static void vio_dma_iommu_unmap_page(struct device *dev, dma_addr_t dma_handle,
543 enum dma_data_direction direction,
544 struct dma_attrs *attrs)
546 struct vio_dev *viodev = to_vio_dev(dev);
548 dma_iommu_ops.unmap_page(dev, dma_handle, size, direction, attrs);
550 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
553 static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist,
554 int nelems, enum dma_data_direction direction,
555 struct dma_attrs *attrs)
557 struct vio_dev *viodev = to_vio_dev(dev);
558 struct scatterlist *sgl;
560 size_t alloc_size = 0;
562 for (sgl = sglist; count < nelems; count++, sgl++)
563 alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE);
565 if (vio_cmo_alloc(viodev, alloc_size)) {
566 atomic_inc(&viodev->cmo.allocs_failed);
570 ret = dma_iommu_ops.map_sg(dev, sglist, nelems, direction, attrs);
572 if (unlikely(!ret)) {
573 vio_cmo_dealloc(viodev, alloc_size);
574 atomic_inc(&viodev->cmo.allocs_failed);
577 for (sgl = sglist, count = 0; count < ret; count++, sgl++)
578 alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE);
580 vio_cmo_dealloc(viodev, alloc_size);
585 static void vio_dma_iommu_unmap_sg(struct device *dev,
586 struct scatterlist *sglist, int nelems,
587 enum dma_data_direction direction,
588 struct dma_attrs *attrs)
590 struct vio_dev *viodev = to_vio_dev(dev);
591 struct scatterlist *sgl;
592 size_t alloc_size = 0;
595 for (sgl = sglist; count < nelems; count++, sgl++)
596 alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE);
598 dma_iommu_ops.unmap_sg(dev, sglist, nelems, direction, attrs);
600 vio_cmo_dealloc(viodev, alloc_size);
603 struct dma_mapping_ops vio_dma_mapping_ops = {
604 .alloc_coherent = vio_dma_iommu_alloc_coherent,
605 .free_coherent = 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,
614 * vio_cmo_set_dev_desired - Set desired entitlement for a device
616 * @viodev: struct vio_dev for device to alter
617 * @new_desired: new desired entitlement level in bytes
619 * For use by devices to request a change to their entitlement at runtime or
620 * through sysfs. The desired entitlement level is changed and a balancing
621 * of system resources is scheduled to run in the future.
623 void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired)
626 struct vio_cmo_dev_entry *dev_ent;
629 if (!firmware_has_feature(FW_FEATURE_CMO))
632 spin_lock_irqsave(&vio_cmo.lock, flags);
633 if (desired < VIO_CMO_MIN_ENT)
634 desired = VIO_CMO_MIN_ENT;
637 * Changes will not be made for devices not in the device list.
638 * If it is not in the device list, then no driver is loaded
639 * for the device and it can not receive entitlement.
641 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
642 if (viodev == dev_ent->viodev) {
649 /* Increase/decrease in desired device entitlement */
650 if (desired >= viodev->cmo.desired) {
651 /* Just bump the bus and device values prior to a balance*/
652 vio_cmo.desired += desired - viodev->cmo.desired;
653 viodev->cmo.desired = desired;
655 /* Decrease bus and device values for desired entitlement */
656 vio_cmo.desired -= viodev->cmo.desired - desired;
657 viodev->cmo.desired = desired;
659 * If less entitlement is desired than current entitlement, move
660 * any reserve memory in the change region to the excess pool.
662 if (viodev->cmo.entitled > desired) {
663 vio_cmo.reserve.size -= viodev->cmo.entitled - desired;
664 vio_cmo.excess.size += viodev->cmo.entitled - desired;
666 * If entitlement moving from the reserve pool to the
667 * excess pool is currently unused, add to the excess
670 if (viodev->cmo.allocated < viodev->cmo.entitled)
671 vio_cmo.excess.free += viodev->cmo.entitled -
672 max(viodev->cmo.allocated, desired);
673 viodev->cmo.entitled = desired;
676 schedule_delayed_work(&vio_cmo.balance_q, 0);
677 spin_unlock_irqrestore(&vio_cmo.lock, flags);
681 * vio_cmo_bus_probe - Handle CMO specific bus probe activities
683 * @viodev - Pointer to struct vio_dev for device
685 * Determine the devices IO memory entitlement needs, attempting
686 * to satisfy the system minimum entitlement at first and scheduling
687 * a balance operation to take care of the rest at a later time.
689 * Returns: 0 on success, -EINVAL when device doesn't support CMO, and
690 * -ENOMEM when entitlement is not available for device or
694 static int vio_cmo_bus_probe(struct vio_dev *viodev)
696 struct vio_cmo_dev_entry *dev_ent;
697 struct device *dev = &viodev->dev;
698 struct vio_driver *viodrv = to_vio_driver(dev->driver);
703 * Check to see that device has a DMA window and configure
704 * entitlement for the device.
706 if (of_get_property(viodev->dev.archdata.of_node,
707 "ibm,my-dma-window", NULL)) {
708 /* Check that the driver is CMO enabled and get desired DMA */
709 if (!viodrv->get_desired_dma) {
710 dev_err(dev, "%s: device driver does not support CMO\n",
715 viodev->cmo.desired = IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev));
716 if (viodev->cmo.desired < VIO_CMO_MIN_ENT)
717 viodev->cmo.desired = VIO_CMO_MIN_ENT;
718 size = VIO_CMO_MIN_ENT;
720 dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry),
725 dev_ent->viodev = viodev;
726 spin_lock_irqsave(&vio_cmo.lock, flags);
727 list_add(&dev_ent->list, &vio_cmo.device_list);
729 viodev->cmo.desired = 0;
731 spin_lock_irqsave(&vio_cmo.lock, flags);
735 * If the needs for vio_cmo.min have not changed since they
736 * were last set, the number of devices in the OF tree has
737 * been constant and the IO memory for this is already in
740 if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) *
742 /* Updated desired entitlement if device requires it */
744 vio_cmo.desired += (viodev->cmo.desired -
749 tmp = vio_cmo.spare + vio_cmo.excess.free;
751 dev_err(dev, "%s: insufficient free "
752 "entitlement to add device. "
753 "Need %lu, have %lu\n", __func__,
754 size, (vio_cmo.spare + tmp));
755 spin_unlock_irqrestore(&vio_cmo.lock, flags);
759 /* Use excess pool first to fulfill request */
760 tmp = min(size, vio_cmo.excess.free);
761 vio_cmo.excess.free -= tmp;
762 vio_cmo.excess.size -= tmp;
763 vio_cmo.reserve.size += tmp;
765 /* Use spare if excess pool was insufficient */
766 vio_cmo.spare -= size - tmp;
768 /* Update bus accounting */
770 vio_cmo.desired += viodev->cmo.desired;
772 spin_unlock_irqrestore(&vio_cmo.lock, flags);
777 * vio_cmo_bus_remove - Handle CMO specific bus removal activities
779 * @viodev - Pointer to struct vio_dev for device
781 * Remove the device from the cmo device list. The minimum entitlement
782 * will be reserved for the device as long as it is in the system. The
783 * rest of the entitlement the device had been allocated will be returned
786 static void vio_cmo_bus_remove(struct vio_dev *viodev)
788 struct vio_cmo_dev_entry *dev_ent;
792 spin_lock_irqsave(&vio_cmo.lock, flags);
793 if (viodev->cmo.allocated) {
794 dev_err(&viodev->dev, "%s: device had %lu bytes of IO "
795 "allocated after remove operation.\n",
796 __func__, viodev->cmo.allocated);
801 * Remove the device from the device list being maintained for
802 * CMO enabled devices.
804 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
805 if (viodev == dev_ent->viodev) {
806 list_del(&dev_ent->list);
812 * Devices may not require any entitlement and they do not need
813 * to be processed. Otherwise, return the device's entitlement
816 if (viodev->cmo.entitled) {
818 * This device has not yet left the OF tree, it's
819 * minimum entitlement remains in vio_cmo.min and
822 vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT);
825 * Save min allocation for device in reserve as long
826 * as it exists in OF tree as determined by later
829 viodev->cmo.entitled -= VIO_CMO_MIN_ENT;
831 /* Replenish spare from freed reserve pool */
832 if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) {
833 tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT -
835 vio_cmo.spare += tmp;
836 viodev->cmo.entitled -= tmp;
839 /* Remaining reserve goes to excess pool */
840 vio_cmo.excess.size += viodev->cmo.entitled;
841 vio_cmo.excess.free += viodev->cmo.entitled;
842 vio_cmo.reserve.size -= viodev->cmo.entitled;
845 * Until the device is removed it will keep a
846 * minimum entitlement; this will guarantee that
847 * a module unload/load will result in a success.
849 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
850 viodev->cmo.desired = VIO_CMO_MIN_ENT;
851 atomic_set(&viodev->cmo.allocs_failed, 0);
854 spin_unlock_irqrestore(&vio_cmo.lock, flags);
857 static void vio_cmo_set_dma_ops(struct vio_dev *viodev)
859 vio_dma_mapping_ops.dma_supported = dma_iommu_ops.dma_supported;
860 viodev->dev.archdata.dma_ops = &vio_dma_mapping_ops;
864 * vio_cmo_bus_init - CMO entitlement initialization at bus init time
866 * Set up the reserve and excess entitlement pools based on available
867 * system entitlement and the number of devices in the OF tree that
868 * require entitlement in the reserve pool.
870 static void vio_cmo_bus_init(void)
872 struct hvcall_mpp_data mpp_data;
875 memset(&vio_cmo, 0, sizeof(struct vio_cmo));
876 spin_lock_init(&vio_cmo.lock);
877 INIT_LIST_HEAD(&vio_cmo.device_list);
878 INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance);
880 /* Get current system entitlement */
881 err = h_get_mpp(&mpp_data);
884 * On failure, continue with entitlement set to 0, will panic()
885 * later when spare is reserved.
887 if (err != H_SUCCESS) {
888 printk(KERN_ERR "%s: unable to determine system IO "\
889 "entitlement. (%d)\n", __func__, err);
890 vio_cmo.entitled = 0;
892 vio_cmo.entitled = mpp_data.entitled_mem;
895 /* Set reservation and check against entitlement */
896 vio_cmo.spare = VIO_CMO_MIN_ENT;
897 vio_cmo.reserve.size = vio_cmo.spare;
898 vio_cmo.reserve.size += (vio_cmo_num_OF_devs() *
900 if (vio_cmo.reserve.size > vio_cmo.entitled) {
901 printk(KERN_ERR "%s: insufficient system entitlement\n",
903 panic("%s: Insufficient system entitlement", __func__);
906 /* Set the remaining accounting variables */
907 vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size;
908 vio_cmo.excess.free = vio_cmo.excess.size;
909 vio_cmo.min = vio_cmo.reserve.size;
910 vio_cmo.desired = vio_cmo.reserve.size;
913 /* sysfs device functions and data structures for CMO */
915 #define viodev_cmo_rd_attr(name) \
916 static ssize_t viodev_cmo_##name##_show(struct device *dev, \
917 struct device_attribute *attr, \
920 return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name); \
923 static ssize_t viodev_cmo_allocs_failed_show(struct device *dev,
924 struct device_attribute *attr, char *buf)
926 struct vio_dev *viodev = to_vio_dev(dev);
927 return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed));
930 static ssize_t viodev_cmo_allocs_failed_reset(struct device *dev,
931 struct device_attribute *attr, const char *buf, size_t count)
933 struct vio_dev *viodev = to_vio_dev(dev);
934 atomic_set(&viodev->cmo.allocs_failed, 0);
938 static ssize_t viodev_cmo_desired_set(struct device *dev,
939 struct device_attribute *attr, const char *buf, size_t count)
941 struct vio_dev *viodev = to_vio_dev(dev);
945 ret = strict_strtoul(buf, 10, &new_desired);
949 vio_cmo_set_dev_desired(viodev, new_desired);
953 viodev_cmo_rd_attr(desired);
954 viodev_cmo_rd_attr(entitled);
955 viodev_cmo_rd_attr(allocated);
957 static ssize_t name_show(struct device *, struct device_attribute *, char *);
958 static ssize_t devspec_show(struct device *, struct device_attribute *, char *);
959 static struct device_attribute vio_cmo_dev_attrs[] = {
962 __ATTR(cmo_desired, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
963 viodev_cmo_desired_show, viodev_cmo_desired_set),
964 __ATTR(cmo_entitled, S_IRUGO, viodev_cmo_entitled_show, NULL),
965 __ATTR(cmo_allocated, S_IRUGO, viodev_cmo_allocated_show, NULL),
966 __ATTR(cmo_allocs_failed, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
967 viodev_cmo_allocs_failed_show, viodev_cmo_allocs_failed_reset),
971 /* sysfs bus functions and data structures for CMO */
973 #define viobus_cmo_rd_attr(name) \
975 viobus_cmo_##name##_show(struct bus_type *bt, char *buf) \
977 return sprintf(buf, "%lu\n", vio_cmo.name); \
980 #define viobus_cmo_pool_rd_attr(name, var) \
982 viobus_cmo_##name##_pool_show_##var(struct bus_type *bt, char *buf) \
984 return sprintf(buf, "%lu\n", vio_cmo.name.var); \
987 static ssize_t viobus_cmo_high_reset(struct bus_type *bt, const char *buf,
992 spin_lock_irqsave(&vio_cmo.lock, flags);
993 vio_cmo.high = vio_cmo.curr;
994 spin_unlock_irqrestore(&vio_cmo.lock, flags);
999 viobus_cmo_rd_attr(entitled);
1000 viobus_cmo_pool_rd_attr(reserve, size);
1001 viobus_cmo_pool_rd_attr(excess, size);
1002 viobus_cmo_pool_rd_attr(excess, free);
1003 viobus_cmo_rd_attr(spare);
1004 viobus_cmo_rd_attr(min);
1005 viobus_cmo_rd_attr(desired);
1006 viobus_cmo_rd_attr(curr);
1007 viobus_cmo_rd_attr(high);
1009 static struct bus_attribute vio_cmo_bus_attrs[] = {
1010 __ATTR(cmo_entitled, S_IRUGO, viobus_cmo_entitled_show, NULL),
1011 __ATTR(cmo_reserve_size, S_IRUGO, viobus_cmo_reserve_pool_show_size, NULL),
1012 __ATTR(cmo_excess_size, S_IRUGO, viobus_cmo_excess_pool_show_size, NULL),
1013 __ATTR(cmo_excess_free, S_IRUGO, viobus_cmo_excess_pool_show_free, NULL),
1014 __ATTR(cmo_spare, S_IRUGO, viobus_cmo_spare_show, NULL),
1015 __ATTR(cmo_min, S_IRUGO, viobus_cmo_min_show, NULL),
1016 __ATTR(cmo_desired, S_IRUGO, viobus_cmo_desired_show, NULL),
1017 __ATTR(cmo_curr, S_IRUGO, viobus_cmo_curr_show, NULL),
1018 __ATTR(cmo_high, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
1019 viobus_cmo_high_show, viobus_cmo_high_reset),
1023 static void vio_cmo_sysfs_init(void)
1025 vio_bus_type.dev_attrs = vio_cmo_dev_attrs;
1026 vio_bus_type.bus_attrs = vio_cmo_bus_attrs;
1028 #else /* CONFIG_PPC_SMLPAR */
1029 /* Dummy functions for iSeries platform */
1030 int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; }
1031 void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {}
1032 static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; }
1033 static void vio_cmo_bus_remove(struct vio_dev *viodev) {}
1034 static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {}
1035 static void vio_cmo_bus_init(void) {}
1036 static void vio_cmo_sysfs_init(void) { }
1037 #endif /* CONFIG_PPC_SMLPAR */
1038 EXPORT_SYMBOL(vio_cmo_entitlement_update);
1039 EXPORT_SYMBOL(vio_cmo_set_dev_desired);
1041 static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
1043 const unsigned char *dma_window;
1044 struct iommu_table *tbl;
1045 unsigned long offset, size;
1047 if (firmware_has_feature(FW_FEATURE_ISERIES))
1048 return vio_build_iommu_table_iseries(dev);
1050 dma_window = of_get_property(dev->dev.archdata.of_node,
1051 "ibm,my-dma-window", NULL);
1055 tbl = kmalloc(sizeof(*tbl), GFP_KERNEL);
1057 of_parse_dma_window(dev->dev.archdata.of_node, dma_window,
1058 &tbl->it_index, &offset, &size);
1060 /* TCE table size - measured in tce entries */
1061 tbl->it_size = size >> IOMMU_PAGE_SHIFT;
1062 /* offset for VIO should always be 0 */
1063 tbl->it_offset = offset >> IOMMU_PAGE_SHIFT;
1065 tbl->it_type = TCE_VB;
1067 return iommu_init_table(tbl, -1);
1071 * vio_match_device: - Tell if a VIO device has a matching
1072 * VIO device id structure.
1073 * @ids: array of VIO device id structures to search in
1074 * @dev: the VIO device structure to match against
1076 * Used by a driver to check whether a VIO device present in the
1077 * system is in its list of supported devices. Returns the matching
1078 * vio_device_id structure or NULL if there is no match.
1080 static const struct vio_device_id *vio_match_device(
1081 const struct vio_device_id *ids, const struct vio_dev *dev)
1083 while (ids->type[0] != '\0') {
1084 if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) &&
1085 of_device_is_compatible(dev->dev.archdata.of_node,
1094 * Convert from struct device to struct vio_dev and pass to driver.
1095 * dev->driver has already been set by generic code because vio_bus_match
1098 static int vio_bus_probe(struct device *dev)
1100 struct vio_dev *viodev = to_vio_dev(dev);
1101 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1102 const struct vio_device_id *id;
1103 int error = -ENODEV;
1108 id = vio_match_device(viodrv->id_table, viodev);
1110 memset(&viodev->cmo, 0, sizeof(viodev->cmo));
1111 if (firmware_has_feature(FW_FEATURE_CMO)) {
1112 error = vio_cmo_bus_probe(viodev);
1116 error = viodrv->probe(viodev, id);
1117 if (error && firmware_has_feature(FW_FEATURE_CMO))
1118 vio_cmo_bus_remove(viodev);
1124 /* convert from struct device to struct vio_dev and pass to driver. */
1125 static int vio_bus_remove(struct device *dev)
1127 struct vio_dev *viodev = to_vio_dev(dev);
1128 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1129 struct device *devptr;
1133 * Hold a reference to the device after the remove function is called
1134 * to allow for CMO accounting cleanup for the device.
1136 devptr = get_device(dev);
1139 ret = viodrv->remove(viodev);
1141 if (!ret && firmware_has_feature(FW_FEATURE_CMO))
1142 vio_cmo_bus_remove(viodev);
1149 * vio_register_driver: - Register a new vio driver
1150 * @drv: The vio_driver structure to be registered.
1152 int vio_register_driver(struct vio_driver *viodrv)
1154 printk(KERN_DEBUG "%s: driver %s registering\n", __func__,
1155 viodrv->driver.name);
1157 /* fill in 'struct driver' fields */
1158 viodrv->driver.bus = &vio_bus_type;
1160 return driver_register(&viodrv->driver);
1162 EXPORT_SYMBOL(vio_register_driver);
1165 * vio_unregister_driver - Remove registration of vio driver.
1166 * @driver: The vio_driver struct to be removed form registration
1168 void vio_unregister_driver(struct vio_driver *viodrv)
1170 driver_unregister(&viodrv->driver);
1172 EXPORT_SYMBOL(vio_unregister_driver);
1174 /* vio_dev refcount hit 0 */
1175 static void __devinit vio_dev_release(struct device *dev)
1177 /* XXX should free TCE table */
1178 of_node_put(dev->archdata.of_node);
1179 kfree(to_vio_dev(dev));
1183 * vio_register_device_node: - Register a new vio device.
1184 * @of_node: The OF node for this device.
1186 * Creates and initializes a vio_dev structure from the data in
1187 * of_node and adds it to the list of virtual devices.
1188 * Returns a pointer to the created vio_dev or NULL if node has
1189 * NULL device_type or compatible fields.
1191 struct vio_dev *vio_register_device_node(struct device_node *of_node)
1193 struct vio_dev *viodev;
1194 const unsigned int *unit_address;
1196 /* we need the 'device_type' property, in order to match with drivers */
1197 if (of_node->type == NULL) {
1198 printk(KERN_WARNING "%s: node %s missing 'device_type'\n",
1200 of_node->name ? of_node->name : "<unknown>");
1204 unit_address = of_get_property(of_node, "reg", NULL);
1205 if (unit_address == NULL) {
1206 printk(KERN_WARNING "%s: node %s missing 'reg'\n",
1208 of_node->name ? of_node->name : "<unknown>");
1212 /* allocate a vio_dev for this node */
1213 viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
1217 viodev->irq = irq_of_parse_and_map(of_node, 0);
1219 snprintf(viodev->dev.bus_id, BUS_ID_SIZE, "%x", *unit_address);
1220 viodev->name = of_node->name;
1221 viodev->type = of_node->type;
1222 viodev->unit_address = *unit_address;
1223 if (firmware_has_feature(FW_FEATURE_ISERIES)) {
1224 unit_address = of_get_property(of_node,
1225 "linux,unit_address", NULL);
1226 if (unit_address != NULL)
1227 viodev->unit_address = *unit_address;
1229 viodev->dev.archdata.of_node = of_node_get(of_node);
1231 if (firmware_has_feature(FW_FEATURE_CMO))
1232 vio_cmo_set_dma_ops(viodev);
1234 viodev->dev.archdata.dma_ops = &dma_iommu_ops;
1235 viodev->dev.archdata.dma_data = vio_build_iommu_table(viodev);
1236 set_dev_node(&viodev->dev, of_node_to_nid(of_node));
1238 /* init generic 'struct device' fields: */
1239 viodev->dev.parent = &vio_bus_device.dev;
1240 viodev->dev.bus = &vio_bus_type;
1241 viodev->dev.release = vio_dev_release;
1243 /* register with generic device framework */
1244 if (device_register(&viodev->dev)) {
1245 printk(KERN_ERR "%s: failed to register device %s\n",
1246 __func__, viodev->dev.bus_id);
1247 /* XXX free TCE table */
1254 EXPORT_SYMBOL(vio_register_device_node);
1257 * vio_bus_init: - Initialize the virtual IO bus
1259 static int __init vio_bus_init(void)
1262 struct device_node *node_vroot;
1264 if (firmware_has_feature(FW_FEATURE_CMO))
1265 vio_cmo_sysfs_init();
1267 err = bus_register(&vio_bus_type);
1269 printk(KERN_ERR "failed to register VIO bus\n");
1274 * The fake parent of all vio devices, just to give us
1277 err = device_register(&vio_bus_device.dev);
1279 printk(KERN_WARNING "%s: device_register returned %i\n",
1284 if (firmware_has_feature(FW_FEATURE_CMO))
1287 node_vroot = of_find_node_by_name(NULL, "vdevice");
1289 struct device_node *of_node;
1292 * Create struct vio_devices for each virtual device in
1293 * the device tree. Drivers will associate with them later.
1295 for (of_node = node_vroot->child; of_node != NULL;
1296 of_node = of_node->sibling)
1297 vio_register_device_node(of_node);
1298 of_node_put(node_vroot);
1303 __initcall(vio_bus_init);
1305 static ssize_t name_show(struct device *dev,
1306 struct device_attribute *attr, char *buf)
1308 return sprintf(buf, "%s\n", to_vio_dev(dev)->name);
1311 static ssize_t devspec_show(struct device *dev,
1312 struct device_attribute *attr, char *buf)
1314 struct device_node *of_node = dev->archdata.of_node;
1316 return sprintf(buf, "%s\n", of_node ? of_node->full_name : "none");
1319 static struct device_attribute vio_dev_attrs[] = {
1325 void __devinit vio_unregister_device(struct vio_dev *viodev)
1327 device_unregister(&viodev->dev);
1329 EXPORT_SYMBOL(vio_unregister_device);
1331 static int vio_bus_match(struct device *dev, struct device_driver *drv)
1333 const struct vio_dev *vio_dev = to_vio_dev(dev);
1334 struct vio_driver *vio_drv = to_vio_driver(drv);
1335 const struct vio_device_id *ids = vio_drv->id_table;
1337 return (ids != NULL) && (vio_match_device(ids, vio_dev) != NULL);
1340 static int vio_hotplug(struct device *dev, struct kobj_uevent_env *env)
1342 const struct vio_dev *vio_dev = to_vio_dev(dev);
1343 struct device_node *dn;
1346 dn = dev->archdata.of_node;
1349 cp = of_get_property(dn, "compatible", NULL);
1353 add_uevent_var(env, "MODALIAS=vio:T%sS%s", vio_dev->type, cp);
1357 static struct bus_type vio_bus_type = {
1359 .dev_attrs = vio_dev_attrs,
1360 .uevent = vio_hotplug,
1361 .match = vio_bus_match,
1362 .probe = vio_bus_probe,
1363 .remove = vio_bus_remove,
1367 * vio_get_attribute: - get attribute for virtual device
1368 * @vdev: The vio device to get property.
1369 * @which: The property/attribute to be extracted.
1370 * @length: Pointer to length of returned data size (unused if NULL).
1372 * Calls prom.c's of_get_property() to return the value of the
1373 * attribute specified by @which
1375 const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length)
1377 return of_get_property(vdev->dev.archdata.of_node, which, length);
1379 EXPORT_SYMBOL(vio_get_attribute);
1381 #ifdef CONFIG_PPC_PSERIES
1382 /* vio_find_name() - internal because only vio.c knows how we formatted the
1385 static struct vio_dev *vio_find_name(const char *name)
1387 struct device *found;
1389 found = bus_find_device_by_name(&vio_bus_type, NULL, name);
1393 return to_vio_dev(found);
1397 * vio_find_node - find an already-registered vio_dev
1398 * @vnode: device_node of the virtual device we're looking for
1400 struct vio_dev *vio_find_node(struct device_node *vnode)
1402 const uint32_t *unit_address;
1403 char kobj_name[BUS_ID_SIZE];
1405 /* construct the kobject name from the device node */
1406 unit_address = of_get_property(vnode, "reg", NULL);
1409 snprintf(kobj_name, BUS_ID_SIZE, "%x", *unit_address);
1411 return vio_find_name(kobj_name);
1413 EXPORT_SYMBOL(vio_find_node);
1415 int vio_enable_interrupts(struct vio_dev *dev)
1417 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE);
1418 if (rc != H_SUCCESS)
1419 printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc);
1422 EXPORT_SYMBOL(vio_enable_interrupts);
1424 int vio_disable_interrupts(struct vio_dev *dev)
1426 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE);
1427 if (rc != H_SUCCESS)
1428 printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc);
1431 EXPORT_SYMBOL(vio_disable_interrupts);
1432 #endif /* CONFIG_PPC_PSERIES */