Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IBM PowerPC Virtual I/O Infrastructure Support. | |
3 | * | |
a90ab95a | 4 | * Copyright (c) 2003,2008 IBM Corp. |
1da177e4 LT |
5 | * Dave Engebretsen engebret@us.ibm.com |
6 | * Santiago Leon santil@us.ibm.com | |
7 | * Hollis Blanchard <hollisb@us.ibm.com> | |
19dbd0f6 | 8 | * Stephen Rothwell |
a90ab95a | 9 | * Robert Jennings <rcjenn@us.ibm.com> |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or | |
12 | * modify it under the terms of the GNU General Public License | |
13 | * as published by the Free Software Foundation; either version | |
14 | * 2 of the License, or (at your option) any later version. | |
15 | */ | |
16 | ||
c7f0e8cb SR |
17 | #include <linux/types.h> |
18 | #include <linux/device.h> | |
1da177e4 | 19 | #include <linux/init.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
1da177e4 | 21 | #include <linux/console.h> |
1da177e4 | 22 | #include <linux/module.h> |
1da177e4 LT |
23 | #include <linux/mm.h> |
24 | #include <linux/dma-mapping.h> | |
c7f0e8cb SR |
25 | #include <linux/kobject.h> |
26 | ||
1da177e4 LT |
27 | #include <asm/iommu.h> |
28 | #include <asm/dma.h> | |
1da177e4 | 29 | #include <asm/vio.h> |
143dcec2 | 30 | #include <asm/prom.h> |
e10fa773 | 31 | #include <asm/firmware.h> |
c7f0e8cb SR |
32 | #include <asm/tce.h> |
33 | #include <asm/abs_addr.h> | |
34 | #include <asm/page.h> | |
35 | #include <asm/hvcall.h> | |
36 | #include <asm/iseries/vio.h> | |
37 | #include <asm/iseries/hv_types.h> | |
38 | #include <asm/iseries/hv_lp_config.h> | |
39 | #include <asm/iseries/hv_call_xm.h> | |
40 | #include <asm/iseries/iommu.h> | |
41 | ||
6fccab26 SR |
42 | static struct bus_type vio_bus_type; |
43 | ||
c7f0e8cb | 44 | static struct vio_dev vio_bus_device = { /* fake "parent" device */ |
aab0d375 | 45 | .name = "vio", |
ac5b33c9 | 46 | .type = "", |
aab0d375 | 47 | .dev.init_name = "vio", |
ac5b33c9 | 48 | .dev.bus = &vio_bus_type, |
1da177e4 | 49 | }; |
ac5b33c9 | 50 | |
a90ab95a RJ |
51 | #ifdef CONFIG_PPC_SMLPAR |
52 | /** | |
53 | * vio_cmo_pool - A pool of IO memory for CMO use | |
54 | * | |
55 | * @size: The size of the pool in bytes | |
56 | * @free: The amount of free memory in the pool | |
57 | */ | |
58 | struct vio_cmo_pool { | |
59 | size_t size; | |
60 | size_t free; | |
61 | }; | |
62 | ||
63 | /* How many ms to delay queued balance work */ | |
64 | #define VIO_CMO_BALANCE_DELAY 100 | |
65 | ||
66 | /* Portion out IO memory to CMO devices by this chunk size */ | |
67 | #define VIO_CMO_BALANCE_CHUNK 131072 | |
68 | ||
69 | /** | |
70 | * vio_cmo_dev_entry - A device that is CMO-enabled and requires entitlement | |
71 | * | |
72 | * @vio_dev: struct vio_dev pointer | |
73 | * @list: pointer to other devices on bus that are being tracked | |
74 | */ | |
75 | struct vio_cmo_dev_entry { | |
76 | struct vio_dev *viodev; | |
77 | struct list_head list; | |
78 | }; | |
79 | ||
80 | /** | |
81 | * vio_cmo - VIO bus accounting structure for CMO entitlement | |
82 | * | |
83 | * @lock: spinlock for entire structure | |
84 | * @balance_q: work queue for balancing system entitlement | |
85 | * @device_list: list of CMO-enabled devices requiring entitlement | |
86 | * @entitled: total system entitlement in bytes | |
87 | * @reserve: pool of memory from which devices reserve entitlement, incl. spare | |
88 | * @excess: pool of excess entitlement not needed for device reserves or spare | |
89 | * @spare: IO memory for device hotplug functionality | |
90 | * @min: minimum necessary for system operation | |
91 | * @desired: desired memory for system operation | |
92 | * @curr: bytes currently allocated | |
93 | * @high: high water mark for IO data usage | |
94 | */ | |
95 | struct vio_cmo { | |
96 | spinlock_t lock; | |
97 | struct delayed_work balance_q; | |
98 | struct list_head device_list; | |
99 | size_t entitled; | |
100 | struct vio_cmo_pool reserve; | |
101 | struct vio_cmo_pool excess; | |
102 | size_t spare; | |
103 | size_t min; | |
104 | size_t desired; | |
105 | size_t curr; | |
106 | size_t high; | |
107 | } vio_cmo; | |
108 | ||
109 | /** | |
110 | * vio_cmo_OF_devices - Count the number of OF devices that have DMA windows | |
111 | */ | |
112 | static int vio_cmo_num_OF_devs(void) | |
113 | { | |
114 | struct device_node *node_vroot; | |
115 | int count = 0; | |
116 | ||
117 | /* | |
118 | * Count the number of vdevice entries with an | |
119 | * ibm,my-dma-window OF property | |
120 | */ | |
121 | node_vroot = of_find_node_by_name(NULL, "vdevice"); | |
122 | if (node_vroot) { | |
123 | struct device_node *of_node; | |
124 | struct property *prop; | |
125 | ||
126 | for_each_child_of_node(node_vroot, of_node) { | |
127 | prop = of_find_property(of_node, "ibm,my-dma-window", | |
128 | NULL); | |
129 | if (prop) | |
130 | count++; | |
131 | } | |
132 | } | |
133 | of_node_put(node_vroot); | |
134 | return count; | |
135 | } | |
136 | ||
137 | /** | |
138 | * vio_cmo_alloc - allocate IO memory for CMO-enable devices | |
139 | * | |
140 | * @viodev: VIO device requesting IO memory | |
141 | * @size: size of allocation requested | |
142 | * | |
143 | * Allocations come from memory reserved for the devices and any excess | |
144 | * IO memory available to all devices. The spare pool used to service | |
145 | * hotplug must be equal to %VIO_CMO_MIN_ENT for the excess pool to be | |
146 | * made available. | |
147 | * | |
148 | * Return codes: | |
149 | * 0 for successful allocation and -ENOMEM for a failure | |
150 | */ | |
151 | static inline int vio_cmo_alloc(struct vio_dev *viodev, size_t size) | |
152 | { | |
153 | unsigned long flags; | |
154 | size_t reserve_free = 0; | |
155 | size_t excess_free = 0; | |
156 | int ret = -ENOMEM; | |
157 | ||
158 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
159 | ||
160 | /* Determine the amount of free entitlement available in reserve */ | |
161 | if (viodev->cmo.entitled > viodev->cmo.allocated) | |
162 | reserve_free = viodev->cmo.entitled - viodev->cmo.allocated; | |
163 | ||
164 | /* If spare is not fulfilled, the excess pool can not be used. */ | |
165 | if (vio_cmo.spare >= VIO_CMO_MIN_ENT) | |
166 | excess_free = vio_cmo.excess.free; | |
167 | ||
168 | /* The request can be satisfied */ | |
169 | if ((reserve_free + excess_free) >= size) { | |
170 | vio_cmo.curr += size; | |
171 | if (vio_cmo.curr > vio_cmo.high) | |
172 | vio_cmo.high = vio_cmo.curr; | |
173 | viodev->cmo.allocated += size; | |
174 | size -= min(reserve_free, size); | |
175 | vio_cmo.excess.free -= size; | |
176 | ret = 0; | |
177 | } | |
178 | ||
179 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
180 | return ret; | |
181 | } | |
182 | ||
183 | /** | |
184 | * vio_cmo_dealloc - deallocate IO memory from CMO-enable devices | |
185 | * @viodev: VIO device freeing IO memory | |
186 | * @size: size of deallocation | |
187 | * | |
188 | * IO memory is freed by the device back to the correct memory pools. | |
189 | * The spare pool is replenished first from either memory pool, then | |
190 | * the reserve pool is used to reduce device entitlement, the excess | |
191 | * pool is used to increase the reserve pool toward the desired entitlement | |
192 | * target, and then the remaining memory is returned to the pools. | |
193 | * | |
194 | */ | |
195 | static inline void vio_cmo_dealloc(struct vio_dev *viodev, size_t size) | |
196 | { | |
197 | unsigned long flags; | |
198 | size_t spare_needed = 0; | |
199 | size_t excess_freed = 0; | |
200 | size_t reserve_freed = size; | |
201 | size_t tmp; | |
202 | int balance = 0; | |
203 | ||
204 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
205 | vio_cmo.curr -= size; | |
206 | ||
207 | /* Amount of memory freed from the excess pool */ | |
208 | if (viodev->cmo.allocated > viodev->cmo.entitled) { | |
209 | excess_freed = min(reserve_freed, (viodev->cmo.allocated - | |
210 | viodev->cmo.entitled)); | |
211 | reserve_freed -= excess_freed; | |
212 | } | |
213 | ||
214 | /* Remove allocation from device */ | |
215 | viodev->cmo.allocated -= (reserve_freed + excess_freed); | |
216 | ||
217 | /* Spare is a subset of the reserve pool, replenish it first. */ | |
218 | spare_needed = VIO_CMO_MIN_ENT - vio_cmo.spare; | |
219 | ||
220 | /* | |
221 | * Replenish the spare in the reserve pool from the excess pool. | |
222 | * This moves entitlement into the reserve pool. | |
223 | */ | |
224 | if (spare_needed && excess_freed) { | |
225 | tmp = min(excess_freed, spare_needed); | |
226 | vio_cmo.excess.size -= tmp; | |
227 | vio_cmo.reserve.size += tmp; | |
228 | vio_cmo.spare += tmp; | |
229 | excess_freed -= tmp; | |
230 | spare_needed -= tmp; | |
231 | balance = 1; | |
232 | } | |
233 | ||
234 | /* | |
235 | * Replenish the spare in the reserve pool from the reserve pool. | |
236 | * This removes entitlement from the device down to VIO_CMO_MIN_ENT, | |
237 | * if needed, and gives it to the spare pool. The amount of used | |
238 | * memory in this pool does not change. | |
239 | */ | |
240 | if (spare_needed && reserve_freed) { | |
241 | tmp = min(spare_needed, min(reserve_freed, | |
242 | (viodev->cmo.entitled - | |
243 | VIO_CMO_MIN_ENT))); | |
244 | ||
245 | vio_cmo.spare += tmp; | |
246 | viodev->cmo.entitled -= tmp; | |
247 | reserve_freed -= tmp; | |
248 | spare_needed -= tmp; | |
249 | balance = 1; | |
250 | } | |
251 | ||
252 | /* | |
253 | * Increase the reserve pool until the desired allocation is met. | |
254 | * Move an allocation freed from the excess pool into the reserve | |
255 | * pool and schedule a balance operation. | |
256 | */ | |
257 | if (excess_freed && (vio_cmo.desired > vio_cmo.reserve.size)) { | |
258 | tmp = min(excess_freed, (vio_cmo.desired - vio_cmo.reserve.size)); | |
259 | ||
260 | vio_cmo.excess.size -= tmp; | |
261 | vio_cmo.reserve.size += tmp; | |
262 | excess_freed -= tmp; | |
263 | balance = 1; | |
264 | } | |
265 | ||
266 | /* Return memory from the excess pool to that pool */ | |
267 | if (excess_freed) | |
268 | vio_cmo.excess.free += excess_freed; | |
269 | ||
270 | if (balance) | |
271 | schedule_delayed_work(&vio_cmo.balance_q, VIO_CMO_BALANCE_DELAY); | |
272 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
273 | } | |
274 | ||
275 | /** | |
276 | * vio_cmo_entitlement_update - Manage system entitlement changes | |
277 | * | |
278 | * @new_entitlement: new system entitlement to attempt to accommodate | |
279 | * | |
280 | * Increases in entitlement will be used to fulfill the spare entitlement | |
281 | * and the rest is given to the excess pool. Decreases, if they are | |
282 | * possible, come from the excess pool and from unused device entitlement | |
283 | * | |
284 | * Returns: 0 on success, -ENOMEM when change can not be made | |
285 | */ | |
286 | int vio_cmo_entitlement_update(size_t new_entitlement) | |
287 | { | |
288 | struct vio_dev *viodev; | |
289 | struct vio_cmo_dev_entry *dev_ent; | |
290 | unsigned long flags; | |
291 | size_t avail, delta, tmp; | |
292 | ||
293 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
294 | ||
295 | /* Entitlement increases */ | |
296 | if (new_entitlement > vio_cmo.entitled) { | |
297 | delta = new_entitlement - vio_cmo.entitled; | |
298 | ||
299 | /* Fulfill spare allocation */ | |
300 | if (vio_cmo.spare < VIO_CMO_MIN_ENT) { | |
301 | tmp = min(delta, (VIO_CMO_MIN_ENT - vio_cmo.spare)); | |
302 | vio_cmo.spare += tmp; | |
303 | vio_cmo.reserve.size += tmp; | |
304 | delta -= tmp; | |
305 | } | |
306 | ||
307 | /* Remaining new allocation goes to the excess pool */ | |
308 | vio_cmo.entitled += delta; | |
309 | vio_cmo.excess.size += delta; | |
310 | vio_cmo.excess.free += delta; | |
311 | ||
312 | goto out; | |
313 | } | |
314 | ||
315 | /* Entitlement decreases */ | |
316 | delta = vio_cmo.entitled - new_entitlement; | |
317 | avail = vio_cmo.excess.free; | |
318 | ||
319 | /* | |
320 | * Need to check how much unused entitlement each device can | |
321 | * sacrifice to fulfill entitlement change. | |
322 | */ | |
323 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) { | |
324 | if (avail >= delta) | |
325 | break; | |
326 | ||
327 | viodev = dev_ent->viodev; | |
328 | if ((viodev->cmo.entitled > viodev->cmo.allocated) && | |
329 | (viodev->cmo.entitled > VIO_CMO_MIN_ENT)) | |
330 | avail += viodev->cmo.entitled - | |
331 | max_t(size_t, viodev->cmo.allocated, | |
332 | VIO_CMO_MIN_ENT); | |
333 | } | |
334 | ||
335 | if (delta <= avail) { | |
336 | vio_cmo.entitled -= delta; | |
337 | ||
338 | /* Take entitlement from the excess pool first */ | |
339 | tmp = min(vio_cmo.excess.free, delta); | |
340 | vio_cmo.excess.size -= tmp; | |
341 | vio_cmo.excess.free -= tmp; | |
342 | delta -= tmp; | |
343 | ||
344 | /* | |
345 | * Remove all but VIO_CMO_MIN_ENT bytes from devices | |
346 | * until entitlement change is served | |
347 | */ | |
348 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) { | |
349 | if (!delta) | |
350 | break; | |
351 | ||
352 | viodev = dev_ent->viodev; | |
353 | tmp = 0; | |
354 | if ((viodev->cmo.entitled > viodev->cmo.allocated) && | |
355 | (viodev->cmo.entitled > VIO_CMO_MIN_ENT)) | |
356 | tmp = viodev->cmo.entitled - | |
357 | max_t(size_t, viodev->cmo.allocated, | |
358 | VIO_CMO_MIN_ENT); | |
359 | viodev->cmo.entitled -= min(tmp, delta); | |
360 | delta -= min(tmp, delta); | |
361 | } | |
362 | } else { | |
363 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
364 | return -ENOMEM; | |
365 | } | |
366 | ||
367 | out: | |
368 | schedule_delayed_work(&vio_cmo.balance_q, 0); | |
369 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
370 | return 0; | |
371 | } | |
372 | ||
373 | /** | |
374 | * vio_cmo_balance - Balance entitlement among devices | |
375 | * | |
376 | * @work: work queue structure for this operation | |
377 | * | |
378 | * Any system entitlement above the minimum needed for devices, or | |
379 | * already allocated to devices, can be distributed to the devices. | |
380 | * The list of devices is iterated through to recalculate the desired | |
381 | * entitlement level and to determine how much entitlement above the | |
382 | * minimum entitlement is allocated to devices. | |
383 | * | |
384 | * Small chunks of the available entitlement are given to devices until | |
385 | * their requirements are fulfilled or there is no entitlement left to give. | |
386 | * Upon completion sizes of the reserve and excess pools are calculated. | |
387 | * | |
388 | * The system minimum entitlement level is also recalculated here. | |
389 | * Entitlement will be reserved for devices even after vio_bus_remove to | |
390 | * accommodate reloading the driver. The OF tree is walked to count the | |
391 | * number of devices present and this will remove entitlement for devices | |
392 | * that have actually left the system after having vio_bus_remove called. | |
393 | */ | |
394 | static void vio_cmo_balance(struct work_struct *work) | |
395 | { | |
396 | struct vio_cmo *cmo; | |
397 | struct vio_dev *viodev; | |
398 | struct vio_cmo_dev_entry *dev_ent; | |
399 | unsigned long flags; | |
400 | size_t avail = 0, level, chunk, need; | |
401 | int devcount = 0, fulfilled; | |
402 | ||
403 | cmo = container_of(work, struct vio_cmo, balance_q.work); | |
404 | ||
405 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
406 | ||
407 | /* Calculate minimum entitlement and fulfill spare */ | |
408 | cmo->min = vio_cmo_num_OF_devs() * VIO_CMO_MIN_ENT; | |
409 | BUG_ON(cmo->min > cmo->entitled); | |
410 | cmo->spare = min_t(size_t, VIO_CMO_MIN_ENT, (cmo->entitled - cmo->min)); | |
411 | cmo->min += cmo->spare; | |
412 | cmo->desired = cmo->min; | |
413 | ||
414 | /* | |
415 | * Determine how much entitlement is available and reset device | |
416 | * entitlements | |
417 | */ | |
418 | avail = cmo->entitled - cmo->spare; | |
419 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) { | |
420 | viodev = dev_ent->viodev; | |
421 | devcount++; | |
422 | viodev->cmo.entitled = VIO_CMO_MIN_ENT; | |
423 | cmo->desired += (viodev->cmo.desired - VIO_CMO_MIN_ENT); | |
424 | avail -= max_t(size_t, viodev->cmo.allocated, VIO_CMO_MIN_ENT); | |
425 | } | |
426 | ||
427 | /* | |
428 | * Having provided each device with the minimum entitlement, loop | |
429 | * over the devices portioning out the remaining entitlement | |
430 | * until there is nothing left. | |
431 | */ | |
432 | level = VIO_CMO_MIN_ENT; | |
433 | while (avail) { | |
434 | fulfilled = 0; | |
435 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) { | |
436 | viodev = dev_ent->viodev; | |
437 | ||
438 | if (viodev->cmo.desired <= level) { | |
439 | fulfilled++; | |
440 | continue; | |
441 | } | |
442 | ||
443 | /* | |
444 | * Give the device up to VIO_CMO_BALANCE_CHUNK | |
445 | * bytes of entitlement, but do not exceed the | |
446 | * desired level of entitlement for the device. | |
447 | */ | |
448 | chunk = min_t(size_t, avail, VIO_CMO_BALANCE_CHUNK); | |
449 | chunk = min(chunk, (viodev->cmo.desired - | |
450 | viodev->cmo.entitled)); | |
451 | viodev->cmo.entitled += chunk; | |
452 | ||
453 | /* | |
454 | * If the memory for this entitlement increase was | |
455 | * already allocated to the device it does not come | |
456 | * from the available pool being portioned out. | |
457 | */ | |
458 | need = max(viodev->cmo.allocated, viodev->cmo.entitled)- | |
459 | max(viodev->cmo.allocated, level); | |
460 | avail -= need; | |
461 | ||
462 | } | |
463 | if (fulfilled == devcount) | |
464 | break; | |
465 | level += VIO_CMO_BALANCE_CHUNK; | |
466 | } | |
467 | ||
468 | /* Calculate new reserve and excess pool sizes */ | |
469 | cmo->reserve.size = cmo->min; | |
470 | cmo->excess.free = 0; | |
471 | cmo->excess.size = 0; | |
472 | need = 0; | |
473 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) { | |
474 | viodev = dev_ent->viodev; | |
475 | /* Calculated reserve size above the minimum entitlement */ | |
476 | if (viodev->cmo.entitled) | |
477 | cmo->reserve.size += (viodev->cmo.entitled - | |
478 | VIO_CMO_MIN_ENT); | |
479 | /* Calculated used excess entitlement */ | |
480 | if (viodev->cmo.allocated > viodev->cmo.entitled) | |
481 | need += viodev->cmo.allocated - viodev->cmo.entitled; | |
482 | } | |
483 | cmo->excess.size = cmo->entitled - cmo->reserve.size; | |
484 | cmo->excess.free = cmo->excess.size - need; | |
485 | ||
bf6aede7 | 486 | cancel_delayed_work(to_delayed_work(work)); |
a90ab95a RJ |
487 | spin_unlock_irqrestore(&vio_cmo.lock, flags); |
488 | } | |
489 | ||
490 | static void *vio_dma_iommu_alloc_coherent(struct device *dev, size_t size, | |
491 | dma_addr_t *dma_handle, gfp_t flag) | |
492 | { | |
493 | struct vio_dev *viodev = to_vio_dev(dev); | |
494 | void *ret; | |
495 | ||
69b052e8 | 496 | if (vio_cmo_alloc(viodev, roundup(size, PAGE_SIZE))) { |
a90ab95a RJ |
497 | atomic_inc(&viodev->cmo.allocs_failed); |
498 | return NULL; | |
499 | } | |
500 | ||
501 | ret = dma_iommu_ops.alloc_coherent(dev, size, dma_handle, flag); | |
502 | if (unlikely(ret == NULL)) { | |
69b052e8 | 503 | vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE)); |
a90ab95a RJ |
504 | atomic_inc(&viodev->cmo.allocs_failed); |
505 | } | |
506 | ||
507 | return ret; | |
508 | } | |
509 | ||
510 | static void vio_dma_iommu_free_coherent(struct device *dev, size_t size, | |
511 | void *vaddr, dma_addr_t dma_handle) | |
512 | { | |
513 | struct vio_dev *viodev = to_vio_dev(dev); | |
514 | ||
515 | dma_iommu_ops.free_coherent(dev, size, vaddr, dma_handle); | |
516 | ||
69b052e8 | 517 | vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE)); |
a90ab95a RJ |
518 | } |
519 | ||
f9226d57 MN |
520 | static dma_addr_t vio_dma_iommu_map_page(struct device *dev, struct page *page, |
521 | unsigned long offset, size_t size, | |
522 | enum dma_data_direction direction, | |
523 | struct dma_attrs *attrs) | |
a90ab95a RJ |
524 | { |
525 | struct vio_dev *viodev = to_vio_dev(dev); | |
526 | dma_addr_t ret = DMA_ERROR_CODE; | |
527 | ||
528 | if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE))) { | |
529 | atomic_inc(&viodev->cmo.allocs_failed); | |
530 | return ret; | |
531 | } | |
532 | ||
f9226d57 | 533 | ret = dma_iommu_ops.map_page(dev, page, offset, size, direction, attrs); |
0764bf63 | 534 | if (unlikely(dma_mapping_error(dev, ret))) { |
a90ab95a RJ |
535 | vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE)); |
536 | atomic_inc(&viodev->cmo.allocs_failed); | |
537 | } | |
538 | ||
539 | return ret; | |
540 | } | |
541 | ||
f9226d57 MN |
542 | static void vio_dma_iommu_unmap_page(struct device *dev, dma_addr_t dma_handle, |
543 | size_t size, | |
544 | enum dma_data_direction direction, | |
545 | struct dma_attrs *attrs) | |
a90ab95a RJ |
546 | { |
547 | struct vio_dev *viodev = to_vio_dev(dev); | |
548 | ||
f9226d57 | 549 | dma_iommu_ops.unmap_page(dev, dma_handle, size, direction, attrs); |
a90ab95a RJ |
550 | |
551 | vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE)); | |
552 | } | |
553 | ||
554 | static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist, | |
555 | int nelems, enum dma_data_direction direction, | |
556 | struct dma_attrs *attrs) | |
557 | { | |
558 | struct vio_dev *viodev = to_vio_dev(dev); | |
559 | struct scatterlist *sgl; | |
560 | int ret, count = 0; | |
561 | size_t alloc_size = 0; | |
562 | ||
563 | for (sgl = sglist; count < nelems; count++, sgl++) | |
564 | alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE); | |
565 | ||
566 | if (vio_cmo_alloc(viodev, alloc_size)) { | |
567 | atomic_inc(&viodev->cmo.allocs_failed); | |
568 | return 0; | |
569 | } | |
570 | ||
571 | ret = dma_iommu_ops.map_sg(dev, sglist, nelems, direction, attrs); | |
572 | ||
573 | if (unlikely(!ret)) { | |
574 | vio_cmo_dealloc(viodev, alloc_size); | |
575 | atomic_inc(&viodev->cmo.allocs_failed); | |
69b052e8 | 576 | return ret; |
a90ab95a RJ |
577 | } |
578 | ||
579 | for (sgl = sglist, count = 0; count < ret; count++, sgl++) | |
580 | alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE); | |
581 | if (alloc_size) | |
582 | vio_cmo_dealloc(viodev, alloc_size); | |
583 | ||
584 | return ret; | |
585 | } | |
586 | ||
587 | static void vio_dma_iommu_unmap_sg(struct device *dev, | |
588 | struct scatterlist *sglist, int nelems, | |
589 | enum dma_data_direction direction, | |
590 | struct dma_attrs *attrs) | |
591 | { | |
592 | struct vio_dev *viodev = to_vio_dev(dev); | |
593 | struct scatterlist *sgl; | |
594 | size_t alloc_size = 0; | |
595 | int count = 0; | |
596 | ||
597 | for (sgl = sglist; count < nelems; count++, sgl++) | |
598 | alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE); | |
599 | ||
600 | dma_iommu_ops.unmap_sg(dev, sglist, nelems, direction, attrs); | |
601 | ||
602 | vio_cmo_dealloc(viodev, alloc_size); | |
603 | } | |
604 | ||
45223c54 | 605 | struct dma_map_ops vio_dma_mapping_ops = { |
a90ab95a RJ |
606 | .alloc_coherent = vio_dma_iommu_alloc_coherent, |
607 | .free_coherent = vio_dma_iommu_free_coherent, | |
a90ab95a RJ |
608 | .map_sg = vio_dma_iommu_map_sg, |
609 | .unmap_sg = vio_dma_iommu_unmap_sg, | |
f9226d57 MN |
610 | .map_page = vio_dma_iommu_map_page, |
611 | .unmap_page = vio_dma_iommu_unmap_page, | |
612 | ||
a90ab95a RJ |
613 | }; |
614 | ||
615 | /** | |
616 | * vio_cmo_set_dev_desired - Set desired entitlement for a device | |
617 | * | |
618 | * @viodev: struct vio_dev for device to alter | |
619 | * @new_desired: new desired entitlement level in bytes | |
620 | * | |
621 | * For use by devices to request a change to their entitlement at runtime or | |
622 | * through sysfs. The desired entitlement level is changed and a balancing | |
623 | * of system resources is scheduled to run in the future. | |
624 | */ | |
625 | void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) | |
626 | { | |
627 | unsigned long flags; | |
628 | struct vio_cmo_dev_entry *dev_ent; | |
629 | int found = 0; | |
630 | ||
631 | if (!firmware_has_feature(FW_FEATURE_CMO)) | |
632 | return; | |
633 | ||
634 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
635 | if (desired < VIO_CMO_MIN_ENT) | |
636 | desired = VIO_CMO_MIN_ENT; | |
637 | ||
638 | /* | |
639 | * Changes will not be made for devices not in the device list. | |
640 | * If it is not in the device list, then no driver is loaded | |
641 | * for the device and it can not receive entitlement. | |
642 | */ | |
643 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) | |
644 | if (viodev == dev_ent->viodev) { | |
645 | found = 1; | |
646 | break; | |
647 | } | |
648 | if (!found) | |
649 | return; | |
650 | ||
651 | /* Increase/decrease in desired device entitlement */ | |
652 | if (desired >= viodev->cmo.desired) { | |
653 | /* Just bump the bus and device values prior to a balance*/ | |
654 | vio_cmo.desired += desired - viodev->cmo.desired; | |
655 | viodev->cmo.desired = desired; | |
656 | } else { | |
657 | /* Decrease bus and device values for desired entitlement */ | |
658 | vio_cmo.desired -= viodev->cmo.desired - desired; | |
659 | viodev->cmo.desired = desired; | |
660 | /* | |
661 | * If less entitlement is desired than current entitlement, move | |
662 | * any reserve memory in the change region to the excess pool. | |
663 | */ | |
664 | if (viodev->cmo.entitled > desired) { | |
665 | vio_cmo.reserve.size -= viodev->cmo.entitled - desired; | |
666 | vio_cmo.excess.size += viodev->cmo.entitled - desired; | |
667 | /* | |
668 | * If entitlement moving from the reserve pool to the | |
669 | * excess pool is currently unused, add to the excess | |
670 | * free counter. | |
671 | */ | |
672 | if (viodev->cmo.allocated < viodev->cmo.entitled) | |
673 | vio_cmo.excess.free += viodev->cmo.entitled - | |
674 | max(viodev->cmo.allocated, desired); | |
675 | viodev->cmo.entitled = desired; | |
676 | } | |
677 | } | |
678 | schedule_delayed_work(&vio_cmo.balance_q, 0); | |
679 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
680 | } | |
681 | ||
682 | /** | |
683 | * vio_cmo_bus_probe - Handle CMO specific bus probe activities | |
684 | * | |
685 | * @viodev - Pointer to struct vio_dev for device | |
686 | * | |
687 | * Determine the devices IO memory entitlement needs, attempting | |
688 | * to satisfy the system minimum entitlement at first and scheduling | |
689 | * a balance operation to take care of the rest at a later time. | |
690 | * | |
691 | * Returns: 0 on success, -EINVAL when device doesn't support CMO, and | |
692 | * -ENOMEM when entitlement is not available for device or | |
693 | * device entry. | |
694 | * | |
695 | */ | |
696 | static int vio_cmo_bus_probe(struct vio_dev *viodev) | |
697 | { | |
698 | struct vio_cmo_dev_entry *dev_ent; | |
699 | struct device *dev = &viodev->dev; | |
700 | struct vio_driver *viodrv = to_vio_driver(dev->driver); | |
701 | unsigned long flags; | |
702 | size_t size; | |
703 | ||
704 | /* | |
705 | * Check to see that device has a DMA window and configure | |
706 | * entitlement for the device. | |
707 | */ | |
708 | if (of_get_property(viodev->dev.archdata.of_node, | |
709 | "ibm,my-dma-window", NULL)) { | |
710 | /* Check that the driver is CMO enabled and get desired DMA */ | |
711 | if (!viodrv->get_desired_dma) { | |
712 | dev_err(dev, "%s: device driver does not support CMO\n", | |
713 | __func__); | |
714 | return -EINVAL; | |
715 | } | |
716 | ||
717 | viodev->cmo.desired = IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev)); | |
718 | if (viodev->cmo.desired < VIO_CMO_MIN_ENT) | |
719 | viodev->cmo.desired = VIO_CMO_MIN_ENT; | |
720 | size = VIO_CMO_MIN_ENT; | |
721 | ||
722 | dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry), | |
723 | GFP_KERNEL); | |
724 | if (!dev_ent) | |
725 | return -ENOMEM; | |
726 | ||
727 | dev_ent->viodev = viodev; | |
728 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
729 | list_add(&dev_ent->list, &vio_cmo.device_list); | |
730 | } else { | |
731 | viodev->cmo.desired = 0; | |
732 | size = 0; | |
733 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
734 | } | |
735 | ||
736 | /* | |
737 | * If the needs for vio_cmo.min have not changed since they | |
738 | * were last set, the number of devices in the OF tree has | |
739 | * been constant and the IO memory for this is already in | |
740 | * the reserve pool. | |
741 | */ | |
742 | if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) * | |
743 | VIO_CMO_MIN_ENT)) { | |
744 | /* Updated desired entitlement if device requires it */ | |
745 | if (size) | |
746 | vio_cmo.desired += (viodev->cmo.desired - | |
747 | VIO_CMO_MIN_ENT); | |
748 | } else { | |
749 | size_t tmp; | |
750 | ||
751 | tmp = vio_cmo.spare + vio_cmo.excess.free; | |
752 | if (tmp < size) { | |
753 | dev_err(dev, "%s: insufficient free " | |
754 | "entitlement to add device. " | |
755 | "Need %lu, have %lu\n", __func__, | |
756 | size, (vio_cmo.spare + tmp)); | |
757 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
758 | return -ENOMEM; | |
759 | } | |
760 | ||
761 | /* Use excess pool first to fulfill request */ | |
762 | tmp = min(size, vio_cmo.excess.free); | |
763 | vio_cmo.excess.free -= tmp; | |
764 | vio_cmo.excess.size -= tmp; | |
765 | vio_cmo.reserve.size += tmp; | |
766 | ||
767 | /* Use spare if excess pool was insufficient */ | |
768 | vio_cmo.spare -= size - tmp; | |
769 | ||
770 | /* Update bus accounting */ | |
771 | vio_cmo.min += size; | |
772 | vio_cmo.desired += viodev->cmo.desired; | |
773 | } | |
774 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
775 | return 0; | |
776 | } | |
777 | ||
778 | /** | |
779 | * vio_cmo_bus_remove - Handle CMO specific bus removal activities | |
780 | * | |
781 | * @viodev - Pointer to struct vio_dev for device | |
782 | * | |
783 | * Remove the device from the cmo device list. The minimum entitlement | |
784 | * will be reserved for the device as long as it is in the system. The | |
785 | * rest of the entitlement the device had been allocated will be returned | |
786 | * to the system. | |
787 | */ | |
788 | static void vio_cmo_bus_remove(struct vio_dev *viodev) | |
789 | { | |
790 | struct vio_cmo_dev_entry *dev_ent; | |
791 | unsigned long flags; | |
792 | size_t tmp; | |
793 | ||
794 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
795 | if (viodev->cmo.allocated) { | |
796 | dev_err(&viodev->dev, "%s: device had %lu bytes of IO " | |
797 | "allocated after remove operation.\n", | |
798 | __func__, viodev->cmo.allocated); | |
799 | BUG(); | |
800 | } | |
801 | ||
802 | /* | |
803 | * Remove the device from the device list being maintained for | |
804 | * CMO enabled devices. | |
805 | */ | |
806 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) | |
807 | if (viodev == dev_ent->viodev) { | |
808 | list_del(&dev_ent->list); | |
809 | kfree(dev_ent); | |
810 | break; | |
811 | } | |
812 | ||
813 | /* | |
814 | * Devices may not require any entitlement and they do not need | |
815 | * to be processed. Otherwise, return the device's entitlement | |
816 | * back to the pools. | |
817 | */ | |
818 | if (viodev->cmo.entitled) { | |
819 | /* | |
820 | * This device has not yet left the OF tree, it's | |
821 | * minimum entitlement remains in vio_cmo.min and | |
822 | * vio_cmo.desired | |
823 | */ | |
824 | vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT); | |
825 | ||
826 | /* | |
827 | * Save min allocation for device in reserve as long | |
828 | * as it exists in OF tree as determined by later | |
829 | * balance operation | |
830 | */ | |
831 | viodev->cmo.entitled -= VIO_CMO_MIN_ENT; | |
832 | ||
833 | /* Replenish spare from freed reserve pool */ | |
834 | if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) { | |
835 | tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT - | |
836 | vio_cmo.spare)); | |
837 | vio_cmo.spare += tmp; | |
838 | viodev->cmo.entitled -= tmp; | |
839 | } | |
840 | ||
841 | /* Remaining reserve goes to excess pool */ | |
842 | vio_cmo.excess.size += viodev->cmo.entitled; | |
843 | vio_cmo.excess.free += viodev->cmo.entitled; | |
844 | vio_cmo.reserve.size -= viodev->cmo.entitled; | |
845 | ||
846 | /* | |
847 | * Until the device is removed it will keep a | |
848 | * minimum entitlement; this will guarantee that | |
849 | * a module unload/load will result in a success. | |
850 | */ | |
851 | viodev->cmo.entitled = VIO_CMO_MIN_ENT; | |
852 | viodev->cmo.desired = VIO_CMO_MIN_ENT; | |
853 | atomic_set(&viodev->cmo.allocs_failed, 0); | |
854 | } | |
855 | ||
856 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
857 | } | |
858 | ||
859 | static void vio_cmo_set_dma_ops(struct vio_dev *viodev) | |
860 | { | |
861 | vio_dma_mapping_ops.dma_supported = dma_iommu_ops.dma_supported; | |
862 | viodev->dev.archdata.dma_ops = &vio_dma_mapping_ops; | |
863 | } | |
864 | ||
865 | /** | |
866 | * vio_cmo_bus_init - CMO entitlement initialization at bus init time | |
867 | * | |
868 | * Set up the reserve and excess entitlement pools based on available | |
869 | * system entitlement and the number of devices in the OF tree that | |
870 | * require entitlement in the reserve pool. | |
871 | */ | |
872 | static void vio_cmo_bus_init(void) | |
873 | { | |
874 | struct hvcall_mpp_data mpp_data; | |
875 | int err; | |
876 | ||
877 | memset(&vio_cmo, 0, sizeof(struct vio_cmo)); | |
878 | spin_lock_init(&vio_cmo.lock); | |
879 | INIT_LIST_HEAD(&vio_cmo.device_list); | |
880 | INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance); | |
881 | ||
882 | /* Get current system entitlement */ | |
883 | err = h_get_mpp(&mpp_data); | |
884 | ||
885 | /* | |
886 | * On failure, continue with entitlement set to 0, will panic() | |
887 | * later when spare is reserved. | |
888 | */ | |
889 | if (err != H_SUCCESS) { | |
890 | printk(KERN_ERR "%s: unable to determine system IO "\ | |
891 | "entitlement. (%d)\n", __func__, err); | |
892 | vio_cmo.entitled = 0; | |
893 | } else { | |
894 | vio_cmo.entitled = mpp_data.entitled_mem; | |
895 | } | |
896 | ||
897 | /* Set reservation and check against entitlement */ | |
898 | vio_cmo.spare = VIO_CMO_MIN_ENT; | |
899 | vio_cmo.reserve.size = vio_cmo.spare; | |
900 | vio_cmo.reserve.size += (vio_cmo_num_OF_devs() * | |
901 | VIO_CMO_MIN_ENT); | |
902 | if (vio_cmo.reserve.size > vio_cmo.entitled) { | |
903 | printk(KERN_ERR "%s: insufficient system entitlement\n", | |
904 | __func__); | |
905 | panic("%s: Insufficient system entitlement", __func__); | |
906 | } | |
907 | ||
908 | /* Set the remaining accounting variables */ | |
909 | vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size; | |
910 | vio_cmo.excess.free = vio_cmo.excess.size; | |
911 | vio_cmo.min = vio_cmo.reserve.size; | |
912 | vio_cmo.desired = vio_cmo.reserve.size; | |
913 | } | |
914 | ||
915 | /* sysfs device functions and data structures for CMO */ | |
916 | ||
917 | #define viodev_cmo_rd_attr(name) \ | |
918 | static ssize_t viodev_cmo_##name##_show(struct device *dev, \ | |
919 | struct device_attribute *attr, \ | |
920 | char *buf) \ | |
921 | { \ | |
922 | return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name); \ | |
923 | } | |
924 | ||
925 | static ssize_t viodev_cmo_allocs_failed_show(struct device *dev, | |
926 | struct device_attribute *attr, char *buf) | |
927 | { | |
928 | struct vio_dev *viodev = to_vio_dev(dev); | |
929 | return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed)); | |
930 | } | |
931 | ||
932 | static ssize_t viodev_cmo_allocs_failed_reset(struct device *dev, | |
933 | struct device_attribute *attr, const char *buf, size_t count) | |
934 | { | |
935 | struct vio_dev *viodev = to_vio_dev(dev); | |
936 | atomic_set(&viodev->cmo.allocs_failed, 0); | |
937 | return count; | |
938 | } | |
939 | ||
940 | static ssize_t viodev_cmo_desired_set(struct device *dev, | |
941 | struct device_attribute *attr, const char *buf, size_t count) | |
942 | { | |
943 | struct vio_dev *viodev = to_vio_dev(dev); | |
944 | size_t new_desired; | |
945 | int ret; | |
946 | ||
947 | ret = strict_strtoul(buf, 10, &new_desired); | |
948 | if (ret) | |
949 | return ret; | |
950 | ||
951 | vio_cmo_set_dev_desired(viodev, new_desired); | |
952 | return count; | |
953 | } | |
954 | ||
955 | viodev_cmo_rd_attr(desired); | |
956 | viodev_cmo_rd_attr(entitled); | |
957 | viodev_cmo_rd_attr(allocated); | |
958 | ||
959 | static ssize_t name_show(struct device *, struct device_attribute *, char *); | |
960 | static ssize_t devspec_show(struct device *, struct device_attribute *, char *); | |
961 | static struct device_attribute vio_cmo_dev_attrs[] = { | |
962 | __ATTR_RO(name), | |
963 | __ATTR_RO(devspec), | |
964 | __ATTR(cmo_desired, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH, | |
965 | viodev_cmo_desired_show, viodev_cmo_desired_set), | |
966 | __ATTR(cmo_entitled, S_IRUGO, viodev_cmo_entitled_show, NULL), | |
967 | __ATTR(cmo_allocated, S_IRUGO, viodev_cmo_allocated_show, NULL), | |
968 | __ATTR(cmo_allocs_failed, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH, | |
969 | viodev_cmo_allocs_failed_show, viodev_cmo_allocs_failed_reset), | |
970 | __ATTR_NULL | |
971 | }; | |
972 | ||
973 | /* sysfs bus functions and data structures for CMO */ | |
974 | ||
975 | #define viobus_cmo_rd_attr(name) \ | |
976 | static ssize_t \ | |
977 | viobus_cmo_##name##_show(struct bus_type *bt, char *buf) \ | |
978 | { \ | |
979 | return sprintf(buf, "%lu\n", vio_cmo.name); \ | |
980 | } | |
981 | ||
982 | #define viobus_cmo_pool_rd_attr(name, var) \ | |
983 | static ssize_t \ | |
984 | viobus_cmo_##name##_pool_show_##var(struct bus_type *bt, char *buf) \ | |
985 | { \ | |
986 | return sprintf(buf, "%lu\n", vio_cmo.name.var); \ | |
987 | } | |
988 | ||
989 | static ssize_t viobus_cmo_high_reset(struct bus_type *bt, const char *buf, | |
990 | size_t count) | |
991 | { | |
992 | unsigned long flags; | |
993 | ||
994 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
995 | vio_cmo.high = vio_cmo.curr; | |
996 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
997 | ||
998 | return count; | |
999 | } | |
1000 | ||
1001 | viobus_cmo_rd_attr(entitled); | |
1002 | viobus_cmo_pool_rd_attr(reserve, size); | |
1003 | viobus_cmo_pool_rd_attr(excess, size); | |
1004 | viobus_cmo_pool_rd_attr(excess, free); | |
1005 | viobus_cmo_rd_attr(spare); | |
1006 | viobus_cmo_rd_attr(min); | |
1007 | viobus_cmo_rd_attr(desired); | |
1008 | viobus_cmo_rd_attr(curr); | |
1009 | viobus_cmo_rd_attr(high); | |
1010 | ||
1011 | static struct bus_attribute vio_cmo_bus_attrs[] = { | |
1012 | __ATTR(cmo_entitled, S_IRUGO, viobus_cmo_entitled_show, NULL), | |
1013 | __ATTR(cmo_reserve_size, S_IRUGO, viobus_cmo_reserve_pool_show_size, NULL), | |
1014 | __ATTR(cmo_excess_size, S_IRUGO, viobus_cmo_excess_pool_show_size, NULL), | |
1015 | __ATTR(cmo_excess_free, S_IRUGO, viobus_cmo_excess_pool_show_free, NULL), | |
1016 | __ATTR(cmo_spare, S_IRUGO, viobus_cmo_spare_show, NULL), | |
1017 | __ATTR(cmo_min, S_IRUGO, viobus_cmo_min_show, NULL), | |
1018 | __ATTR(cmo_desired, S_IRUGO, viobus_cmo_desired_show, NULL), | |
1019 | __ATTR(cmo_curr, S_IRUGO, viobus_cmo_curr_show, NULL), | |
1020 | __ATTR(cmo_high, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH, | |
1021 | viobus_cmo_high_show, viobus_cmo_high_reset), | |
1022 | __ATTR_NULL | |
1023 | }; | |
1024 | ||
1025 | static void vio_cmo_sysfs_init(void) | |
1026 | { | |
1027 | vio_bus_type.dev_attrs = vio_cmo_dev_attrs; | |
1028 | vio_bus_type.bus_attrs = vio_cmo_bus_attrs; | |
1029 | } | |
1030 | #else /* CONFIG_PPC_SMLPAR */ | |
1031 | /* Dummy functions for iSeries platform */ | |
1032 | int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; } | |
1033 | void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {} | |
1034 | static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; } | |
1035 | static void vio_cmo_bus_remove(struct vio_dev *viodev) {} | |
1036 | static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {} | |
b9fa49a9 NL |
1037 | static void vio_cmo_bus_init(void) {} |
1038 | static void vio_cmo_sysfs_init(void) { } | |
a90ab95a RJ |
1039 | #endif /* CONFIG_PPC_SMLPAR */ |
1040 | EXPORT_SYMBOL(vio_cmo_entitlement_update); | |
1041 | EXPORT_SYMBOL(vio_cmo_set_dev_desired); | |
1042 | ||
c7f0e8cb SR |
1043 | static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev) |
1044 | { | |
dd9b67ab SR |
1045 | const unsigned char *dma_window; |
1046 | struct iommu_table *tbl; | |
1047 | unsigned long offset, size; | |
1048 | ||
1049 | if (firmware_has_feature(FW_FEATURE_ISERIES)) | |
1050 | return vio_build_iommu_table_iseries(dev); | |
c7f0e8cb | 1051 | |
dd9b67ab SR |
1052 | dma_window = of_get_property(dev->dev.archdata.of_node, |
1053 | "ibm,my-dma-window", NULL); | |
1054 | if (!dma_window) | |
1055 | return NULL; | |
c7f0e8cb | 1056 | |
dd9b67ab | 1057 | tbl = kmalloc(sizeof(*tbl), GFP_KERNEL); |
0f337274 | 1058 | if (tbl == NULL) |
1059 | return NULL; | |
c7f0e8cb | 1060 | |
dd9b67ab SR |
1061 | of_parse_dma_window(dev->dev.archdata.of_node, dma_window, |
1062 | &tbl->it_index, &offset, &size); | |
c7f0e8cb | 1063 | |
dd9b67ab SR |
1064 | /* TCE table size - measured in tce entries */ |
1065 | tbl->it_size = size >> IOMMU_PAGE_SHIFT; | |
1066 | /* offset for VIO should always be 0 */ | |
1067 | tbl->it_offset = offset >> IOMMU_PAGE_SHIFT; | |
1068 | tbl->it_busno = 0; | |
1069 | tbl->it_type = TCE_VB; | |
c7f0e8cb | 1070 | |
dd9b67ab | 1071 | return iommu_init_table(tbl, -1); |
c7f0e8cb | 1072 | } |
1da177e4 | 1073 | |
e10fa773 SR |
1074 | /** |
1075 | * vio_match_device: - Tell if a VIO device has a matching | |
1076 | * VIO device id structure. | |
1077 | * @ids: array of VIO device id structures to search in | |
1078 | * @dev: the VIO device structure to match against | |
1079 | * | |
1080 | * Used by a driver to check whether a VIO device present in the | |
1081 | * system is in its list of supported devices. Returns the matching | |
1082 | * vio_device_id structure or NULL if there is no match. | |
1083 | */ | |
1084 | static const struct vio_device_id *vio_match_device( | |
1085 | const struct vio_device_id *ids, const struct vio_dev *dev) | |
1086 | { | |
1087 | while (ids->type[0] != '\0') { | |
dd721ffd | 1088 | if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) && |
55b61fec | 1089 | of_device_is_compatible(dev->dev.archdata.of_node, |
12d04eef | 1090 | ids->compat)) |
e10fa773 SR |
1091 | return ids; |
1092 | ids++; | |
1093 | } | |
1094 | return NULL; | |
1095 | } | |
1096 | ||
5c0b4b87 SR |
1097 | /* |
1098 | * Convert from struct device to struct vio_dev and pass to driver. | |
1da177e4 | 1099 | * dev->driver has already been set by generic code because vio_bus_match |
5c0b4b87 SR |
1100 | * succeeded. |
1101 | */ | |
1da177e4 LT |
1102 | static int vio_bus_probe(struct device *dev) |
1103 | { | |
1104 | struct vio_dev *viodev = to_vio_dev(dev); | |
1105 | struct vio_driver *viodrv = to_vio_driver(dev->driver); | |
1106 | const struct vio_device_id *id; | |
1107 | int error = -ENODEV; | |
1108 | ||
1da177e4 LT |
1109 | if (!viodrv->probe) |
1110 | return error; | |
1111 | ||
1112 | id = vio_match_device(viodrv->id_table, viodev); | |
a90ab95a RJ |
1113 | if (id) { |
1114 | memset(&viodev->cmo, 0, sizeof(viodev->cmo)); | |
1115 | if (firmware_has_feature(FW_FEATURE_CMO)) { | |
1116 | error = vio_cmo_bus_probe(viodev); | |
1117 | if (error) | |
1118 | return error; | |
1119 | } | |
1da177e4 | 1120 | error = viodrv->probe(viodev, id); |
cd5aeb9f | 1121 | if (error && firmware_has_feature(FW_FEATURE_CMO)) |
a90ab95a RJ |
1122 | vio_cmo_bus_remove(viodev); |
1123 | } | |
1da177e4 LT |
1124 | |
1125 | return error; | |
1126 | } | |
1127 | ||
1128 | /* convert from struct device to struct vio_dev and pass to driver. */ | |
1129 | static int vio_bus_remove(struct device *dev) | |
1130 | { | |
1131 | struct vio_dev *viodev = to_vio_dev(dev); | |
1132 | struct vio_driver *viodrv = to_vio_driver(dev->driver); | |
a90ab95a RJ |
1133 | struct device *devptr; |
1134 | int ret = 1; | |
1135 | ||
1136 | /* | |
1137 | * Hold a reference to the device after the remove function is called | |
1138 | * to allow for CMO accounting cleanup for the device. | |
1139 | */ | |
1140 | devptr = get_device(dev); | |
1da177e4 | 1141 | |
5c0b4b87 | 1142 | if (viodrv->remove) |
a90ab95a | 1143 | ret = viodrv->remove(viodev); |
1da177e4 | 1144 | |
a90ab95a RJ |
1145 | if (!ret && firmware_has_feature(FW_FEATURE_CMO)) |
1146 | vio_cmo_bus_remove(viodev); | |
1147 | ||
1148 | put_device(devptr); | |
1149 | return ret; | |
1da177e4 LT |
1150 | } |
1151 | ||
1152 | /** | |
1153 | * vio_register_driver: - Register a new vio driver | |
1154 | * @drv: The vio_driver structure to be registered. | |
1155 | */ | |
1156 | int vio_register_driver(struct vio_driver *viodrv) | |
1157 | { | |
e48b1b45 | 1158 | printk(KERN_DEBUG "%s: driver %s registering\n", __func__, |
6fdf5392 | 1159 | viodrv->driver.name); |
1da177e4 LT |
1160 | |
1161 | /* fill in 'struct driver' fields */ | |
1da177e4 | 1162 | viodrv->driver.bus = &vio_bus_type; |
1da177e4 LT |
1163 | |
1164 | return driver_register(&viodrv->driver); | |
1165 | } | |
1166 | EXPORT_SYMBOL(vio_register_driver); | |
1167 | ||
1168 | /** | |
1169 | * vio_unregister_driver - Remove registration of vio driver. | |
1170 | * @driver: The vio_driver struct to be removed form registration | |
1171 | */ | |
1172 | void vio_unregister_driver(struct vio_driver *viodrv) | |
1173 | { | |
1174 | driver_unregister(&viodrv->driver); | |
1175 | } | |
1176 | EXPORT_SYMBOL(vio_unregister_driver); | |
1177 | ||
c7f0e8cb SR |
1178 | /* vio_dev refcount hit 0 */ |
1179 | static void __devinit vio_dev_release(struct device *dev) | |
1180 | { | |
6690faeb MK |
1181 | /* XXX should free TCE table */ |
1182 | of_node_put(dev->archdata.of_node); | |
c7f0e8cb SR |
1183 | kfree(to_vio_dev(dev)); |
1184 | } | |
1185 | ||
1da177e4 | 1186 | /** |
e10fa773 SR |
1187 | * vio_register_device_node: - Register a new vio device. |
1188 | * @of_node: The OF node for this device. | |
1da177e4 | 1189 | * |
e10fa773 | 1190 | * Creates and initializes a vio_dev structure from the data in |
12d04eef | 1191 | * of_node and adds it to the list of virtual devices. |
e10fa773 SR |
1192 | * Returns a pointer to the created vio_dev or NULL if node has |
1193 | * NULL device_type or compatible fields. | |
1da177e4 | 1194 | */ |
de7d812d | 1195 | struct vio_dev *vio_register_device_node(struct device_node *of_node) |
1da177e4 | 1196 | { |
e10fa773 | 1197 | struct vio_dev *viodev; |
a7f67bdf | 1198 | const unsigned int *unit_address; |
e10fa773 SR |
1199 | |
1200 | /* we need the 'device_type' property, in order to match with drivers */ | |
1201 | if (of_node->type == NULL) { | |
1202 | printk(KERN_WARNING "%s: node %s missing 'device_type'\n", | |
e48b1b45 | 1203 | __func__, |
e10fa773 SR |
1204 | of_node->name ? of_node->name : "<unknown>"); |
1205 | return NULL; | |
1da177e4 | 1206 | } |
e10fa773 | 1207 | |
e2eb6392 | 1208 | unit_address = of_get_property(of_node, "reg", NULL); |
e10fa773 SR |
1209 | if (unit_address == NULL) { |
1210 | printk(KERN_WARNING "%s: node %s missing 'reg'\n", | |
e48b1b45 | 1211 | __func__, |
e10fa773 SR |
1212 | of_node->name ? of_node->name : "<unknown>"); |
1213 | return NULL; | |
1214 | } | |
1215 | ||
1216 | /* allocate a vio_dev for this node */ | |
1217 | viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL); | |
1218 | if (viodev == NULL) | |
1219 | return NULL; | |
1220 | ||
0ebfff14 | 1221 | viodev->irq = irq_of_parse_and_map(of_node, 0); |
e10fa773 | 1222 | |
aab0d375 | 1223 | dev_set_name(&viodev->dev, "%x", *unit_address); |
e10fa773 SR |
1224 | viodev->name = of_node->name; |
1225 | viodev->type = of_node->type; | |
1226 | viodev->unit_address = *unit_address; | |
1227 | if (firmware_has_feature(FW_FEATURE_ISERIES)) { | |
e2eb6392 | 1228 | unit_address = of_get_property(of_node, |
e10fa773 SR |
1229 | "linux,unit_address", NULL); |
1230 | if (unit_address != NULL) | |
1231 | viodev->unit_address = *unit_address; | |
1232 | } | |
12d04eef | 1233 | viodev->dev.archdata.of_node = of_node_get(of_node); |
a90ab95a RJ |
1234 | |
1235 | if (firmware_has_feature(FW_FEATURE_CMO)) | |
1236 | vio_cmo_set_dma_ops(viodev); | |
1237 | else | |
1238 | viodev->dev.archdata.dma_ops = &dma_iommu_ops; | |
738ef42e | 1239 | set_iommu_table_base(&viodev->dev, vio_build_iommu_table(viodev)); |
8fae0353 | 1240 | set_dev_node(&viodev->dev, of_node_to_nid(of_node)); |
c7f0e8cb SR |
1241 | |
1242 | /* init generic 'struct device' fields: */ | |
1243 | viodev->dev.parent = &vio_bus_device.dev; | |
1244 | viodev->dev.bus = &vio_bus_type; | |
1245 | viodev->dev.release = vio_dev_release; | |
e10fa773 SR |
1246 | |
1247 | /* register with generic device framework */ | |
c7f0e8cb SR |
1248 | if (device_register(&viodev->dev)) { |
1249 | printk(KERN_ERR "%s: failed to register device %s\n", | |
aab0d375 | 1250 | __func__, dev_name(&viodev->dev)); |
e10fa773 SR |
1251 | /* XXX free TCE table */ |
1252 | kfree(viodev); | |
1253 | return NULL; | |
1254 | } | |
1255 | ||
1256 | return viodev; | |
1da177e4 | 1257 | } |
e10fa773 | 1258 | EXPORT_SYMBOL(vio_register_device_node); |
1da177e4 | 1259 | |
1da177e4 LT |
1260 | /** |
1261 | * vio_bus_init: - Initialize the virtual IO bus | |
1262 | */ | |
c7f0e8cb | 1263 | static int __init vio_bus_init(void) |
1da177e4 LT |
1264 | { |
1265 | int err; | |
e10fa773 | 1266 | struct device_node *node_vroot; |
1da177e4 | 1267 | |
a90ab95a RJ |
1268 | if (firmware_has_feature(FW_FEATURE_CMO)) |
1269 | vio_cmo_sysfs_init(); | |
1270 | ||
1da177e4 LT |
1271 | err = bus_register(&vio_bus_type); |
1272 | if (err) { | |
1273 | printk(KERN_ERR "failed to register VIO bus\n"); | |
1274 | return err; | |
1275 | } | |
1276 | ||
5c0b4b87 SR |
1277 | /* |
1278 | * The fake parent of all vio devices, just to give us | |
3e494c80 SR |
1279 | * a nice directory |
1280 | */ | |
ac5b33c9 | 1281 | err = device_register(&vio_bus_device.dev); |
1da177e4 | 1282 | if (err) { |
3e494c80 | 1283 | printk(KERN_WARNING "%s: device_register returned %i\n", |
e48b1b45 | 1284 | __func__, err); |
1da177e4 LT |
1285 | return err; |
1286 | } | |
1287 | ||
a90ab95a RJ |
1288 | if (firmware_has_feature(FW_FEATURE_CMO)) |
1289 | vio_cmo_bus_init(); | |
1290 | ||
30686ba6 | 1291 | node_vroot = of_find_node_by_name(NULL, "vdevice"); |
e10fa773 SR |
1292 | if (node_vroot) { |
1293 | struct device_node *of_node; | |
1294 | ||
1295 | /* | |
1296 | * Create struct vio_devices for each virtual device in | |
1297 | * the device tree. Drivers will associate with them later. | |
1298 | */ | |
1299 | for (of_node = node_vroot->child; of_node != NULL; | |
c5467262 | 1300 | of_node = of_node->sibling) |
e10fa773 | 1301 | vio_register_device_node(of_node); |
30686ba6 | 1302 | of_node_put(node_vroot); |
e10fa773 SR |
1303 | } |
1304 | ||
3e494c80 SR |
1305 | return 0; |
1306 | } | |
c7f0e8cb | 1307 | __initcall(vio_bus_init); |
1da177e4 | 1308 | |
e10fa773 | 1309 | static ssize_t name_show(struct device *dev, |
5c0b4b87 | 1310 | struct device_attribute *attr, char *buf) |
1da177e4 LT |
1311 | { |
1312 | return sprintf(buf, "%s\n", to_vio_dev(dev)->name); | |
1313 | } | |
e10fa773 SR |
1314 | |
1315 | static ssize_t devspec_show(struct device *dev, | |
1316 | struct device_attribute *attr, char *buf) | |
1317 | { | |
12d04eef | 1318 | struct device_node *of_node = dev->archdata.of_node; |
e10fa773 SR |
1319 | |
1320 | return sprintf(buf, "%s\n", of_node ? of_node->full_name : "none"); | |
1321 | } | |
1322 | ||
1323 | static struct device_attribute vio_dev_attrs[] = { | |
1324 | __ATTR_RO(name), | |
1325 | __ATTR_RO(devspec), | |
1326 | __ATTR_NULL | |
1327 | }; | |
1da177e4 | 1328 | |
1da177e4 LT |
1329 | void __devinit vio_unregister_device(struct vio_dev *viodev) |
1330 | { | |
1da177e4 LT |
1331 | device_unregister(&viodev->dev); |
1332 | } | |
1333 | EXPORT_SYMBOL(vio_unregister_device); | |
1334 | ||
1da177e4 LT |
1335 | static int vio_bus_match(struct device *dev, struct device_driver *drv) |
1336 | { | |
1337 | const struct vio_dev *vio_dev = to_vio_dev(dev); | |
1338 | struct vio_driver *vio_drv = to_vio_driver(drv); | |
1339 | const struct vio_device_id *ids = vio_drv->id_table; | |
1da177e4 | 1340 | |
5c0b4b87 | 1341 | return (ids != NULL) && (vio_match_device(ids, vio_dev) != NULL); |
1da177e4 LT |
1342 | } |
1343 | ||
7eff2e7a | 1344 | static int vio_hotplug(struct device *dev, struct kobj_uevent_env *env) |
143dcec2 OH |
1345 | { |
1346 | const struct vio_dev *vio_dev = to_vio_dev(dev); | |
12d04eef | 1347 | struct device_node *dn; |
a7f67bdf | 1348 | const char *cp; |
143dcec2 | 1349 | |
12d04eef | 1350 | dn = dev->archdata.of_node; |
e10fa773 | 1351 | if (!dn) |
143dcec2 | 1352 | return -ENODEV; |
7eff2e7a | 1353 | cp = of_get_property(dn, "compatible", NULL); |
143dcec2 OH |
1354 | if (!cp) |
1355 | return -ENODEV; | |
1356 | ||
7eff2e7a | 1357 | add_uevent_var(env, "MODALIAS=vio:T%sS%s", vio_dev->type, cp); |
143dcec2 OH |
1358 | return 0; |
1359 | } | |
1360 | ||
6fccab26 | 1361 | static struct bus_type vio_bus_type = { |
1da177e4 | 1362 | .name = "vio", |
e10fa773 | 1363 | .dev_attrs = vio_dev_attrs, |
312c004d | 1364 | .uevent = vio_hotplug, |
1da177e4 | 1365 | .match = vio_bus_match, |
2f53a80f RK |
1366 | .probe = vio_bus_probe, |
1367 | .remove = vio_bus_remove, | |
1da177e4 | 1368 | }; |
e10fa773 SR |
1369 | |
1370 | /** | |
1371 | * vio_get_attribute: - get attribute for virtual device | |
1372 | * @vdev: The vio device to get property. | |
1373 | * @which: The property/attribute to be extracted. | |
1374 | * @length: Pointer to length of returned data size (unused if NULL). | |
1375 | * | |
e2eb6392 | 1376 | * Calls prom.c's of_get_property() to return the value of the |
e10fa773 SR |
1377 | * attribute specified by @which |
1378 | */ | |
1379 | const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length) | |
1380 | { | |
e2eb6392 | 1381 | return of_get_property(vdev->dev.archdata.of_node, which, length); |
e10fa773 SR |
1382 | } |
1383 | EXPORT_SYMBOL(vio_get_attribute); | |
c7f0e8cb SR |
1384 | |
1385 | #ifdef CONFIG_PPC_PSERIES | |
1386 | /* vio_find_name() - internal because only vio.c knows how we formatted the | |
1387 | * kobject name | |
c7f0e8cb | 1388 | */ |
c847c853 | 1389 | static struct vio_dev *vio_find_name(const char *name) |
c7f0e8cb | 1390 | { |
c847c853 | 1391 | struct device *found; |
c7f0e8cb | 1392 | |
c847c853 | 1393 | found = bus_find_device_by_name(&vio_bus_type, NULL, name); |
c7f0e8cb SR |
1394 | if (!found) |
1395 | return NULL; | |
1396 | ||
c847c853 | 1397 | return to_vio_dev(found); |
c7f0e8cb SR |
1398 | } |
1399 | ||
1400 | /** | |
1401 | * vio_find_node - find an already-registered vio_dev | |
1402 | * @vnode: device_node of the virtual device we're looking for | |
1403 | */ | |
1404 | struct vio_dev *vio_find_node(struct device_node *vnode) | |
1405 | { | |
a7f67bdf | 1406 | const uint32_t *unit_address; |
aab0d375 | 1407 | char kobj_name[20]; |
c7f0e8cb SR |
1408 | |
1409 | /* construct the kobject name from the device node */ | |
e2eb6392 | 1410 | unit_address = of_get_property(vnode, "reg", NULL); |
c7f0e8cb SR |
1411 | if (!unit_address) |
1412 | return NULL; | |
aab0d375 | 1413 | snprintf(kobj_name, sizeof(kobj_name), "%x", *unit_address); |
c7f0e8cb SR |
1414 | |
1415 | return vio_find_name(kobj_name); | |
1416 | } | |
1417 | EXPORT_SYMBOL(vio_find_node); | |
1418 | ||
1419 | int vio_enable_interrupts(struct vio_dev *dev) | |
1420 | { | |
1421 | int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE); | |
1422 | if (rc != H_SUCCESS) | |
1423 | printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc); | |
1424 | return rc; | |
1425 | } | |
1426 | EXPORT_SYMBOL(vio_enable_interrupts); | |
1427 | ||
1428 | int vio_disable_interrupts(struct vio_dev *dev) | |
1429 | { | |
1430 | int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE); | |
1431 | if (rc != H_SUCCESS) | |
1432 | printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc); | |
1433 | return rc; | |
1434 | } | |
1435 | EXPORT_SYMBOL(vio_disable_interrupts); | |
1436 | #endif /* CONFIG_PPC_PSERIES */ |