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) { | |
732eacc0 | 241 | tmp = min3(spare_needed, reserve_freed, (viodev->cmo.entitled - VIO_CMO_MIN_ENT)); |
a90ab95a RJ |
242 | |
243 | vio_cmo.spare += tmp; | |
244 | viodev->cmo.entitled -= tmp; | |
245 | reserve_freed -= tmp; | |
246 | spare_needed -= tmp; | |
247 | balance = 1; | |
248 | } | |
249 | ||
250 | /* | |
251 | * Increase the reserve pool until the desired allocation is met. | |
252 | * Move an allocation freed from the excess pool into the reserve | |
253 | * pool and schedule a balance operation. | |
254 | */ | |
255 | if (excess_freed && (vio_cmo.desired > vio_cmo.reserve.size)) { | |
256 | tmp = min(excess_freed, (vio_cmo.desired - vio_cmo.reserve.size)); | |
257 | ||
258 | vio_cmo.excess.size -= tmp; | |
259 | vio_cmo.reserve.size += tmp; | |
260 | excess_freed -= tmp; | |
261 | balance = 1; | |
262 | } | |
263 | ||
264 | /* Return memory from the excess pool to that pool */ | |
265 | if (excess_freed) | |
266 | vio_cmo.excess.free += excess_freed; | |
267 | ||
268 | if (balance) | |
269 | schedule_delayed_work(&vio_cmo.balance_q, VIO_CMO_BALANCE_DELAY); | |
270 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
271 | } | |
272 | ||
273 | /** | |
274 | * vio_cmo_entitlement_update - Manage system entitlement changes | |
275 | * | |
276 | * @new_entitlement: new system entitlement to attempt to accommodate | |
277 | * | |
278 | * Increases in entitlement will be used to fulfill the spare entitlement | |
279 | * and the rest is given to the excess pool. Decreases, if they are | |
280 | * possible, come from the excess pool and from unused device entitlement | |
281 | * | |
282 | * Returns: 0 on success, -ENOMEM when change can not be made | |
283 | */ | |
284 | int vio_cmo_entitlement_update(size_t new_entitlement) | |
285 | { | |
286 | struct vio_dev *viodev; | |
287 | struct vio_cmo_dev_entry *dev_ent; | |
288 | unsigned long flags; | |
289 | size_t avail, delta, tmp; | |
290 | ||
291 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
292 | ||
293 | /* Entitlement increases */ | |
294 | if (new_entitlement > vio_cmo.entitled) { | |
295 | delta = new_entitlement - vio_cmo.entitled; | |
296 | ||
297 | /* Fulfill spare allocation */ | |
298 | if (vio_cmo.spare < VIO_CMO_MIN_ENT) { | |
299 | tmp = min(delta, (VIO_CMO_MIN_ENT - vio_cmo.spare)); | |
300 | vio_cmo.spare += tmp; | |
301 | vio_cmo.reserve.size += tmp; | |
302 | delta -= tmp; | |
303 | } | |
304 | ||
305 | /* Remaining new allocation goes to the excess pool */ | |
306 | vio_cmo.entitled += delta; | |
307 | vio_cmo.excess.size += delta; | |
308 | vio_cmo.excess.free += delta; | |
309 | ||
310 | goto out; | |
311 | } | |
312 | ||
313 | /* Entitlement decreases */ | |
314 | delta = vio_cmo.entitled - new_entitlement; | |
315 | avail = vio_cmo.excess.free; | |
316 | ||
317 | /* | |
318 | * Need to check how much unused entitlement each device can | |
319 | * sacrifice to fulfill entitlement change. | |
320 | */ | |
321 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) { | |
322 | if (avail >= delta) | |
323 | break; | |
324 | ||
325 | viodev = dev_ent->viodev; | |
326 | if ((viodev->cmo.entitled > viodev->cmo.allocated) && | |
327 | (viodev->cmo.entitled > VIO_CMO_MIN_ENT)) | |
328 | avail += viodev->cmo.entitled - | |
329 | max_t(size_t, viodev->cmo.allocated, | |
330 | VIO_CMO_MIN_ENT); | |
331 | } | |
332 | ||
333 | if (delta <= avail) { | |
334 | vio_cmo.entitled -= delta; | |
335 | ||
336 | /* Take entitlement from the excess pool first */ | |
337 | tmp = min(vio_cmo.excess.free, delta); | |
338 | vio_cmo.excess.size -= tmp; | |
339 | vio_cmo.excess.free -= tmp; | |
340 | delta -= tmp; | |
341 | ||
342 | /* | |
343 | * Remove all but VIO_CMO_MIN_ENT bytes from devices | |
344 | * until entitlement change is served | |
345 | */ | |
346 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) { | |
347 | if (!delta) | |
348 | break; | |
349 | ||
350 | viodev = dev_ent->viodev; | |
351 | tmp = 0; | |
352 | if ((viodev->cmo.entitled > viodev->cmo.allocated) && | |
353 | (viodev->cmo.entitled > VIO_CMO_MIN_ENT)) | |
354 | tmp = viodev->cmo.entitled - | |
355 | max_t(size_t, viodev->cmo.allocated, | |
356 | VIO_CMO_MIN_ENT); | |
357 | viodev->cmo.entitled -= min(tmp, delta); | |
358 | delta -= min(tmp, delta); | |
359 | } | |
360 | } else { | |
361 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
362 | return -ENOMEM; | |
363 | } | |
364 | ||
365 | out: | |
366 | schedule_delayed_work(&vio_cmo.balance_q, 0); | |
367 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
368 | return 0; | |
369 | } | |
370 | ||
371 | /** | |
372 | * vio_cmo_balance - Balance entitlement among devices | |
373 | * | |
374 | * @work: work queue structure for this operation | |
375 | * | |
376 | * Any system entitlement above the minimum needed for devices, or | |
377 | * already allocated to devices, can be distributed to the devices. | |
378 | * The list of devices is iterated through to recalculate the desired | |
379 | * entitlement level and to determine how much entitlement above the | |
380 | * minimum entitlement is allocated to devices. | |
381 | * | |
382 | * Small chunks of the available entitlement are given to devices until | |
383 | * their requirements are fulfilled or there is no entitlement left to give. | |
384 | * Upon completion sizes of the reserve and excess pools are calculated. | |
385 | * | |
386 | * The system minimum entitlement level is also recalculated here. | |
387 | * Entitlement will be reserved for devices even after vio_bus_remove to | |
388 | * accommodate reloading the driver. The OF tree is walked to count the | |
389 | * number of devices present and this will remove entitlement for devices | |
390 | * that have actually left the system after having vio_bus_remove called. | |
391 | */ | |
392 | static void vio_cmo_balance(struct work_struct *work) | |
393 | { | |
394 | struct vio_cmo *cmo; | |
395 | struct vio_dev *viodev; | |
396 | struct vio_cmo_dev_entry *dev_ent; | |
397 | unsigned long flags; | |
398 | size_t avail = 0, level, chunk, need; | |
399 | int devcount = 0, fulfilled; | |
400 | ||
401 | cmo = container_of(work, struct vio_cmo, balance_q.work); | |
402 | ||
403 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
404 | ||
405 | /* Calculate minimum entitlement and fulfill spare */ | |
406 | cmo->min = vio_cmo_num_OF_devs() * VIO_CMO_MIN_ENT; | |
407 | BUG_ON(cmo->min > cmo->entitled); | |
408 | cmo->spare = min_t(size_t, VIO_CMO_MIN_ENT, (cmo->entitled - cmo->min)); | |
409 | cmo->min += cmo->spare; | |
410 | cmo->desired = cmo->min; | |
411 | ||
412 | /* | |
413 | * Determine how much entitlement is available and reset device | |
414 | * entitlements | |
415 | */ | |
416 | avail = cmo->entitled - cmo->spare; | |
417 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) { | |
418 | viodev = dev_ent->viodev; | |
419 | devcount++; | |
420 | viodev->cmo.entitled = VIO_CMO_MIN_ENT; | |
421 | cmo->desired += (viodev->cmo.desired - VIO_CMO_MIN_ENT); | |
422 | avail -= max_t(size_t, viodev->cmo.allocated, VIO_CMO_MIN_ENT); | |
423 | } | |
424 | ||
425 | /* | |
426 | * Having provided each device with the minimum entitlement, loop | |
427 | * over the devices portioning out the remaining entitlement | |
428 | * until there is nothing left. | |
429 | */ | |
430 | level = VIO_CMO_MIN_ENT; | |
431 | while (avail) { | |
432 | fulfilled = 0; | |
433 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) { | |
434 | viodev = dev_ent->viodev; | |
435 | ||
436 | if (viodev->cmo.desired <= level) { | |
437 | fulfilled++; | |
438 | continue; | |
439 | } | |
440 | ||
441 | /* | |
442 | * Give the device up to VIO_CMO_BALANCE_CHUNK | |
443 | * bytes of entitlement, but do not exceed the | |
444 | * desired level of entitlement for the device. | |
445 | */ | |
446 | chunk = min_t(size_t, avail, VIO_CMO_BALANCE_CHUNK); | |
447 | chunk = min(chunk, (viodev->cmo.desired - | |
448 | viodev->cmo.entitled)); | |
449 | viodev->cmo.entitled += chunk; | |
450 | ||
451 | /* | |
452 | * If the memory for this entitlement increase was | |
453 | * already allocated to the device it does not come | |
454 | * from the available pool being portioned out. | |
455 | */ | |
456 | need = max(viodev->cmo.allocated, viodev->cmo.entitled)- | |
457 | max(viodev->cmo.allocated, level); | |
458 | avail -= need; | |
459 | ||
460 | } | |
461 | if (fulfilled == devcount) | |
462 | break; | |
463 | level += VIO_CMO_BALANCE_CHUNK; | |
464 | } | |
465 | ||
466 | /* Calculate new reserve and excess pool sizes */ | |
467 | cmo->reserve.size = cmo->min; | |
468 | cmo->excess.free = 0; | |
469 | cmo->excess.size = 0; | |
470 | need = 0; | |
471 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) { | |
472 | viodev = dev_ent->viodev; | |
473 | /* Calculated reserve size above the minimum entitlement */ | |
474 | if (viodev->cmo.entitled) | |
475 | cmo->reserve.size += (viodev->cmo.entitled - | |
476 | VIO_CMO_MIN_ENT); | |
477 | /* Calculated used excess entitlement */ | |
478 | if (viodev->cmo.allocated > viodev->cmo.entitled) | |
479 | need += viodev->cmo.allocated - viodev->cmo.entitled; | |
480 | } | |
481 | cmo->excess.size = cmo->entitled - cmo->reserve.size; | |
482 | cmo->excess.free = cmo->excess.size - need; | |
483 | ||
bf6aede7 | 484 | cancel_delayed_work(to_delayed_work(work)); |
a90ab95a RJ |
485 | spin_unlock_irqrestore(&vio_cmo.lock, flags); |
486 | } | |
487 | ||
488 | static void *vio_dma_iommu_alloc_coherent(struct device *dev, size_t size, | |
489 | dma_addr_t *dma_handle, gfp_t flag) | |
490 | { | |
491 | struct vio_dev *viodev = to_vio_dev(dev); | |
492 | void *ret; | |
493 | ||
69b052e8 | 494 | if (vio_cmo_alloc(viodev, roundup(size, PAGE_SIZE))) { |
a90ab95a RJ |
495 | atomic_inc(&viodev->cmo.allocs_failed); |
496 | return NULL; | |
497 | } | |
498 | ||
499 | ret = dma_iommu_ops.alloc_coherent(dev, size, dma_handle, flag); | |
500 | if (unlikely(ret == NULL)) { | |
69b052e8 | 501 | vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE)); |
a90ab95a RJ |
502 | atomic_inc(&viodev->cmo.allocs_failed); |
503 | } | |
504 | ||
505 | return ret; | |
506 | } | |
507 | ||
508 | static void vio_dma_iommu_free_coherent(struct device *dev, size_t size, | |
509 | void *vaddr, dma_addr_t dma_handle) | |
510 | { | |
511 | struct vio_dev *viodev = to_vio_dev(dev); | |
512 | ||
513 | dma_iommu_ops.free_coherent(dev, size, vaddr, dma_handle); | |
514 | ||
69b052e8 | 515 | vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE)); |
a90ab95a RJ |
516 | } |
517 | ||
f9226d57 MN |
518 | static dma_addr_t vio_dma_iommu_map_page(struct device *dev, struct page *page, |
519 | unsigned long offset, size_t size, | |
520 | enum dma_data_direction direction, | |
521 | struct dma_attrs *attrs) | |
a90ab95a RJ |
522 | { |
523 | struct vio_dev *viodev = to_vio_dev(dev); | |
524 | dma_addr_t ret = DMA_ERROR_CODE; | |
525 | ||
526 | if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE))) { | |
527 | atomic_inc(&viodev->cmo.allocs_failed); | |
528 | return ret; | |
529 | } | |
530 | ||
f9226d57 | 531 | ret = dma_iommu_ops.map_page(dev, page, offset, size, direction, attrs); |
0764bf63 | 532 | if (unlikely(dma_mapping_error(dev, ret))) { |
a90ab95a RJ |
533 | vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE)); |
534 | atomic_inc(&viodev->cmo.allocs_failed); | |
535 | } | |
536 | ||
537 | return ret; | |
538 | } | |
539 | ||
f9226d57 MN |
540 | static void vio_dma_iommu_unmap_page(struct device *dev, dma_addr_t dma_handle, |
541 | size_t size, | |
542 | enum dma_data_direction direction, | |
543 | struct dma_attrs *attrs) | |
a90ab95a RJ |
544 | { |
545 | struct vio_dev *viodev = to_vio_dev(dev); | |
546 | ||
f9226d57 | 547 | dma_iommu_ops.unmap_page(dev, dma_handle, size, direction, attrs); |
a90ab95a RJ |
548 | |
549 | vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE)); | |
550 | } | |
551 | ||
552 | static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist, | |
553 | int nelems, enum dma_data_direction direction, | |
554 | struct dma_attrs *attrs) | |
555 | { | |
556 | struct vio_dev *viodev = to_vio_dev(dev); | |
557 | struct scatterlist *sgl; | |
558 | int ret, count = 0; | |
559 | size_t alloc_size = 0; | |
560 | ||
561 | for (sgl = sglist; count < nelems; count++, sgl++) | |
562 | alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE); | |
563 | ||
564 | if (vio_cmo_alloc(viodev, alloc_size)) { | |
565 | atomic_inc(&viodev->cmo.allocs_failed); | |
566 | return 0; | |
567 | } | |
568 | ||
569 | ret = dma_iommu_ops.map_sg(dev, sglist, nelems, direction, attrs); | |
570 | ||
571 | if (unlikely(!ret)) { | |
572 | vio_cmo_dealloc(viodev, alloc_size); | |
573 | atomic_inc(&viodev->cmo.allocs_failed); | |
69b052e8 | 574 | return ret; |
a90ab95a RJ |
575 | } |
576 | ||
577 | for (sgl = sglist, count = 0; count < ret; count++, sgl++) | |
578 | alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE); | |
579 | if (alloc_size) | |
580 | vio_cmo_dealloc(viodev, alloc_size); | |
581 | ||
582 | return ret; | |
583 | } | |
584 | ||
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) | |
589 | { | |
590 | struct vio_dev *viodev = to_vio_dev(dev); | |
591 | struct scatterlist *sgl; | |
592 | size_t alloc_size = 0; | |
593 | int count = 0; | |
594 | ||
595 | for (sgl = sglist; count < nelems; count++, sgl++) | |
596 | alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE); | |
597 | ||
598 | dma_iommu_ops.unmap_sg(dev, sglist, nelems, direction, attrs); | |
599 | ||
600 | vio_cmo_dealloc(viodev, alloc_size); | |
601 | } | |
602 | ||
6d283d78 NA |
603 | static int vio_dma_iommu_dma_supported(struct device *dev, u64 mask) |
604 | { | |
605 | return dma_iommu_ops.dma_supported(dev, mask); | |
606 | } | |
607 | ||
d24f9c69 MM |
608 | static u64 vio_dma_get_required_mask(struct device *dev) |
609 | { | |
610 | return dma_iommu_ops.get_required_mask(dev); | |
611 | } | |
612 | ||
45223c54 | 613 | struct dma_map_ops vio_dma_mapping_ops = { |
a90ab95a RJ |
614 | .alloc_coherent = vio_dma_iommu_alloc_coherent, |
615 | .free_coherent = vio_dma_iommu_free_coherent, | |
a90ab95a RJ |
616 | .map_sg = vio_dma_iommu_map_sg, |
617 | .unmap_sg = vio_dma_iommu_unmap_sg, | |
f9226d57 MN |
618 | .map_page = vio_dma_iommu_map_page, |
619 | .unmap_page = vio_dma_iommu_unmap_page, | |
6d283d78 | 620 | .dma_supported = vio_dma_iommu_dma_supported, |
d24f9c69 | 621 | .get_required_mask = vio_dma_get_required_mask, |
a90ab95a RJ |
622 | }; |
623 | ||
624 | /** | |
625 | * vio_cmo_set_dev_desired - Set desired entitlement for a device | |
626 | * | |
627 | * @viodev: struct vio_dev for device to alter | |
628 | * @new_desired: new desired entitlement level in bytes | |
629 | * | |
630 | * For use by devices to request a change to their entitlement at runtime or | |
631 | * through sysfs. The desired entitlement level is changed and a balancing | |
632 | * of system resources is scheduled to run in the future. | |
633 | */ | |
634 | void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) | |
635 | { | |
636 | unsigned long flags; | |
637 | struct vio_cmo_dev_entry *dev_ent; | |
638 | int found = 0; | |
639 | ||
640 | if (!firmware_has_feature(FW_FEATURE_CMO)) | |
641 | return; | |
642 | ||
643 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
644 | if (desired < VIO_CMO_MIN_ENT) | |
645 | desired = VIO_CMO_MIN_ENT; | |
646 | ||
647 | /* | |
648 | * Changes will not be made for devices not in the device list. | |
649 | * If it is not in the device list, then no driver is loaded | |
650 | * for the device and it can not receive entitlement. | |
651 | */ | |
652 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) | |
653 | if (viodev == dev_ent->viodev) { | |
654 | found = 1; | |
655 | break; | |
656 | } | |
f6d8c8bb JL |
657 | if (!found) { |
658 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
a90ab95a | 659 | return; |
f6d8c8bb | 660 | } |
a90ab95a RJ |
661 | |
662 | /* Increase/decrease in desired device entitlement */ | |
663 | if (desired >= viodev->cmo.desired) { | |
664 | /* Just bump the bus and device values prior to a balance*/ | |
665 | vio_cmo.desired += desired - viodev->cmo.desired; | |
666 | viodev->cmo.desired = desired; | |
667 | } else { | |
668 | /* Decrease bus and device values for desired entitlement */ | |
669 | vio_cmo.desired -= viodev->cmo.desired - desired; | |
670 | viodev->cmo.desired = desired; | |
671 | /* | |
672 | * If less entitlement is desired than current entitlement, move | |
673 | * any reserve memory in the change region to the excess pool. | |
674 | */ | |
675 | if (viodev->cmo.entitled > desired) { | |
676 | vio_cmo.reserve.size -= viodev->cmo.entitled - desired; | |
677 | vio_cmo.excess.size += viodev->cmo.entitled - desired; | |
678 | /* | |
679 | * If entitlement moving from the reserve pool to the | |
680 | * excess pool is currently unused, add to the excess | |
681 | * free counter. | |
682 | */ | |
683 | if (viodev->cmo.allocated < viodev->cmo.entitled) | |
684 | vio_cmo.excess.free += viodev->cmo.entitled - | |
685 | max(viodev->cmo.allocated, desired); | |
686 | viodev->cmo.entitled = desired; | |
687 | } | |
688 | } | |
689 | schedule_delayed_work(&vio_cmo.balance_q, 0); | |
690 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
691 | } | |
692 | ||
693 | /** | |
694 | * vio_cmo_bus_probe - Handle CMO specific bus probe activities | |
695 | * | |
696 | * @viodev - Pointer to struct vio_dev for device | |
697 | * | |
698 | * Determine the devices IO memory entitlement needs, attempting | |
699 | * to satisfy the system minimum entitlement at first and scheduling | |
700 | * a balance operation to take care of the rest at a later time. | |
701 | * | |
702 | * Returns: 0 on success, -EINVAL when device doesn't support CMO, and | |
703 | * -ENOMEM when entitlement is not available for device or | |
704 | * device entry. | |
705 | * | |
706 | */ | |
707 | static int vio_cmo_bus_probe(struct vio_dev *viodev) | |
708 | { | |
709 | struct vio_cmo_dev_entry *dev_ent; | |
710 | struct device *dev = &viodev->dev; | |
711 | struct vio_driver *viodrv = to_vio_driver(dev->driver); | |
712 | unsigned long flags; | |
713 | size_t size; | |
714 | ||
715 | /* | |
716 | * Check to see that device has a DMA window and configure | |
717 | * entitlement for the device. | |
718 | */ | |
58f9b0b0 | 719 | if (of_get_property(viodev->dev.of_node, |
a90ab95a RJ |
720 | "ibm,my-dma-window", NULL)) { |
721 | /* Check that the driver is CMO enabled and get desired DMA */ | |
722 | if (!viodrv->get_desired_dma) { | |
723 | dev_err(dev, "%s: device driver does not support CMO\n", | |
724 | __func__); | |
725 | return -EINVAL; | |
726 | } | |
727 | ||
728 | viodev->cmo.desired = IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev)); | |
729 | if (viodev->cmo.desired < VIO_CMO_MIN_ENT) | |
730 | viodev->cmo.desired = VIO_CMO_MIN_ENT; | |
731 | size = VIO_CMO_MIN_ENT; | |
732 | ||
733 | dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry), | |
734 | GFP_KERNEL); | |
735 | if (!dev_ent) | |
736 | return -ENOMEM; | |
737 | ||
738 | dev_ent->viodev = viodev; | |
739 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
740 | list_add(&dev_ent->list, &vio_cmo.device_list); | |
741 | } else { | |
742 | viodev->cmo.desired = 0; | |
743 | size = 0; | |
744 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
745 | } | |
746 | ||
747 | /* | |
748 | * If the needs for vio_cmo.min have not changed since they | |
749 | * were last set, the number of devices in the OF tree has | |
750 | * been constant and the IO memory for this is already in | |
751 | * the reserve pool. | |
752 | */ | |
753 | if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) * | |
754 | VIO_CMO_MIN_ENT)) { | |
755 | /* Updated desired entitlement if device requires it */ | |
756 | if (size) | |
757 | vio_cmo.desired += (viodev->cmo.desired - | |
758 | VIO_CMO_MIN_ENT); | |
759 | } else { | |
760 | size_t tmp; | |
761 | ||
762 | tmp = vio_cmo.spare + vio_cmo.excess.free; | |
763 | if (tmp < size) { | |
764 | dev_err(dev, "%s: insufficient free " | |
765 | "entitlement to add device. " | |
766 | "Need %lu, have %lu\n", __func__, | |
767 | size, (vio_cmo.spare + tmp)); | |
768 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
769 | return -ENOMEM; | |
770 | } | |
771 | ||
772 | /* Use excess pool first to fulfill request */ | |
773 | tmp = min(size, vio_cmo.excess.free); | |
774 | vio_cmo.excess.free -= tmp; | |
775 | vio_cmo.excess.size -= tmp; | |
776 | vio_cmo.reserve.size += tmp; | |
777 | ||
778 | /* Use spare if excess pool was insufficient */ | |
779 | vio_cmo.spare -= size - tmp; | |
780 | ||
781 | /* Update bus accounting */ | |
782 | vio_cmo.min += size; | |
783 | vio_cmo.desired += viodev->cmo.desired; | |
784 | } | |
785 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
786 | return 0; | |
787 | } | |
788 | ||
789 | /** | |
790 | * vio_cmo_bus_remove - Handle CMO specific bus removal activities | |
791 | * | |
792 | * @viodev - Pointer to struct vio_dev for device | |
793 | * | |
794 | * Remove the device from the cmo device list. The minimum entitlement | |
795 | * will be reserved for the device as long as it is in the system. The | |
796 | * rest of the entitlement the device had been allocated will be returned | |
797 | * to the system. | |
798 | */ | |
799 | static void vio_cmo_bus_remove(struct vio_dev *viodev) | |
800 | { | |
801 | struct vio_cmo_dev_entry *dev_ent; | |
802 | unsigned long flags; | |
803 | size_t tmp; | |
804 | ||
805 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
806 | if (viodev->cmo.allocated) { | |
807 | dev_err(&viodev->dev, "%s: device had %lu bytes of IO " | |
808 | "allocated after remove operation.\n", | |
809 | __func__, viodev->cmo.allocated); | |
810 | BUG(); | |
811 | } | |
812 | ||
813 | /* | |
814 | * Remove the device from the device list being maintained for | |
815 | * CMO enabled devices. | |
816 | */ | |
817 | list_for_each_entry(dev_ent, &vio_cmo.device_list, list) | |
818 | if (viodev == dev_ent->viodev) { | |
819 | list_del(&dev_ent->list); | |
820 | kfree(dev_ent); | |
821 | break; | |
822 | } | |
823 | ||
824 | /* | |
825 | * Devices may not require any entitlement and they do not need | |
826 | * to be processed. Otherwise, return the device's entitlement | |
827 | * back to the pools. | |
828 | */ | |
829 | if (viodev->cmo.entitled) { | |
830 | /* | |
831 | * This device has not yet left the OF tree, it's | |
832 | * minimum entitlement remains in vio_cmo.min and | |
833 | * vio_cmo.desired | |
834 | */ | |
835 | vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT); | |
836 | ||
837 | /* | |
838 | * Save min allocation for device in reserve as long | |
839 | * as it exists in OF tree as determined by later | |
840 | * balance operation | |
841 | */ | |
842 | viodev->cmo.entitled -= VIO_CMO_MIN_ENT; | |
843 | ||
844 | /* Replenish spare from freed reserve pool */ | |
845 | if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) { | |
846 | tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT - | |
847 | vio_cmo.spare)); | |
848 | vio_cmo.spare += tmp; | |
849 | viodev->cmo.entitled -= tmp; | |
850 | } | |
851 | ||
852 | /* Remaining reserve goes to excess pool */ | |
853 | vio_cmo.excess.size += viodev->cmo.entitled; | |
854 | vio_cmo.excess.free += viodev->cmo.entitled; | |
855 | vio_cmo.reserve.size -= viodev->cmo.entitled; | |
856 | ||
857 | /* | |
858 | * Until the device is removed it will keep a | |
859 | * minimum entitlement; this will guarantee that | |
860 | * a module unload/load will result in a success. | |
861 | */ | |
862 | viodev->cmo.entitled = VIO_CMO_MIN_ENT; | |
863 | viodev->cmo.desired = VIO_CMO_MIN_ENT; | |
864 | atomic_set(&viodev->cmo.allocs_failed, 0); | |
865 | } | |
866 | ||
867 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
868 | } | |
869 | ||
870 | static void vio_cmo_set_dma_ops(struct vio_dev *viodev) | |
871 | { | |
6d283d78 | 872 | set_dma_ops(&viodev->dev, &vio_dma_mapping_ops); |
a90ab95a RJ |
873 | } |
874 | ||
875 | /** | |
876 | * vio_cmo_bus_init - CMO entitlement initialization at bus init time | |
877 | * | |
878 | * Set up the reserve and excess entitlement pools based on available | |
879 | * system entitlement and the number of devices in the OF tree that | |
880 | * require entitlement in the reserve pool. | |
881 | */ | |
882 | static void vio_cmo_bus_init(void) | |
883 | { | |
884 | struct hvcall_mpp_data mpp_data; | |
885 | int err; | |
886 | ||
887 | memset(&vio_cmo, 0, sizeof(struct vio_cmo)); | |
888 | spin_lock_init(&vio_cmo.lock); | |
889 | INIT_LIST_HEAD(&vio_cmo.device_list); | |
890 | INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance); | |
891 | ||
892 | /* Get current system entitlement */ | |
893 | err = h_get_mpp(&mpp_data); | |
894 | ||
895 | /* | |
896 | * On failure, continue with entitlement set to 0, will panic() | |
897 | * later when spare is reserved. | |
898 | */ | |
899 | if (err != H_SUCCESS) { | |
900 | printk(KERN_ERR "%s: unable to determine system IO "\ | |
901 | "entitlement. (%d)\n", __func__, err); | |
902 | vio_cmo.entitled = 0; | |
903 | } else { | |
904 | vio_cmo.entitled = mpp_data.entitled_mem; | |
905 | } | |
906 | ||
907 | /* Set reservation and check against entitlement */ | |
908 | vio_cmo.spare = VIO_CMO_MIN_ENT; | |
909 | vio_cmo.reserve.size = vio_cmo.spare; | |
910 | vio_cmo.reserve.size += (vio_cmo_num_OF_devs() * | |
911 | VIO_CMO_MIN_ENT); | |
912 | if (vio_cmo.reserve.size > vio_cmo.entitled) { | |
913 | printk(KERN_ERR "%s: insufficient system entitlement\n", | |
914 | __func__); | |
915 | panic("%s: Insufficient system entitlement", __func__); | |
916 | } | |
917 | ||
918 | /* Set the remaining accounting variables */ | |
919 | vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size; | |
920 | vio_cmo.excess.free = vio_cmo.excess.size; | |
921 | vio_cmo.min = vio_cmo.reserve.size; | |
922 | vio_cmo.desired = vio_cmo.reserve.size; | |
923 | } | |
924 | ||
925 | /* sysfs device functions and data structures for CMO */ | |
926 | ||
927 | #define viodev_cmo_rd_attr(name) \ | |
928 | static ssize_t viodev_cmo_##name##_show(struct device *dev, \ | |
929 | struct device_attribute *attr, \ | |
930 | char *buf) \ | |
931 | { \ | |
932 | return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name); \ | |
933 | } | |
934 | ||
935 | static ssize_t viodev_cmo_allocs_failed_show(struct device *dev, | |
936 | struct device_attribute *attr, char *buf) | |
937 | { | |
938 | struct vio_dev *viodev = to_vio_dev(dev); | |
939 | return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed)); | |
940 | } | |
941 | ||
942 | static ssize_t viodev_cmo_allocs_failed_reset(struct device *dev, | |
943 | struct device_attribute *attr, const char *buf, size_t count) | |
944 | { | |
945 | struct vio_dev *viodev = to_vio_dev(dev); | |
946 | atomic_set(&viodev->cmo.allocs_failed, 0); | |
947 | return count; | |
948 | } | |
949 | ||
950 | static ssize_t viodev_cmo_desired_set(struct device *dev, | |
951 | struct device_attribute *attr, const char *buf, size_t count) | |
952 | { | |
953 | struct vio_dev *viodev = to_vio_dev(dev); | |
954 | size_t new_desired; | |
955 | int ret; | |
956 | ||
957 | ret = strict_strtoul(buf, 10, &new_desired); | |
958 | if (ret) | |
959 | return ret; | |
960 | ||
961 | vio_cmo_set_dev_desired(viodev, new_desired); | |
962 | return count; | |
963 | } | |
964 | ||
965 | viodev_cmo_rd_attr(desired); | |
966 | viodev_cmo_rd_attr(entitled); | |
967 | viodev_cmo_rd_attr(allocated); | |
968 | ||
969 | static ssize_t name_show(struct device *, struct device_attribute *, char *); | |
970 | static ssize_t devspec_show(struct device *, struct device_attribute *, char *); | |
578b7cd1 BH |
971 | static ssize_t modalias_show(struct device *dev, struct device_attribute *attr, |
972 | char *buf); | |
a90ab95a RJ |
973 | static struct device_attribute vio_cmo_dev_attrs[] = { |
974 | __ATTR_RO(name), | |
975 | __ATTR_RO(devspec), | |
578b7cd1 | 976 | __ATTR_RO(modalias), |
a90ab95a RJ |
977 | __ATTR(cmo_desired, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH, |
978 | viodev_cmo_desired_show, viodev_cmo_desired_set), | |
979 | __ATTR(cmo_entitled, S_IRUGO, viodev_cmo_entitled_show, NULL), | |
980 | __ATTR(cmo_allocated, S_IRUGO, viodev_cmo_allocated_show, NULL), | |
981 | __ATTR(cmo_allocs_failed, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH, | |
982 | viodev_cmo_allocs_failed_show, viodev_cmo_allocs_failed_reset), | |
983 | __ATTR_NULL | |
984 | }; | |
985 | ||
986 | /* sysfs bus functions and data structures for CMO */ | |
987 | ||
988 | #define viobus_cmo_rd_attr(name) \ | |
989 | static ssize_t \ | |
990 | viobus_cmo_##name##_show(struct bus_type *bt, char *buf) \ | |
991 | { \ | |
992 | return sprintf(buf, "%lu\n", vio_cmo.name); \ | |
993 | } | |
994 | ||
995 | #define viobus_cmo_pool_rd_attr(name, var) \ | |
996 | static ssize_t \ | |
997 | viobus_cmo_##name##_pool_show_##var(struct bus_type *bt, char *buf) \ | |
998 | { \ | |
999 | return sprintf(buf, "%lu\n", vio_cmo.name.var); \ | |
1000 | } | |
1001 | ||
1002 | static ssize_t viobus_cmo_high_reset(struct bus_type *bt, const char *buf, | |
1003 | size_t count) | |
1004 | { | |
1005 | unsigned long flags; | |
1006 | ||
1007 | spin_lock_irqsave(&vio_cmo.lock, flags); | |
1008 | vio_cmo.high = vio_cmo.curr; | |
1009 | spin_unlock_irqrestore(&vio_cmo.lock, flags); | |
1010 | ||
1011 | return count; | |
1012 | } | |
1013 | ||
1014 | viobus_cmo_rd_attr(entitled); | |
1015 | viobus_cmo_pool_rd_attr(reserve, size); | |
1016 | viobus_cmo_pool_rd_attr(excess, size); | |
1017 | viobus_cmo_pool_rd_attr(excess, free); | |
1018 | viobus_cmo_rd_attr(spare); | |
1019 | viobus_cmo_rd_attr(min); | |
1020 | viobus_cmo_rd_attr(desired); | |
1021 | viobus_cmo_rd_attr(curr); | |
1022 | viobus_cmo_rd_attr(high); | |
1023 | ||
1024 | static struct bus_attribute vio_cmo_bus_attrs[] = { | |
1025 | __ATTR(cmo_entitled, S_IRUGO, viobus_cmo_entitled_show, NULL), | |
1026 | __ATTR(cmo_reserve_size, S_IRUGO, viobus_cmo_reserve_pool_show_size, NULL), | |
1027 | __ATTR(cmo_excess_size, S_IRUGO, viobus_cmo_excess_pool_show_size, NULL), | |
1028 | __ATTR(cmo_excess_free, S_IRUGO, viobus_cmo_excess_pool_show_free, NULL), | |
1029 | __ATTR(cmo_spare, S_IRUGO, viobus_cmo_spare_show, NULL), | |
1030 | __ATTR(cmo_min, S_IRUGO, viobus_cmo_min_show, NULL), | |
1031 | __ATTR(cmo_desired, S_IRUGO, viobus_cmo_desired_show, NULL), | |
1032 | __ATTR(cmo_curr, S_IRUGO, viobus_cmo_curr_show, NULL), | |
1033 | __ATTR(cmo_high, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH, | |
1034 | viobus_cmo_high_show, viobus_cmo_high_reset), | |
1035 | __ATTR_NULL | |
1036 | }; | |
1037 | ||
1038 | static void vio_cmo_sysfs_init(void) | |
1039 | { | |
1040 | vio_bus_type.dev_attrs = vio_cmo_dev_attrs; | |
1041 | vio_bus_type.bus_attrs = vio_cmo_bus_attrs; | |
1042 | } | |
1043 | #else /* CONFIG_PPC_SMLPAR */ | |
1044 | /* Dummy functions for iSeries platform */ | |
1045 | int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; } | |
1046 | void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {} | |
1047 | static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; } | |
1048 | static void vio_cmo_bus_remove(struct vio_dev *viodev) {} | |
1049 | static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {} | |
b9fa49a9 NL |
1050 | static void vio_cmo_bus_init(void) {} |
1051 | static void vio_cmo_sysfs_init(void) { } | |
a90ab95a RJ |
1052 | #endif /* CONFIG_PPC_SMLPAR */ |
1053 | EXPORT_SYMBOL(vio_cmo_entitlement_update); | |
1054 | EXPORT_SYMBOL(vio_cmo_set_dev_desired); | |
1055 | ||
c7f0e8cb SR |
1056 | static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev) |
1057 | { | |
dd9b67ab SR |
1058 | const unsigned char *dma_window; |
1059 | struct iommu_table *tbl; | |
1060 | unsigned long offset, size; | |
1061 | ||
1062 | if (firmware_has_feature(FW_FEATURE_ISERIES)) | |
1063 | return vio_build_iommu_table_iseries(dev); | |
c7f0e8cb | 1064 | |
58f9b0b0 | 1065 | dma_window = of_get_property(dev->dev.of_node, |
dd9b67ab SR |
1066 | "ibm,my-dma-window", NULL); |
1067 | if (!dma_window) | |
1068 | return NULL; | |
c7f0e8cb | 1069 | |
7aa241fd | 1070 | tbl = kzalloc(sizeof(*tbl), GFP_KERNEL); |
0f337274 | 1071 | if (tbl == NULL) |
1072 | return NULL; | |
c7f0e8cb | 1073 | |
58f9b0b0 | 1074 | of_parse_dma_window(dev->dev.of_node, dma_window, |
dd9b67ab | 1075 | &tbl->it_index, &offset, &size); |
c7f0e8cb | 1076 | |
dd9b67ab SR |
1077 | /* TCE table size - measured in tce entries */ |
1078 | tbl->it_size = size >> IOMMU_PAGE_SHIFT; | |
1079 | /* offset for VIO should always be 0 */ | |
1080 | tbl->it_offset = offset >> IOMMU_PAGE_SHIFT; | |
1081 | tbl->it_busno = 0; | |
1082 | tbl->it_type = TCE_VB; | |
7aa241fd | 1083 | tbl->it_blocksize = 16; |
c7f0e8cb | 1084 | |
dd9b67ab | 1085 | return iommu_init_table(tbl, -1); |
c7f0e8cb | 1086 | } |
1da177e4 | 1087 | |
e10fa773 SR |
1088 | /** |
1089 | * vio_match_device: - Tell if a VIO device has a matching | |
1090 | * VIO device id structure. | |
1091 | * @ids: array of VIO device id structures to search in | |
1092 | * @dev: the VIO device structure to match against | |
1093 | * | |
1094 | * Used by a driver to check whether a VIO device present in the | |
1095 | * system is in its list of supported devices. Returns the matching | |
1096 | * vio_device_id structure or NULL if there is no match. | |
1097 | */ | |
1098 | static const struct vio_device_id *vio_match_device( | |
1099 | const struct vio_device_id *ids, const struct vio_dev *dev) | |
1100 | { | |
1101 | while (ids->type[0] != '\0') { | |
dd721ffd | 1102 | if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) && |
58f9b0b0 | 1103 | of_device_is_compatible(dev->dev.of_node, |
12d04eef | 1104 | ids->compat)) |
e10fa773 SR |
1105 | return ids; |
1106 | ids++; | |
1107 | } | |
1108 | return NULL; | |
1109 | } | |
1110 | ||
5c0b4b87 SR |
1111 | /* |
1112 | * Convert from struct device to struct vio_dev and pass to driver. | |
1da177e4 | 1113 | * dev->driver has already been set by generic code because vio_bus_match |
5c0b4b87 SR |
1114 | * succeeded. |
1115 | */ | |
1da177e4 LT |
1116 | static int vio_bus_probe(struct device *dev) |
1117 | { | |
1118 | struct vio_dev *viodev = to_vio_dev(dev); | |
1119 | struct vio_driver *viodrv = to_vio_driver(dev->driver); | |
1120 | const struct vio_device_id *id; | |
1121 | int error = -ENODEV; | |
1122 | ||
1da177e4 LT |
1123 | if (!viodrv->probe) |
1124 | return error; | |
1125 | ||
1126 | id = vio_match_device(viodrv->id_table, viodev); | |
a90ab95a RJ |
1127 | if (id) { |
1128 | memset(&viodev->cmo, 0, sizeof(viodev->cmo)); | |
1129 | if (firmware_has_feature(FW_FEATURE_CMO)) { | |
1130 | error = vio_cmo_bus_probe(viodev); | |
1131 | if (error) | |
1132 | return error; | |
1133 | } | |
1da177e4 | 1134 | error = viodrv->probe(viodev, id); |
cd5aeb9f | 1135 | if (error && firmware_has_feature(FW_FEATURE_CMO)) |
a90ab95a RJ |
1136 | vio_cmo_bus_remove(viodev); |
1137 | } | |
1da177e4 LT |
1138 | |
1139 | return error; | |
1140 | } | |
1141 | ||
1142 | /* convert from struct device to struct vio_dev and pass to driver. */ | |
1143 | static int vio_bus_remove(struct device *dev) | |
1144 | { | |
1145 | struct vio_dev *viodev = to_vio_dev(dev); | |
1146 | struct vio_driver *viodrv = to_vio_driver(dev->driver); | |
a90ab95a RJ |
1147 | struct device *devptr; |
1148 | int ret = 1; | |
1149 | ||
1150 | /* | |
1151 | * Hold a reference to the device after the remove function is called | |
1152 | * to allow for CMO accounting cleanup for the device. | |
1153 | */ | |
1154 | devptr = get_device(dev); | |
1da177e4 | 1155 | |
5c0b4b87 | 1156 | if (viodrv->remove) |
a90ab95a | 1157 | ret = viodrv->remove(viodev); |
1da177e4 | 1158 | |
a90ab95a RJ |
1159 | if (!ret && firmware_has_feature(FW_FEATURE_CMO)) |
1160 | vio_cmo_bus_remove(viodev); | |
1161 | ||
1162 | put_device(devptr); | |
1163 | return ret; | |
1da177e4 LT |
1164 | } |
1165 | ||
1166 | /** | |
1167 | * vio_register_driver: - Register a new vio driver | |
1168 | * @drv: The vio_driver structure to be registered. | |
1169 | */ | |
1170 | int vio_register_driver(struct vio_driver *viodrv) | |
1171 | { | |
e48b1b45 | 1172 | printk(KERN_DEBUG "%s: driver %s registering\n", __func__, |
6fdf5392 | 1173 | viodrv->driver.name); |
1da177e4 LT |
1174 | |
1175 | /* fill in 'struct driver' fields */ | |
1da177e4 | 1176 | viodrv->driver.bus = &vio_bus_type; |
1da177e4 LT |
1177 | |
1178 | return driver_register(&viodrv->driver); | |
1179 | } | |
1180 | EXPORT_SYMBOL(vio_register_driver); | |
1181 | ||
1182 | /** | |
1183 | * vio_unregister_driver - Remove registration of vio driver. | |
1184 | * @driver: The vio_driver struct to be removed form registration | |
1185 | */ | |
1186 | void vio_unregister_driver(struct vio_driver *viodrv) | |
1187 | { | |
1188 | driver_unregister(&viodrv->driver); | |
1189 | } | |
1190 | EXPORT_SYMBOL(vio_unregister_driver); | |
1191 | ||
c7f0e8cb SR |
1192 | /* vio_dev refcount hit 0 */ |
1193 | static void __devinit vio_dev_release(struct device *dev) | |
1194 | { | |
45848e0f NA |
1195 | struct iommu_table *tbl = get_iommu_table_base(dev); |
1196 | ||
1197 | /* iSeries uses a common table for all vio devices */ | |
1198 | if (!firmware_has_feature(FW_FEATURE_ISERIES) && tbl) | |
1199 | iommu_free_table(tbl, dev->of_node ? | |
1200 | dev->of_node->full_name : dev_name(dev)); | |
58f9b0b0 | 1201 | of_node_put(dev->of_node); |
c7f0e8cb SR |
1202 | kfree(to_vio_dev(dev)); |
1203 | } | |
1204 | ||
1da177e4 | 1205 | /** |
e10fa773 SR |
1206 | * vio_register_device_node: - Register a new vio device. |
1207 | * @of_node: The OF node for this device. | |
1da177e4 | 1208 | * |
e10fa773 | 1209 | * Creates and initializes a vio_dev structure from the data in |
12d04eef | 1210 | * of_node and adds it to the list of virtual devices. |
e10fa773 SR |
1211 | * Returns a pointer to the created vio_dev or NULL if node has |
1212 | * NULL device_type or compatible fields. | |
1da177e4 | 1213 | */ |
de7d812d | 1214 | struct vio_dev *vio_register_device_node(struct device_node *of_node) |
1da177e4 | 1215 | { |
e10fa773 | 1216 | struct vio_dev *viodev; |
a7f67bdf | 1217 | const unsigned int *unit_address; |
e10fa773 SR |
1218 | |
1219 | /* we need the 'device_type' property, in order to match with drivers */ | |
1220 | if (of_node->type == NULL) { | |
1221 | printk(KERN_WARNING "%s: node %s missing 'device_type'\n", | |
e48b1b45 | 1222 | __func__, |
e10fa773 SR |
1223 | of_node->name ? of_node->name : "<unknown>"); |
1224 | return NULL; | |
1da177e4 | 1225 | } |
e10fa773 | 1226 | |
e2eb6392 | 1227 | unit_address = of_get_property(of_node, "reg", NULL); |
e10fa773 SR |
1228 | if (unit_address == NULL) { |
1229 | printk(KERN_WARNING "%s: node %s missing 'reg'\n", | |
e48b1b45 | 1230 | __func__, |
e10fa773 SR |
1231 | of_node->name ? of_node->name : "<unknown>"); |
1232 | return NULL; | |
1233 | } | |
1234 | ||
1235 | /* allocate a vio_dev for this node */ | |
1236 | viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL); | |
1237 | if (viodev == NULL) | |
1238 | return NULL; | |
1239 | ||
0ebfff14 | 1240 | viodev->irq = irq_of_parse_and_map(of_node, 0); |
e10fa773 | 1241 | |
aab0d375 | 1242 | dev_set_name(&viodev->dev, "%x", *unit_address); |
e10fa773 SR |
1243 | viodev->name = of_node->name; |
1244 | viodev->type = of_node->type; | |
1245 | viodev->unit_address = *unit_address; | |
1246 | if (firmware_has_feature(FW_FEATURE_ISERIES)) { | |
e2eb6392 | 1247 | unit_address = of_get_property(of_node, |
e10fa773 SR |
1248 | "linux,unit_address", NULL); |
1249 | if (unit_address != NULL) | |
1250 | viodev->unit_address = *unit_address; | |
1251 | } | |
d706c1b0 | 1252 | viodev->dev.of_node = of_node_get(of_node); |
a90ab95a RJ |
1253 | |
1254 | if (firmware_has_feature(FW_FEATURE_CMO)) | |
1255 | vio_cmo_set_dma_ops(viodev); | |
1256 | else | |
6d283d78 | 1257 | set_dma_ops(&viodev->dev, &dma_iommu_ops); |
738ef42e | 1258 | set_iommu_table_base(&viodev->dev, vio_build_iommu_table(viodev)); |
8fae0353 | 1259 | set_dev_node(&viodev->dev, of_node_to_nid(of_node)); |
c7f0e8cb SR |
1260 | |
1261 | /* init generic 'struct device' fields: */ | |
1262 | viodev->dev.parent = &vio_bus_device.dev; | |
1263 | viodev->dev.bus = &vio_bus_type; | |
1264 | viodev->dev.release = vio_dev_release; | |
b3c73856 NA |
1265 | /* needed to ensure proper operation of coherent allocations |
1266 | * later, in case driver doesn't set it explicitly */ | |
1267 | dma_set_mask(&viodev->dev, DMA_BIT_MASK(64)); | |
1268 | dma_set_coherent_mask(&viodev->dev, DMA_BIT_MASK(64)); | |
e10fa773 SR |
1269 | |
1270 | /* register with generic device framework */ | |
c7f0e8cb SR |
1271 | if (device_register(&viodev->dev)) { |
1272 | printk(KERN_ERR "%s: failed to register device %s\n", | |
aab0d375 | 1273 | __func__, dev_name(&viodev->dev)); |
edea8f6f | 1274 | put_device(&viodev->dev); |
e10fa773 SR |
1275 | return NULL; |
1276 | } | |
1277 | ||
1278 | return viodev; | |
1da177e4 | 1279 | } |
e10fa773 | 1280 | EXPORT_SYMBOL(vio_register_device_node); |
1da177e4 | 1281 | |
1da177e4 LT |
1282 | /** |
1283 | * vio_bus_init: - Initialize the virtual IO bus | |
1284 | */ | |
c7f0e8cb | 1285 | static int __init vio_bus_init(void) |
1da177e4 LT |
1286 | { |
1287 | int err; | |
e10fa773 | 1288 | struct device_node *node_vroot; |
1da177e4 | 1289 | |
a90ab95a RJ |
1290 | if (firmware_has_feature(FW_FEATURE_CMO)) |
1291 | vio_cmo_sysfs_init(); | |
1292 | ||
1da177e4 LT |
1293 | err = bus_register(&vio_bus_type); |
1294 | if (err) { | |
1295 | printk(KERN_ERR "failed to register VIO bus\n"); | |
1296 | return err; | |
1297 | } | |
1298 | ||
5c0b4b87 SR |
1299 | /* |
1300 | * The fake parent of all vio devices, just to give us | |
3e494c80 SR |
1301 | * a nice directory |
1302 | */ | |
ac5b33c9 | 1303 | err = device_register(&vio_bus_device.dev); |
1da177e4 | 1304 | if (err) { |
3e494c80 | 1305 | printk(KERN_WARNING "%s: device_register returned %i\n", |
e48b1b45 | 1306 | __func__, err); |
1da177e4 LT |
1307 | return err; |
1308 | } | |
1309 | ||
a90ab95a RJ |
1310 | if (firmware_has_feature(FW_FEATURE_CMO)) |
1311 | vio_cmo_bus_init(); | |
1312 | ||
30686ba6 | 1313 | node_vroot = of_find_node_by_name(NULL, "vdevice"); |
e10fa773 SR |
1314 | if (node_vroot) { |
1315 | struct device_node *of_node; | |
1316 | ||
1317 | /* | |
1318 | * Create struct vio_devices for each virtual device in | |
1319 | * the device tree. Drivers will associate with them later. | |
1320 | */ | |
1321 | for (of_node = node_vroot->child; of_node != NULL; | |
c5467262 | 1322 | of_node = of_node->sibling) |
e10fa773 | 1323 | vio_register_device_node(of_node); |
30686ba6 | 1324 | of_node_put(node_vroot); |
e10fa773 SR |
1325 | } |
1326 | ||
3e494c80 SR |
1327 | return 0; |
1328 | } | |
c7f0e8cb | 1329 | __initcall(vio_bus_init); |
1da177e4 | 1330 | |
e10fa773 | 1331 | static ssize_t name_show(struct device *dev, |
5c0b4b87 | 1332 | struct device_attribute *attr, char *buf) |
1da177e4 LT |
1333 | { |
1334 | return sprintf(buf, "%s\n", to_vio_dev(dev)->name); | |
1335 | } | |
e10fa773 SR |
1336 | |
1337 | static ssize_t devspec_show(struct device *dev, | |
1338 | struct device_attribute *attr, char *buf) | |
1339 | { | |
58f9b0b0 | 1340 | struct device_node *of_node = dev->of_node; |
e10fa773 SR |
1341 | |
1342 | return sprintf(buf, "%s\n", of_node ? of_node->full_name : "none"); | |
1343 | } | |
1344 | ||
578b7cd1 BH |
1345 | static ssize_t modalias_show(struct device *dev, struct device_attribute *attr, |
1346 | char *buf) | |
1347 | { | |
1348 | const struct vio_dev *vio_dev = to_vio_dev(dev); | |
1349 | struct device_node *dn; | |
1350 | const char *cp; | |
1351 | ||
cf9b59e9 | 1352 | dn = dev->of_node; |
578b7cd1 BH |
1353 | if (!dn) |
1354 | return -ENODEV; | |
1355 | cp = of_get_property(dn, "compatible", NULL); | |
1356 | if (!cp) | |
1357 | return -ENODEV; | |
1358 | ||
1359 | return sprintf(buf, "vio:T%sS%s\n", vio_dev->type, cp); | |
1360 | } | |
1361 | ||
e10fa773 SR |
1362 | static struct device_attribute vio_dev_attrs[] = { |
1363 | __ATTR_RO(name), | |
1364 | __ATTR_RO(devspec), | |
578b7cd1 | 1365 | __ATTR_RO(modalias), |
e10fa773 SR |
1366 | __ATTR_NULL |
1367 | }; | |
1da177e4 | 1368 | |
1da177e4 LT |
1369 | void __devinit vio_unregister_device(struct vio_dev *viodev) |
1370 | { | |
1da177e4 LT |
1371 | device_unregister(&viodev->dev); |
1372 | } | |
1373 | EXPORT_SYMBOL(vio_unregister_device); | |
1374 | ||
1da177e4 LT |
1375 | static int vio_bus_match(struct device *dev, struct device_driver *drv) |
1376 | { | |
1377 | const struct vio_dev *vio_dev = to_vio_dev(dev); | |
1378 | struct vio_driver *vio_drv = to_vio_driver(drv); | |
1379 | const struct vio_device_id *ids = vio_drv->id_table; | |
1da177e4 | 1380 | |
5c0b4b87 | 1381 | return (ids != NULL) && (vio_match_device(ids, vio_dev) != NULL); |
1da177e4 LT |
1382 | } |
1383 | ||
7eff2e7a | 1384 | static int vio_hotplug(struct device *dev, struct kobj_uevent_env *env) |
143dcec2 OH |
1385 | { |
1386 | const struct vio_dev *vio_dev = to_vio_dev(dev); | |
12d04eef | 1387 | struct device_node *dn; |
a7f67bdf | 1388 | const char *cp; |
143dcec2 | 1389 | |
58f9b0b0 | 1390 | dn = dev->of_node; |
e10fa773 | 1391 | if (!dn) |
143dcec2 | 1392 | return -ENODEV; |
7eff2e7a | 1393 | cp = of_get_property(dn, "compatible", NULL); |
143dcec2 OH |
1394 | if (!cp) |
1395 | return -ENODEV; | |
1396 | ||
7eff2e7a | 1397 | add_uevent_var(env, "MODALIAS=vio:T%sS%s", vio_dev->type, cp); |
143dcec2 OH |
1398 | return 0; |
1399 | } | |
1400 | ||
6fccab26 | 1401 | static struct bus_type vio_bus_type = { |
1da177e4 | 1402 | .name = "vio", |
e10fa773 | 1403 | .dev_attrs = vio_dev_attrs, |
312c004d | 1404 | .uevent = vio_hotplug, |
1da177e4 | 1405 | .match = vio_bus_match, |
2f53a80f RK |
1406 | .probe = vio_bus_probe, |
1407 | .remove = vio_bus_remove, | |
a1263c71 | 1408 | .pm = GENERIC_SUBSYS_PM_OPS, |
1da177e4 | 1409 | }; |
e10fa773 SR |
1410 | |
1411 | /** | |
1412 | * vio_get_attribute: - get attribute for virtual device | |
1413 | * @vdev: The vio device to get property. | |
1414 | * @which: The property/attribute to be extracted. | |
1415 | * @length: Pointer to length of returned data size (unused if NULL). | |
1416 | * | |
e2eb6392 | 1417 | * Calls prom.c's of_get_property() to return the value of the |
e10fa773 SR |
1418 | * attribute specified by @which |
1419 | */ | |
1420 | const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length) | |
1421 | { | |
58f9b0b0 | 1422 | return of_get_property(vdev->dev.of_node, which, length); |
e10fa773 SR |
1423 | } |
1424 | EXPORT_SYMBOL(vio_get_attribute); | |
c7f0e8cb SR |
1425 | |
1426 | #ifdef CONFIG_PPC_PSERIES | |
1427 | /* vio_find_name() - internal because only vio.c knows how we formatted the | |
1428 | * kobject name | |
c7f0e8cb | 1429 | */ |
c847c853 | 1430 | static struct vio_dev *vio_find_name(const char *name) |
c7f0e8cb | 1431 | { |
c847c853 | 1432 | struct device *found; |
c7f0e8cb | 1433 | |
c847c853 | 1434 | found = bus_find_device_by_name(&vio_bus_type, NULL, name); |
c7f0e8cb SR |
1435 | if (!found) |
1436 | return NULL; | |
1437 | ||
c847c853 | 1438 | return to_vio_dev(found); |
c7f0e8cb SR |
1439 | } |
1440 | ||
1441 | /** | |
1442 | * vio_find_node - find an already-registered vio_dev | |
1443 | * @vnode: device_node of the virtual device we're looking for | |
1444 | */ | |
1445 | struct vio_dev *vio_find_node(struct device_node *vnode) | |
1446 | { | |
a7f67bdf | 1447 | const uint32_t *unit_address; |
aab0d375 | 1448 | char kobj_name[20]; |
c7f0e8cb SR |
1449 | |
1450 | /* construct the kobject name from the device node */ | |
e2eb6392 | 1451 | unit_address = of_get_property(vnode, "reg", NULL); |
c7f0e8cb SR |
1452 | if (!unit_address) |
1453 | return NULL; | |
aab0d375 | 1454 | snprintf(kobj_name, sizeof(kobj_name), "%x", *unit_address); |
c7f0e8cb SR |
1455 | |
1456 | return vio_find_name(kobj_name); | |
1457 | } | |
1458 | EXPORT_SYMBOL(vio_find_node); | |
1459 | ||
1460 | int vio_enable_interrupts(struct vio_dev *dev) | |
1461 | { | |
1462 | int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE); | |
1463 | if (rc != H_SUCCESS) | |
1464 | printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc); | |
1465 | return rc; | |
1466 | } | |
1467 | EXPORT_SYMBOL(vio_enable_interrupts); | |
1468 | ||
1469 | int vio_disable_interrupts(struct vio_dev *dev) | |
1470 | { | |
1471 | int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE); | |
1472 | if (rc != H_SUCCESS) | |
1473 | printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc); | |
1474 | return rc; | |
1475 | } | |
1476 | EXPORT_SYMBOL(vio_disable_interrupts); | |
1477 | #endif /* CONFIG_PPC_PSERIES */ |