Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 2001 Mike Corrigan & Dave Engebretsen, IBM Corporation |
3 | * | |
4 | * Rewrite, cleanup, new allocation schemes, virtual merging: | |
5 | * Copyright (C) 2004 Olof Johansson, IBM Corporation | |
6 | * and Ben. Herrenschmidt, IBM Corporation | |
7 | * | |
8 | * Dynamic DMA mapping support, bus-independent parts. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or modify | |
11 | * it under the terms of the GNU General Public License as published by | |
12 | * the Free Software Foundation; either version 2 of the License, or | |
13 | * (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | * GNU General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public License | |
21 | * along with this program; if not, write to the Free Software | |
22 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
23 | */ | |
24 | ||
25 | ||
1da177e4 LT |
26 | #include <linux/init.h> |
27 | #include <linux/types.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/mm.h> | |
30 | #include <linux/spinlock.h> | |
31 | #include <linux/string.h> | |
32 | #include <linux/dma-mapping.h> | |
a66022c4 | 33 | #include <linux/bitmap.h> |
fb3475e9 | 34 | #include <linux/iommu-helper.h> |
62a8bd6c | 35 | #include <linux/crash_dump.h> |
b4c3a872 | 36 | #include <linux/hash.h> |
d6b9a81b AB |
37 | #include <linux/fault-inject.h> |
38 | #include <linux/pci.h> | |
4e13c1ac AK |
39 | #include <linux/iommu.h> |
40 | #include <linux/sched.h> | |
1da177e4 LT |
41 | #include <asm/io.h> |
42 | #include <asm/prom.h> | |
43 | #include <asm/iommu.h> | |
44 | #include <asm/pci-bridge.h> | |
45 | #include <asm/machdep.h> | |
5f50867b | 46 | #include <asm/kdump.h> |
3ccc00a7 | 47 | #include <asm/fadump.h> |
d6b9a81b | 48 | #include <asm/vio.h> |
4e13c1ac | 49 | #include <asm/tce.h> |
1da177e4 LT |
50 | |
51 | #define DBG(...) | |
52 | ||
191aee58 | 53 | static int novmerge; |
56997559 | 54 | |
6490c490 RJ |
55 | static void __iommu_free(struct iommu_table *, dma_addr_t, unsigned int); |
56 | ||
1da177e4 LT |
57 | static int __init setup_iommu(char *str) |
58 | { | |
59 | if (!strcmp(str, "novmerge")) | |
60 | novmerge = 1; | |
61 | else if (!strcmp(str, "vmerge")) | |
62 | novmerge = 0; | |
63 | return 1; | |
64 | } | |
65 | ||
66 | __setup("iommu=", setup_iommu); | |
67 | ||
b4c3a872 AB |
68 | static DEFINE_PER_CPU(unsigned int, iommu_pool_hash); |
69 | ||
70 | /* | |
71 | * We precalculate the hash to avoid doing it on every allocation. | |
72 | * | |
73 | * The hash is important to spread CPUs across all the pools. For example, | |
74 | * on a POWER7 with 4 way SMT we want interrupts on the primary threads and | |
75 | * with 4 pools all primary threads would map to the same pool. | |
76 | */ | |
77 | static int __init setup_iommu_pool_hash(void) | |
78 | { | |
79 | unsigned int i; | |
80 | ||
81 | for_each_possible_cpu(i) | |
82 | per_cpu(iommu_pool_hash, i) = hash_32(i, IOMMU_POOL_HASHBITS); | |
83 | ||
84 | return 0; | |
85 | } | |
86 | subsys_initcall(setup_iommu_pool_hash); | |
87 | ||
d6b9a81b AB |
88 | #ifdef CONFIG_FAIL_IOMMU |
89 | ||
90 | static DECLARE_FAULT_ATTR(fail_iommu); | |
91 | ||
92 | static int __init setup_fail_iommu(char *str) | |
93 | { | |
94 | return setup_fault_attr(&fail_iommu, str); | |
95 | } | |
96 | __setup("fail_iommu=", setup_fail_iommu); | |
97 | ||
98 | static bool should_fail_iommu(struct device *dev) | |
99 | { | |
100 | return dev->archdata.fail_iommu && should_fail(&fail_iommu, 1); | |
101 | } | |
102 | ||
103 | static int __init fail_iommu_debugfs(void) | |
104 | { | |
105 | struct dentry *dir = fault_create_debugfs_attr("fail_iommu", | |
106 | NULL, &fail_iommu); | |
107 | ||
8c6ffba0 | 108 | return PTR_ERR_OR_ZERO(dir); |
d6b9a81b AB |
109 | } |
110 | late_initcall(fail_iommu_debugfs); | |
111 | ||
112 | static ssize_t fail_iommu_show(struct device *dev, | |
113 | struct device_attribute *attr, char *buf) | |
114 | { | |
115 | return sprintf(buf, "%d\n", dev->archdata.fail_iommu); | |
116 | } | |
117 | ||
118 | static ssize_t fail_iommu_store(struct device *dev, | |
119 | struct device_attribute *attr, const char *buf, | |
120 | size_t count) | |
121 | { | |
122 | int i; | |
123 | ||
124 | if (count > 0 && sscanf(buf, "%d", &i) > 0) | |
125 | dev->archdata.fail_iommu = (i == 0) ? 0 : 1; | |
126 | ||
127 | return count; | |
128 | } | |
129 | ||
130 | static DEVICE_ATTR(fail_iommu, S_IRUGO|S_IWUSR, fail_iommu_show, | |
131 | fail_iommu_store); | |
132 | ||
133 | static int fail_iommu_bus_notify(struct notifier_block *nb, | |
134 | unsigned long action, void *data) | |
135 | { | |
136 | struct device *dev = data; | |
137 | ||
138 | if (action == BUS_NOTIFY_ADD_DEVICE) { | |
139 | if (device_create_file(dev, &dev_attr_fail_iommu)) | |
140 | pr_warn("Unable to create IOMMU fault injection sysfs " | |
141 | "entries\n"); | |
142 | } else if (action == BUS_NOTIFY_DEL_DEVICE) { | |
143 | device_remove_file(dev, &dev_attr_fail_iommu); | |
144 | } | |
145 | ||
146 | return 0; | |
147 | } | |
148 | ||
149 | static struct notifier_block fail_iommu_bus_notifier = { | |
150 | .notifier_call = fail_iommu_bus_notify | |
151 | }; | |
152 | ||
153 | static int __init fail_iommu_setup(void) | |
154 | { | |
155 | #ifdef CONFIG_PCI | |
156 | bus_register_notifier(&pci_bus_type, &fail_iommu_bus_notifier); | |
157 | #endif | |
158 | #ifdef CONFIG_IBMVIO | |
159 | bus_register_notifier(&vio_bus_type, &fail_iommu_bus_notifier); | |
160 | #endif | |
161 | ||
162 | return 0; | |
163 | } | |
164 | /* | |
165 | * Must execute after PCI and VIO subsystem have initialised but before | |
166 | * devices are probed. | |
167 | */ | |
168 | arch_initcall(fail_iommu_setup); | |
169 | #else | |
170 | static inline bool should_fail_iommu(struct device *dev) | |
171 | { | |
172 | return false; | |
173 | } | |
174 | #endif | |
175 | ||
fb3475e9 FT |
176 | static unsigned long iommu_range_alloc(struct device *dev, |
177 | struct iommu_table *tbl, | |
1da177e4 LT |
178 | unsigned long npages, |
179 | unsigned long *handle, | |
7daa411b | 180 | unsigned long mask, |
1da177e4 LT |
181 | unsigned int align_order) |
182 | { | |
fb3475e9 | 183 | unsigned long n, end, start; |
1da177e4 LT |
184 | unsigned long limit; |
185 | int largealloc = npages > 15; | |
186 | int pass = 0; | |
187 | unsigned long align_mask; | |
fb3475e9 | 188 | unsigned long boundary_size; |
d3622137 | 189 | unsigned long flags; |
b4c3a872 AB |
190 | unsigned int pool_nr; |
191 | struct iommu_pool *pool; | |
1da177e4 LT |
192 | |
193 | align_mask = 0xffffffffffffffffl >> (64 - align_order); | |
194 | ||
195 | /* This allocator was derived from x86_64's bit string search */ | |
196 | ||
197 | /* Sanity check */ | |
13a2eea1 | 198 | if (unlikely(npages == 0)) { |
1da177e4 LT |
199 | if (printk_ratelimit()) |
200 | WARN_ON(1); | |
201 | return DMA_ERROR_CODE; | |
202 | } | |
203 | ||
d6b9a81b AB |
204 | if (should_fail_iommu(dev)) |
205 | return DMA_ERROR_CODE; | |
206 | ||
b4c3a872 AB |
207 | /* |
208 | * We don't need to disable preemption here because any CPU can | |
209 | * safely use any IOMMU pool. | |
210 | */ | |
69111bac | 211 | pool_nr = __this_cpu_read(iommu_pool_hash) & (tbl->nr_pools - 1); |
d3622137 | 212 | |
b4c3a872 AB |
213 | if (largealloc) |
214 | pool = &(tbl->large_pool); | |
1da177e4 | 215 | else |
b4c3a872 | 216 | pool = &(tbl->pools[pool_nr]); |
1da177e4 | 217 | |
b4c3a872 AB |
218 | spin_lock_irqsave(&(pool->lock), flags); |
219 | ||
220 | again: | |
d900bd73 AB |
221 | if ((pass == 0) && handle && *handle && |
222 | (*handle >= pool->start) && (*handle < pool->end)) | |
b4c3a872 AB |
223 | start = *handle; |
224 | else | |
225 | start = pool->hint; | |
1da177e4 | 226 | |
b4c3a872 | 227 | limit = pool->end; |
1da177e4 LT |
228 | |
229 | /* The case below can happen if we have a small segment appended | |
230 | * to a large, or when the previous alloc was at the very end of | |
231 | * the available space. If so, go back to the initial start. | |
232 | */ | |
233 | if (start >= limit) | |
b4c3a872 | 234 | start = pool->start; |
1da177e4 | 235 | |
7daa411b OJ |
236 | if (limit + tbl->it_offset > mask) { |
237 | limit = mask - tbl->it_offset + 1; | |
238 | /* If we're constrained on address range, first try | |
239 | * at the masked hint to avoid O(n) search complexity, | |
b4c3a872 | 240 | * but on second pass, start at 0 in pool 0. |
7daa411b | 241 | */ |
b4c3a872 | 242 | if ((start & mask) >= limit || pass > 0) { |
d900bd73 | 243 | spin_unlock(&(pool->lock)); |
b4c3a872 | 244 | pool = &(tbl->pools[0]); |
d900bd73 | 245 | spin_lock(&(pool->lock)); |
b4c3a872 AB |
246 | start = pool->start; |
247 | } else { | |
7daa411b | 248 | start &= mask; |
b4c3a872 | 249 | } |
7daa411b OJ |
250 | } |
251 | ||
fb3475e9 FT |
252 | if (dev) |
253 | boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1, | |
d0847757 | 254 | 1 << tbl->it_page_shift); |
fb3475e9 | 255 | else |
d0847757 | 256 | boundary_size = ALIGN(1UL << 32, 1 << tbl->it_page_shift); |
fb3475e9 | 257 | /* 4GB boundary for iseries_hv_alloc and iseries_hv_map */ |
1da177e4 | 258 | |
d0847757 AP |
259 | n = iommu_area_alloc(tbl->it_map, limit, start, npages, tbl->it_offset, |
260 | boundary_size >> tbl->it_page_shift, align_mask); | |
fb3475e9 | 261 | if (n == -1) { |
b4c3a872 AB |
262 | if (likely(pass == 0)) { |
263 | /* First try the pool from the start */ | |
264 | pool->hint = pool->start; | |
1da177e4 LT |
265 | pass++; |
266 | goto again; | |
b4c3a872 AB |
267 | |
268 | } else if (pass <= tbl->nr_pools) { | |
269 | /* Now try scanning all the other pools */ | |
270 | spin_unlock(&(pool->lock)); | |
271 | pool_nr = (pool_nr + 1) & (tbl->nr_pools - 1); | |
272 | pool = &tbl->pools[pool_nr]; | |
273 | spin_lock(&(pool->lock)); | |
274 | pool->hint = pool->start; | |
275 | pass++; | |
276 | goto again; | |
277 | ||
1da177e4 | 278 | } else { |
b4c3a872 AB |
279 | /* Give up */ |
280 | spin_unlock_irqrestore(&(pool->lock), flags); | |
1da177e4 LT |
281 | return DMA_ERROR_CODE; |
282 | } | |
283 | } | |
284 | ||
fb3475e9 | 285 | end = n + npages; |
1da177e4 LT |
286 | |
287 | /* Bump the hint to a new block for small allocs. */ | |
288 | if (largealloc) { | |
289 | /* Don't bump to new block to avoid fragmentation */ | |
b4c3a872 | 290 | pool->hint = end; |
1da177e4 LT |
291 | } else { |
292 | /* Overflow will be taken care of at the next allocation */ | |
b4c3a872 | 293 | pool->hint = (end + tbl->it_blocksize - 1) & |
1da177e4 LT |
294 | ~(tbl->it_blocksize - 1); |
295 | } | |
296 | ||
297 | /* Update handle for SG allocations */ | |
298 | if (handle) | |
299 | *handle = end; | |
300 | ||
b4c3a872 AB |
301 | spin_unlock_irqrestore(&(pool->lock), flags); |
302 | ||
1da177e4 LT |
303 | return n; |
304 | } | |
305 | ||
fb3475e9 FT |
306 | static dma_addr_t iommu_alloc(struct device *dev, struct iommu_table *tbl, |
307 | void *page, unsigned int npages, | |
308 | enum dma_data_direction direction, | |
4f3dd8a0 | 309 | unsigned long mask, unsigned int align_order, |
00085f1e | 310 | unsigned long attrs) |
1da177e4 | 311 | { |
d3622137 | 312 | unsigned long entry; |
1da177e4 | 313 | dma_addr_t ret = DMA_ERROR_CODE; |
6490c490 | 314 | int build_fail; |
7daa411b | 315 | |
fb3475e9 | 316 | entry = iommu_range_alloc(dev, tbl, npages, NULL, mask, align_order); |
1da177e4 | 317 | |
0e4bc95d | 318 | if (unlikely(entry == DMA_ERROR_CODE)) |
1da177e4 | 319 | return DMA_ERROR_CODE; |
1da177e4 LT |
320 | |
321 | entry += tbl->it_offset; /* Offset into real TCE table */ | |
d0847757 | 322 | ret = entry << tbl->it_page_shift; /* Set the return dma address */ |
1da177e4 LT |
323 | |
324 | /* Put the TCEs in the HW table */ | |
da004c36 | 325 | build_fail = tbl->it_ops->set(tbl, entry, npages, |
d0847757 AP |
326 | (unsigned long)page & |
327 | IOMMU_PAGE_MASK(tbl), direction, attrs); | |
6490c490 | 328 | |
da004c36 | 329 | /* tbl->it_ops->set() only returns non-zero for transient errors. |
6490c490 RJ |
330 | * Clean up the table bitmap in this case and return |
331 | * DMA_ERROR_CODE. For all other errors the functionality is | |
332 | * not altered. | |
333 | */ | |
334 | if (unlikely(build_fail)) { | |
335 | __iommu_free(tbl, ret, npages); | |
6490c490 RJ |
336 | return DMA_ERROR_CODE; |
337 | } | |
1da177e4 LT |
338 | |
339 | /* Flush/invalidate TLB caches if necessary */ | |
da004c36 AK |
340 | if (tbl->it_ops->flush) |
341 | tbl->it_ops->flush(tbl); | |
1da177e4 | 342 | |
1da177e4 LT |
343 | /* Make sure updates are seen by hardware */ |
344 | mb(); | |
345 | ||
346 | return ret; | |
347 | } | |
348 | ||
67ca1415 AB |
349 | static bool iommu_free_check(struct iommu_table *tbl, dma_addr_t dma_addr, |
350 | unsigned int npages) | |
1da177e4 LT |
351 | { |
352 | unsigned long entry, free_entry; | |
1da177e4 | 353 | |
d0847757 | 354 | entry = dma_addr >> tbl->it_page_shift; |
1da177e4 LT |
355 | free_entry = entry - tbl->it_offset; |
356 | ||
357 | if (((free_entry + npages) > tbl->it_size) || | |
358 | (entry < tbl->it_offset)) { | |
359 | if (printk_ratelimit()) { | |
360 | printk(KERN_INFO "iommu_free: invalid entry\n"); | |
361 | printk(KERN_INFO "\tentry = 0x%lx\n", entry); | |
fe333321 IM |
362 | printk(KERN_INFO "\tdma_addr = 0x%llx\n", (u64)dma_addr); |
363 | printk(KERN_INFO "\tTable = 0x%llx\n", (u64)tbl); | |
364 | printk(KERN_INFO "\tbus# = 0x%llx\n", (u64)tbl->it_busno); | |
365 | printk(KERN_INFO "\tsize = 0x%llx\n", (u64)tbl->it_size); | |
366 | printk(KERN_INFO "\tstartOff = 0x%llx\n", (u64)tbl->it_offset); | |
367 | printk(KERN_INFO "\tindex = 0x%llx\n", (u64)tbl->it_index); | |
1da177e4 LT |
368 | WARN_ON(1); |
369 | } | |
67ca1415 AB |
370 | |
371 | return false; | |
1da177e4 LT |
372 | } |
373 | ||
67ca1415 AB |
374 | return true; |
375 | } | |
376 | ||
b4c3a872 AB |
377 | static struct iommu_pool *get_pool(struct iommu_table *tbl, |
378 | unsigned long entry) | |
379 | { | |
380 | struct iommu_pool *p; | |
381 | unsigned long largepool_start = tbl->large_pool.start; | |
382 | ||
383 | /* The large pool is the last pool at the top of the table */ | |
384 | if (entry >= largepool_start) { | |
385 | p = &tbl->large_pool; | |
386 | } else { | |
387 | unsigned int pool_nr = entry / tbl->poolsize; | |
388 | ||
389 | BUG_ON(pool_nr > tbl->nr_pools); | |
390 | p = &tbl->pools[pool_nr]; | |
391 | } | |
392 | ||
393 | return p; | |
394 | } | |
395 | ||
67ca1415 AB |
396 | static void __iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr, |
397 | unsigned int npages) | |
1da177e4 | 398 | { |
67ca1415 | 399 | unsigned long entry, free_entry; |
1da177e4 | 400 | unsigned long flags; |
b4c3a872 | 401 | struct iommu_pool *pool; |
1da177e4 | 402 | |
d0847757 | 403 | entry = dma_addr >> tbl->it_page_shift; |
67ca1415 AB |
404 | free_entry = entry - tbl->it_offset; |
405 | ||
b4c3a872 AB |
406 | pool = get_pool(tbl, free_entry); |
407 | ||
67ca1415 AB |
408 | if (!iommu_free_check(tbl, dma_addr, npages)) |
409 | return; | |
410 | ||
da004c36 | 411 | tbl->it_ops->clear(tbl, entry, npages); |
67ca1415 | 412 | |
b4c3a872 | 413 | spin_lock_irqsave(&(pool->lock), flags); |
67ca1415 | 414 | bitmap_clear(tbl->it_map, free_entry, npages); |
b4c3a872 | 415 | spin_unlock_irqrestore(&(pool->lock), flags); |
67ca1415 AB |
416 | } |
417 | ||
418 | static void iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr, | |
419 | unsigned int npages) | |
420 | { | |
421 | __iommu_free(tbl, dma_addr, npages); | |
1da177e4 LT |
422 | |
423 | /* Make sure TLB cache is flushed if the HW needs it. We do | |
424 | * not do an mb() here on purpose, it is not needed on any of | |
425 | * the current platforms. | |
426 | */ | |
da004c36 AK |
427 | if (tbl->it_ops->flush) |
428 | tbl->it_ops->flush(tbl); | |
1da177e4 LT |
429 | } |
430 | ||
0690cbd2 JR |
431 | int ppc_iommu_map_sg(struct device *dev, struct iommu_table *tbl, |
432 | struct scatterlist *sglist, int nelems, | |
433 | unsigned long mask, enum dma_data_direction direction, | |
00085f1e | 434 | unsigned long attrs) |
1da177e4 LT |
435 | { |
436 | dma_addr_t dma_next = 0, dma_addr; | |
1da177e4 | 437 | struct scatterlist *s, *outs, *segstart; |
6490c490 | 438 | int outcount, incount, i, build_fail = 0; |
d262c32a | 439 | unsigned int align; |
1da177e4 | 440 | unsigned long handle; |
740c3ce6 | 441 | unsigned int max_seg_size; |
1da177e4 LT |
442 | |
443 | BUG_ON(direction == DMA_NONE); | |
444 | ||
445 | if ((nelems == 0) || !tbl) | |
446 | return 0; | |
447 | ||
448 | outs = s = segstart = &sglist[0]; | |
449 | outcount = 1; | |
ac9af7cb | 450 | incount = nelems; |
1da177e4 LT |
451 | handle = 0; |
452 | ||
453 | /* Init first segment length for backout at failure */ | |
454 | outs->dma_length = 0; | |
455 | ||
5d2efba6 | 456 | DBG("sg mapping %d elements:\n", nelems); |
1da177e4 | 457 | |
740c3ce6 | 458 | max_seg_size = dma_get_max_seg_size(dev); |
78bdc310 | 459 | for_each_sg(sglist, s, nelems, i) { |
1da177e4 LT |
460 | unsigned long vaddr, npages, entry, slen; |
461 | ||
462 | slen = s->length; | |
463 | /* Sanity check */ | |
464 | if (slen == 0) { | |
465 | dma_next = 0; | |
466 | continue; | |
467 | } | |
468 | /* Allocate iommu entries for that segment */ | |
58b053e4 | 469 | vaddr = (unsigned long) sg_virt(s); |
d0847757 | 470 | npages = iommu_num_pages(vaddr, slen, IOMMU_PAGE_SIZE(tbl)); |
d262c32a | 471 | align = 0; |
d0847757 | 472 | if (tbl->it_page_shift < PAGE_SHIFT && slen >= PAGE_SIZE && |
d262c32a | 473 | (vaddr & ~PAGE_MASK) == 0) |
d0847757 | 474 | align = PAGE_SHIFT - tbl->it_page_shift; |
fb3475e9 | 475 | entry = iommu_range_alloc(dev, tbl, npages, &handle, |
d0847757 | 476 | mask >> tbl->it_page_shift, align); |
1da177e4 LT |
477 | |
478 | DBG(" - vaddr: %lx, size: %lx\n", vaddr, slen); | |
479 | ||
480 | /* Handle failure */ | |
481 | if (unlikely(entry == DMA_ERROR_CODE)) { | |
482 | if (printk_ratelimit()) | |
4dfa9c47 AB |
483 | dev_info(dev, "iommu_alloc failed, tbl %p " |
484 | "vaddr %lx npages %lu\n", tbl, vaddr, | |
485 | npages); | |
1da177e4 LT |
486 | goto failure; |
487 | } | |
488 | ||
489 | /* Convert entry to a dma_addr_t */ | |
490 | entry += tbl->it_offset; | |
d0847757 AP |
491 | dma_addr = entry << tbl->it_page_shift; |
492 | dma_addr |= (s->offset & ~IOMMU_PAGE_MASK(tbl)); | |
1da177e4 | 493 | |
5d2efba6 | 494 | DBG(" - %lu pages, entry: %lx, dma_addr: %lx\n", |
1da177e4 LT |
495 | npages, entry, dma_addr); |
496 | ||
497 | /* Insert into HW table */ | |
da004c36 | 498 | build_fail = tbl->it_ops->set(tbl, entry, npages, |
d0847757 AP |
499 | vaddr & IOMMU_PAGE_MASK(tbl), |
500 | direction, attrs); | |
6490c490 RJ |
501 | if(unlikely(build_fail)) |
502 | goto failure; | |
1da177e4 LT |
503 | |
504 | /* If we are in an open segment, try merging */ | |
505 | if (segstart != s) { | |
506 | DBG(" - trying merge...\n"); | |
507 | /* We cannot merge if: | |
508 | * - allocated dma_addr isn't contiguous to previous allocation | |
509 | */ | |
740c3ce6 FT |
510 | if (novmerge || (dma_addr != dma_next) || |
511 | (outs->dma_length + s->length > max_seg_size)) { | |
1da177e4 LT |
512 | /* Can't merge: create a new segment */ |
513 | segstart = s; | |
78bdc310 JA |
514 | outcount++; |
515 | outs = sg_next(outs); | |
1da177e4 LT |
516 | DBG(" can't merge, new segment.\n"); |
517 | } else { | |
518 | outs->dma_length += s->length; | |
5d2efba6 | 519 | DBG(" merged, new len: %ux\n", outs->dma_length); |
1da177e4 LT |
520 | } |
521 | } | |
522 | ||
523 | if (segstart == s) { | |
524 | /* This is a new segment, fill entries */ | |
525 | DBG(" - filling new segment.\n"); | |
526 | outs->dma_address = dma_addr; | |
527 | outs->dma_length = slen; | |
528 | } | |
529 | ||
530 | /* Calculate next page pointer for contiguous check */ | |
531 | dma_next = dma_addr + slen; | |
532 | ||
533 | DBG(" - dma next is: %lx\n", dma_next); | |
534 | } | |
535 | ||
536 | /* Flush/invalidate TLB caches if necessary */ | |
da004c36 AK |
537 | if (tbl->it_ops->flush) |
538 | tbl->it_ops->flush(tbl); | |
1da177e4 | 539 | |
1da177e4 LT |
540 | DBG("mapped %d elements:\n", outcount); |
541 | ||
0690cbd2 | 542 | /* For the sake of ppc_iommu_unmap_sg, we clear out the length in the |
1da177e4 LT |
543 | * next entry of the sglist if we didn't fill the list completely |
544 | */ | |
ac9af7cb | 545 | if (outcount < incount) { |
78bdc310 | 546 | outs = sg_next(outs); |
1da177e4 LT |
547 | outs->dma_address = DMA_ERROR_CODE; |
548 | outs->dma_length = 0; | |
549 | } | |
a958a264 JM |
550 | |
551 | /* Make sure updates are seen by hardware */ | |
552 | mb(); | |
553 | ||
1da177e4 LT |
554 | return outcount; |
555 | ||
556 | failure: | |
78bdc310 | 557 | for_each_sg(sglist, s, nelems, i) { |
1da177e4 LT |
558 | if (s->dma_length != 0) { |
559 | unsigned long vaddr, npages; | |
560 | ||
d0847757 | 561 | vaddr = s->dma_address & IOMMU_PAGE_MASK(tbl); |
2994a3b2 | 562 | npages = iommu_num_pages(s->dma_address, s->dma_length, |
d0847757 | 563 | IOMMU_PAGE_SIZE(tbl)); |
d3622137 | 564 | __iommu_free(tbl, vaddr, npages); |
a958a264 JM |
565 | s->dma_address = DMA_ERROR_CODE; |
566 | s->dma_length = 0; | |
1da177e4 | 567 | } |
78bdc310 JA |
568 | if (s == outs) |
569 | break; | |
1da177e4 | 570 | } |
1da177e4 LT |
571 | return 0; |
572 | } | |
573 | ||
574 | ||
0690cbd2 JR |
575 | void ppc_iommu_unmap_sg(struct iommu_table *tbl, struct scatterlist *sglist, |
576 | int nelems, enum dma_data_direction direction, | |
00085f1e | 577 | unsigned long attrs) |
1da177e4 | 578 | { |
78bdc310 | 579 | struct scatterlist *sg; |
1da177e4 LT |
580 | |
581 | BUG_ON(direction == DMA_NONE); | |
582 | ||
583 | if (!tbl) | |
584 | return; | |
585 | ||
78bdc310 | 586 | sg = sglist; |
1da177e4 LT |
587 | while (nelems--) { |
588 | unsigned int npages; | |
78bdc310 | 589 | dma_addr_t dma_handle = sg->dma_address; |
1da177e4 | 590 | |
78bdc310 | 591 | if (sg->dma_length == 0) |
1da177e4 | 592 | break; |
2994a3b2 | 593 | npages = iommu_num_pages(dma_handle, sg->dma_length, |
d0847757 | 594 | IOMMU_PAGE_SIZE(tbl)); |
d3622137 | 595 | __iommu_free(tbl, dma_handle, npages); |
78bdc310 | 596 | sg = sg_next(sg); |
1da177e4 LT |
597 | } |
598 | ||
599 | /* Flush/invalidate TLBs if necessary. As for iommu_free(), we | |
600 | * do not do an mb() here, the affected platforms do not need it | |
601 | * when freeing. | |
602 | */ | |
da004c36 AK |
603 | if (tbl->it_ops->flush) |
604 | tbl->it_ops->flush(tbl); | |
1da177e4 LT |
605 | } |
606 | ||
54622f10 MK |
607 | static void iommu_table_clear(struct iommu_table *tbl) |
608 | { | |
3ccc00a7 MS |
609 | /* |
610 | * In case of firmware assisted dump system goes through clean | |
611 | * reboot process at the time of system crash. Hence it's safe to | |
612 | * clear the TCE entries if firmware assisted dump is active. | |
613 | */ | |
614 | if (!is_kdump_kernel() || is_fadump_active()) { | |
54622f10 | 615 | /* Clear the table in case firmware left allocations in it */ |
da004c36 | 616 | tbl->it_ops->clear(tbl, tbl->it_offset, tbl->it_size); |
54622f10 MK |
617 | return; |
618 | } | |
619 | ||
620 | #ifdef CONFIG_CRASH_DUMP | |
da004c36 | 621 | if (tbl->it_ops->get) { |
54622f10 MK |
622 | unsigned long index, tceval, tcecount = 0; |
623 | ||
624 | /* Reserve the existing mappings left by the first kernel. */ | |
625 | for (index = 0; index < tbl->it_size; index++) { | |
da004c36 | 626 | tceval = tbl->it_ops->get(tbl, index + tbl->it_offset); |
54622f10 MK |
627 | /* |
628 | * Freed TCE entry contains 0x7fffffffffffffff on JS20 | |
629 | */ | |
630 | if (tceval && (tceval != 0x7fffffffffffffffUL)) { | |
631 | __set_bit(index, tbl->it_map); | |
632 | tcecount++; | |
633 | } | |
634 | } | |
635 | ||
636 | if ((tbl->it_size - tcecount) < KDUMP_MIN_TCE_ENTRIES) { | |
637 | printk(KERN_WARNING "TCE table is full; freeing "); | |
638 | printk(KERN_WARNING "%d entries for the kdump boot\n", | |
639 | KDUMP_MIN_TCE_ENTRIES); | |
640 | for (index = tbl->it_size - KDUMP_MIN_TCE_ENTRIES; | |
641 | index < tbl->it_size; index++) | |
642 | __clear_bit(index, tbl->it_map); | |
643 | } | |
644 | } | |
645 | #endif | |
646 | } | |
647 | ||
1da177e4 LT |
648 | /* |
649 | * Build a iommu_table structure. This contains a bit map which | |
650 | * is used to manage allocation of the tce space. | |
651 | */ | |
ca1588e7 | 652 | struct iommu_table *iommu_init_table(struct iommu_table *tbl, int nid) |
1da177e4 LT |
653 | { |
654 | unsigned long sz; | |
655 | static int welcomed = 0; | |
ca1588e7 | 656 | struct page *page; |
b4c3a872 AB |
657 | unsigned int i; |
658 | struct iommu_pool *p; | |
1da177e4 | 659 | |
da004c36 AK |
660 | BUG_ON(!tbl->it_ops); |
661 | ||
1da177e4 | 662 | /* number of bytes needed for the bitmap */ |
c5a0809a | 663 | sz = BITS_TO_LONGS(tbl->it_size) * sizeof(unsigned long); |
1da177e4 | 664 | |
1cf389df | 665 | page = alloc_pages_node(nid, GFP_KERNEL, get_order(sz)); |
ca1588e7 | 666 | if (!page) |
1da177e4 | 667 | panic("iommu_init_table: Can't allocate %ld bytes\n", sz); |
ca1588e7 | 668 | tbl->it_map = page_address(page); |
1da177e4 LT |
669 | memset(tbl->it_map, 0, sz); |
670 | ||
d12b524f TLSC |
671 | /* |
672 | * Reserve page 0 so it will not be used for any mappings. | |
673 | * This avoids buggy drivers that consider page 0 to be invalid | |
674 | * to crash the machine or even lose data. | |
675 | */ | |
676 | if (tbl->it_offset == 0) | |
677 | set_bit(0, tbl->it_map); | |
678 | ||
b4c3a872 | 679 | /* We only split the IOMMU table if we have 1GB or more of space */ |
d0847757 | 680 | if ((tbl->it_size << tbl->it_page_shift) >= (1UL * 1024 * 1024 * 1024)) |
b4c3a872 AB |
681 | tbl->nr_pools = IOMMU_NR_POOLS; |
682 | else | |
683 | tbl->nr_pools = 1; | |
684 | ||
685 | /* We reserve the top 1/4 of the table for large allocations */ | |
dcd261ba | 686 | tbl->poolsize = (tbl->it_size * 3 / 4) / tbl->nr_pools; |
b4c3a872 | 687 | |
dcd261ba | 688 | for (i = 0; i < tbl->nr_pools; i++) { |
b4c3a872 AB |
689 | p = &tbl->pools[i]; |
690 | spin_lock_init(&(p->lock)); | |
691 | p->start = tbl->poolsize * i; | |
692 | p->hint = p->start; | |
693 | p->end = p->start + tbl->poolsize; | |
694 | } | |
695 | ||
696 | p = &tbl->large_pool; | |
697 | spin_lock_init(&(p->lock)); | |
698 | p->start = tbl->poolsize * i; | |
699 | p->hint = p->start; | |
700 | p->end = tbl->it_size; | |
1da177e4 | 701 | |
54622f10 | 702 | iommu_table_clear(tbl); |
d3588ba9 | 703 | |
1da177e4 LT |
704 | if (!welcomed) { |
705 | printk(KERN_INFO "IOMMU table initialized, virtual merging %s\n", | |
706 | novmerge ? "disabled" : "enabled"); | |
707 | welcomed = 1; | |
708 | } | |
709 | ||
710 | return tbl; | |
711 | } | |
712 | ||
68d315f5 | 713 | void iommu_free_table(struct iommu_table *tbl, const char *node_name) |
1da177e4 | 714 | { |
c5a0809a | 715 | unsigned long bitmap_sz; |
1da177e4 LT |
716 | unsigned int order; |
717 | ||
8aca92d8 AK |
718 | if (!tbl) |
719 | return; | |
720 | ||
721 | if (!tbl->it_map) { | |
722 | kfree(tbl); | |
1da177e4 LT |
723 | return; |
724 | } | |
725 | ||
7f966d39 TLSC |
726 | /* |
727 | * In case we have reserved the first bit, we should not emit | |
728 | * the warning below. | |
729 | */ | |
730 | if (tbl->it_offset == 0) | |
731 | clear_bit(0, tbl->it_map); | |
732 | ||
1da177e4 | 733 | /* verify that table contains no entries */ |
c5a0809a AM |
734 | if (!bitmap_empty(tbl->it_map, tbl->it_size)) |
735 | pr_warn("%s: Unexpected TCEs for %s\n", __func__, node_name); | |
1da177e4 LT |
736 | |
737 | /* calculate bitmap size in bytes */ | |
c5a0809a | 738 | bitmap_sz = BITS_TO_LONGS(tbl->it_size) * sizeof(unsigned long); |
1da177e4 LT |
739 | |
740 | /* free bitmap */ | |
741 | order = get_order(bitmap_sz); | |
742 | free_pages((unsigned long) tbl->it_map, order); | |
743 | ||
744 | /* free table */ | |
745 | kfree(tbl); | |
746 | } | |
747 | ||
748 | /* Creates TCEs for a user provided buffer. The user buffer must be | |
f9226d57 MN |
749 | * contiguous real kernel storage (not vmalloc). The address passed here |
750 | * comprises a page address and offset into that page. The dma_addr_t | |
751 | * returned will point to the same byte within the page as was passed in. | |
1da177e4 | 752 | */ |
f9226d57 MN |
753 | dma_addr_t iommu_map_page(struct device *dev, struct iommu_table *tbl, |
754 | struct page *page, unsigned long offset, size_t size, | |
755 | unsigned long mask, enum dma_data_direction direction, | |
00085f1e | 756 | unsigned long attrs) |
1da177e4 LT |
757 | { |
758 | dma_addr_t dma_handle = DMA_ERROR_CODE; | |
f9226d57 | 759 | void *vaddr; |
1da177e4 | 760 | unsigned long uaddr; |
d262c32a | 761 | unsigned int npages, align; |
1da177e4 LT |
762 | |
763 | BUG_ON(direction == DMA_NONE); | |
764 | ||
f9226d57 | 765 | vaddr = page_address(page) + offset; |
1da177e4 | 766 | uaddr = (unsigned long)vaddr; |
d0847757 | 767 | npages = iommu_num_pages(uaddr, size, IOMMU_PAGE_SIZE(tbl)); |
1da177e4 LT |
768 | |
769 | if (tbl) { | |
d262c32a | 770 | align = 0; |
d0847757 | 771 | if (tbl->it_page_shift < PAGE_SHIFT && size >= PAGE_SIZE && |
d262c32a | 772 | ((unsigned long)vaddr & ~PAGE_MASK) == 0) |
d0847757 | 773 | align = PAGE_SHIFT - tbl->it_page_shift; |
d262c32a | 774 | |
fb3475e9 | 775 | dma_handle = iommu_alloc(dev, tbl, vaddr, npages, direction, |
d0847757 | 776 | mask >> tbl->it_page_shift, align, |
4f3dd8a0 | 777 | attrs); |
1da177e4 LT |
778 | if (dma_handle == DMA_ERROR_CODE) { |
779 | if (printk_ratelimit()) { | |
4dfa9c47 AB |
780 | dev_info(dev, "iommu_alloc failed, tbl %p " |
781 | "vaddr %p npages %d\n", tbl, vaddr, | |
782 | npages); | |
1da177e4 LT |
783 | } |
784 | } else | |
d0847757 | 785 | dma_handle |= (uaddr & ~IOMMU_PAGE_MASK(tbl)); |
1da177e4 LT |
786 | } |
787 | ||
788 | return dma_handle; | |
789 | } | |
790 | ||
f9226d57 MN |
791 | void iommu_unmap_page(struct iommu_table *tbl, dma_addr_t dma_handle, |
792 | size_t size, enum dma_data_direction direction, | |
00085f1e | 793 | unsigned long attrs) |
1da177e4 | 794 | { |
5d2efba6 LV |
795 | unsigned int npages; |
796 | ||
1da177e4 LT |
797 | BUG_ON(direction == DMA_NONE); |
798 | ||
5d2efba6 | 799 | if (tbl) { |
d0847757 AP |
800 | npages = iommu_num_pages(dma_handle, size, |
801 | IOMMU_PAGE_SIZE(tbl)); | |
5d2efba6 LV |
802 | iommu_free(tbl, dma_handle, npages); |
803 | } | |
1da177e4 LT |
804 | } |
805 | ||
806 | /* Allocates a contiguous real buffer and creates mappings over it. | |
807 | * Returns the virtual address of the buffer and sets dma_handle | |
808 | * to the dma address (mapping) of the first page. | |
809 | */ | |
fb3475e9 FT |
810 | void *iommu_alloc_coherent(struct device *dev, struct iommu_table *tbl, |
811 | size_t size, dma_addr_t *dma_handle, | |
812 | unsigned long mask, gfp_t flag, int node) | |
1da177e4 LT |
813 | { |
814 | void *ret = NULL; | |
815 | dma_addr_t mapping; | |
5d2efba6 LV |
816 | unsigned int order; |
817 | unsigned int nio_pages, io_order; | |
8eb6c6e3 | 818 | struct page *page; |
1da177e4 LT |
819 | |
820 | size = PAGE_ALIGN(size); | |
1da177e4 LT |
821 | order = get_order(size); |
822 | ||
823 | /* | |
824 | * Client asked for way too much space. This is checked later | |
825 | * anyway. It is easier to debug here for the drivers than in | |
826 | * the tce tables. | |
827 | */ | |
828 | if (order >= IOMAP_MAX_ORDER) { | |
4dfa9c47 AB |
829 | dev_info(dev, "iommu_alloc_consistent size too large: 0x%lx\n", |
830 | size); | |
1da177e4 LT |
831 | return NULL; |
832 | } | |
833 | ||
834 | if (!tbl) | |
835 | return NULL; | |
836 | ||
837 | /* Alloc enough pages (and possibly more) */ | |
05061354 | 838 | page = alloc_pages_node(node, flag, order); |
8eb6c6e3 | 839 | if (!page) |
1da177e4 | 840 | return NULL; |
8eb6c6e3 | 841 | ret = page_address(page); |
1da177e4 LT |
842 | memset(ret, 0, size); |
843 | ||
844 | /* Set up tces to cover the allocated range */ | |
d0847757 AP |
845 | nio_pages = size >> tbl->it_page_shift; |
846 | io_order = get_iommu_order(size, tbl); | |
fb3475e9 | 847 | mapping = iommu_alloc(dev, tbl, ret, nio_pages, DMA_BIDIRECTIONAL, |
00085f1e | 848 | mask >> tbl->it_page_shift, io_order, 0); |
1da177e4 LT |
849 | if (mapping == DMA_ERROR_CODE) { |
850 | free_pages((unsigned long)ret, order); | |
8eb6c6e3 CH |
851 | return NULL; |
852 | } | |
853 | *dma_handle = mapping; | |
1da177e4 LT |
854 | return ret; |
855 | } | |
856 | ||
857 | void iommu_free_coherent(struct iommu_table *tbl, size_t size, | |
858 | void *vaddr, dma_addr_t dma_handle) | |
859 | { | |
1da177e4 | 860 | if (tbl) { |
5d2efba6 LV |
861 | unsigned int nio_pages; |
862 | ||
863 | size = PAGE_ALIGN(size); | |
d0847757 | 864 | nio_pages = size >> tbl->it_page_shift; |
5d2efba6 | 865 | iommu_free(tbl, dma_handle, nio_pages); |
1da177e4 | 866 | size = PAGE_ALIGN(size); |
1da177e4 LT |
867 | free_pages((unsigned long)vaddr, get_order(size)); |
868 | } | |
869 | } | |
4e13c1ac | 870 | |
10b35b2b AK |
871 | unsigned long iommu_direction_to_tce_perm(enum dma_data_direction dir) |
872 | { | |
873 | switch (dir) { | |
874 | case DMA_BIDIRECTIONAL: | |
875 | return TCE_PCI_READ | TCE_PCI_WRITE; | |
876 | case DMA_FROM_DEVICE: | |
877 | return TCE_PCI_WRITE; | |
878 | case DMA_TO_DEVICE: | |
879 | return TCE_PCI_READ; | |
880 | default: | |
881 | return 0; | |
882 | } | |
883 | } | |
884 | EXPORT_SYMBOL_GPL(iommu_direction_to_tce_perm); | |
885 | ||
4e13c1ac AK |
886 | #ifdef CONFIG_IOMMU_API |
887 | /* | |
888 | * SPAPR TCE API | |
889 | */ | |
890 | static void group_release(void *iommu_data) | |
891 | { | |
b348aa65 AK |
892 | struct iommu_table_group *table_group = iommu_data; |
893 | ||
894 | table_group->group = NULL; | |
4e13c1ac AK |
895 | } |
896 | ||
b348aa65 | 897 | void iommu_register_group(struct iommu_table_group *table_group, |
4e13c1ac AK |
898 | int pci_domain_number, unsigned long pe_num) |
899 | { | |
900 | struct iommu_group *grp; | |
901 | char *name; | |
902 | ||
903 | grp = iommu_group_alloc(); | |
904 | if (IS_ERR(grp)) { | |
905 | pr_warn("powerpc iommu api: cannot create new group, err=%ld\n", | |
906 | PTR_ERR(grp)); | |
907 | return; | |
908 | } | |
b348aa65 AK |
909 | table_group->group = grp; |
910 | iommu_group_set_iommudata(grp, table_group, group_release); | |
4e13c1ac AK |
911 | name = kasprintf(GFP_KERNEL, "domain%d-pe%lx", |
912 | pci_domain_number, pe_num); | |
913 | if (!name) | |
914 | return; | |
915 | iommu_group_set_name(grp, name); | |
916 | kfree(name); | |
917 | } | |
918 | ||
919 | enum dma_data_direction iommu_tce_direction(unsigned long tce) | |
920 | { | |
921 | if ((tce & TCE_PCI_READ) && (tce & TCE_PCI_WRITE)) | |
922 | return DMA_BIDIRECTIONAL; | |
923 | else if (tce & TCE_PCI_READ) | |
924 | return DMA_TO_DEVICE; | |
925 | else if (tce & TCE_PCI_WRITE) | |
926 | return DMA_FROM_DEVICE; | |
927 | else | |
928 | return DMA_NONE; | |
929 | } | |
930 | EXPORT_SYMBOL_GPL(iommu_tce_direction); | |
931 | ||
932 | void iommu_flush_tce(struct iommu_table *tbl) | |
933 | { | |
934 | /* Flush/invalidate TLB caches if necessary */ | |
da004c36 AK |
935 | if (tbl->it_ops->flush) |
936 | tbl->it_ops->flush(tbl); | |
4e13c1ac AK |
937 | |
938 | /* Make sure updates are seen by hardware */ | |
939 | mb(); | |
940 | } | |
941 | EXPORT_SYMBOL_GPL(iommu_flush_tce); | |
942 | ||
943 | int iommu_tce_clear_param_check(struct iommu_table *tbl, | |
944 | unsigned long ioba, unsigned long tce_value, | |
945 | unsigned long npages) | |
946 | { | |
da004c36 | 947 | /* tbl->it_ops->clear() does not support any value but 0 */ |
4e13c1ac AK |
948 | if (tce_value) |
949 | return -EINVAL; | |
950 | ||
d0847757 | 951 | if (ioba & ~IOMMU_PAGE_MASK(tbl)) |
4e13c1ac AK |
952 | return -EINVAL; |
953 | ||
d0847757 | 954 | ioba >>= tbl->it_page_shift; |
4e13c1ac AK |
955 | if (ioba < tbl->it_offset) |
956 | return -EINVAL; | |
957 | ||
958 | if ((ioba + npages) > (tbl->it_offset + tbl->it_size)) | |
959 | return -EINVAL; | |
960 | ||
961 | return 0; | |
962 | } | |
963 | EXPORT_SYMBOL_GPL(iommu_tce_clear_param_check); | |
964 | ||
965 | int iommu_tce_put_param_check(struct iommu_table *tbl, | |
966 | unsigned long ioba, unsigned long tce) | |
967 | { | |
05c6cfb9 | 968 | if (tce & ~IOMMU_PAGE_MASK(tbl)) |
4e13c1ac AK |
969 | return -EINVAL; |
970 | ||
d0847757 | 971 | if (ioba & ~IOMMU_PAGE_MASK(tbl)) |
4e13c1ac AK |
972 | return -EINVAL; |
973 | ||
d0847757 | 974 | ioba >>= tbl->it_page_shift; |
4e13c1ac AK |
975 | if (ioba < tbl->it_offset) |
976 | return -EINVAL; | |
977 | ||
978 | if ((ioba + 1) > (tbl->it_offset + tbl->it_size)) | |
979 | return -EINVAL; | |
980 | ||
981 | return 0; | |
982 | } | |
983 | EXPORT_SYMBOL_GPL(iommu_tce_put_param_check); | |
984 | ||
05c6cfb9 AK |
985 | long iommu_tce_xchg(struct iommu_table *tbl, unsigned long entry, |
986 | unsigned long *hpa, enum dma_data_direction *direction) | |
4e13c1ac | 987 | { |
05c6cfb9 | 988 | long ret; |
4e13c1ac | 989 | |
05c6cfb9 | 990 | ret = tbl->it_ops->exchange(tbl, entry, hpa, direction); |
4e13c1ac | 991 | |
05c6cfb9 AK |
992 | if (!ret && ((*direction == DMA_FROM_DEVICE) || |
993 | (*direction == DMA_BIDIRECTIONAL))) | |
994 | SetPageDirty(pfn_to_page(*hpa >> PAGE_SHIFT)); | |
4e13c1ac AK |
995 | |
996 | /* if (unlikely(ret)) | |
997 | pr_err("iommu_tce: %s failed on hwaddr=%lx ioba=%lx kva=%lx ret=%d\n", | |
84f1966e | 998 | __func__, hwaddr, entry << tbl->it_page_shift, |
4e13c1ac AK |
999 | hwaddr, ret); */ |
1000 | ||
1001 | return ret; | |
1002 | } | |
05c6cfb9 | 1003 | EXPORT_SYMBOL_GPL(iommu_tce_xchg); |
4e13c1ac | 1004 | |
4e13c1ac AK |
1005 | int iommu_take_ownership(struct iommu_table *tbl) |
1006 | { | |
b82c75bf AK |
1007 | unsigned long flags, i, sz = (tbl->it_size + 7) >> 3; |
1008 | int ret = 0; | |
1009 | ||
05c6cfb9 AK |
1010 | /* |
1011 | * VFIO does not control TCE entries allocation and the guest | |
1012 | * can write new TCEs on top of existing ones so iommu_tce_build() | |
1013 | * must be able to release old pages. This functionality | |
1014 | * requires exchange() callback defined so if it is not | |
1015 | * implemented, we disallow taking ownership over the table. | |
1016 | */ | |
1017 | if (!tbl->it_ops->exchange) | |
1018 | return -EINVAL; | |
1019 | ||
b82c75bf AK |
1020 | spin_lock_irqsave(&tbl->large_pool.lock, flags); |
1021 | for (i = 0; i < tbl->nr_pools; i++) | |
1022 | spin_lock(&tbl->pools[i].lock); | |
4e13c1ac AK |
1023 | |
1024 | if (tbl->it_offset == 0) | |
1025 | clear_bit(0, tbl->it_map); | |
1026 | ||
1027 | if (!bitmap_empty(tbl->it_map, tbl->it_size)) { | |
1028 | pr_err("iommu_tce: it_map is not empty"); | |
b82c75bf AK |
1029 | ret = -EBUSY; |
1030 | /* Restore bit#0 set by iommu_init_table() */ | |
1031 | if (tbl->it_offset == 0) | |
1032 | set_bit(0, tbl->it_map); | |
1033 | } else { | |
1034 | memset(tbl->it_map, 0xff, sz); | |
4e13c1ac AK |
1035 | } |
1036 | ||
b82c75bf AK |
1037 | for (i = 0; i < tbl->nr_pools; i++) |
1038 | spin_unlock(&tbl->pools[i].lock); | |
1039 | spin_unlock_irqrestore(&tbl->large_pool.lock, flags); | |
4e13c1ac | 1040 | |
b82c75bf | 1041 | return ret; |
4e13c1ac AK |
1042 | } |
1043 | EXPORT_SYMBOL_GPL(iommu_take_ownership); | |
1044 | ||
1045 | void iommu_release_ownership(struct iommu_table *tbl) | |
1046 | { | |
b82c75bf AK |
1047 | unsigned long flags, i, sz = (tbl->it_size + 7) >> 3; |
1048 | ||
1049 | spin_lock_irqsave(&tbl->large_pool.lock, flags); | |
1050 | for (i = 0; i < tbl->nr_pools; i++) | |
1051 | spin_lock(&tbl->pools[i].lock); | |
4e13c1ac | 1052 | |
4e13c1ac AK |
1053 | memset(tbl->it_map, 0, sz); |
1054 | ||
1055 | /* Restore bit#0 set by iommu_init_table() */ | |
1056 | if (tbl->it_offset == 0) | |
1057 | set_bit(0, tbl->it_map); | |
b82c75bf AK |
1058 | |
1059 | for (i = 0; i < tbl->nr_pools; i++) | |
1060 | spin_unlock(&tbl->pools[i].lock); | |
1061 | spin_unlock_irqrestore(&tbl->large_pool.lock, flags); | |
4e13c1ac AK |
1062 | } |
1063 | EXPORT_SYMBOL_GPL(iommu_release_ownership); | |
1064 | ||
d905c5df | 1065 | int iommu_add_device(struct device *dev) |
4e13c1ac AK |
1066 | { |
1067 | struct iommu_table *tbl; | |
0eaf4def | 1068 | struct iommu_table_group_link *tgl; |
4e13c1ac | 1069 | |
763fe0ad GS |
1070 | /* |
1071 | * The sysfs entries should be populated before | |
1072 | * binding IOMMU group. If sysfs entries isn't | |
1073 | * ready, we simply bail. | |
1074 | */ | |
1075 | if (!device_is_registered(dev)) | |
1076 | return -ENOENT; | |
1077 | ||
1078 | if (dev->iommu_group) { | |
1079 | pr_debug("%s: Skipping device %s with iommu group %d\n", | |
1080 | __func__, dev_name(dev), | |
1081 | iommu_group_id(dev->iommu_group)); | |
4e13c1ac AK |
1082 | return -EBUSY; |
1083 | } | |
1084 | ||
1085 | tbl = get_iommu_table_base(dev); | |
0eaf4def | 1086 | if (!tbl) { |
763fe0ad GS |
1087 | pr_debug("%s: Skipping device %s with no tbl\n", |
1088 | __func__, dev_name(dev)); | |
4e13c1ac AK |
1089 | return 0; |
1090 | } | |
1091 | ||
0eaf4def AK |
1092 | tgl = list_first_entry_or_null(&tbl->it_group_list, |
1093 | struct iommu_table_group_link, next); | |
1094 | if (!tgl) { | |
1095 | pr_debug("%s: Skipping device %s with no group\n", | |
1096 | __func__, dev_name(dev)); | |
1097 | return 0; | |
1098 | } | |
763fe0ad GS |
1099 | pr_debug("%s: Adding %s to iommu group %d\n", |
1100 | __func__, dev_name(dev), | |
0eaf4def | 1101 | iommu_group_id(tgl->table_group->group)); |
4e13c1ac | 1102 | |
d0847757 | 1103 | if (PAGE_SIZE < IOMMU_PAGE_SIZE(tbl)) { |
763fe0ad GS |
1104 | pr_err("%s: Invalid IOMMU page size %lx (%lx) on %s\n", |
1105 | __func__, IOMMU_PAGE_SIZE(tbl), | |
1106 | PAGE_SIZE, dev_name(dev)); | |
d0847757 AP |
1107 | return -EINVAL; |
1108 | } | |
1109 | ||
0eaf4def | 1110 | return iommu_group_add_device(tgl->table_group->group, dev); |
4e13c1ac | 1111 | } |
d905c5df | 1112 | EXPORT_SYMBOL_GPL(iommu_add_device); |
4e13c1ac | 1113 | |
d905c5df | 1114 | void iommu_del_device(struct device *dev) |
4e13c1ac | 1115 | { |
0c4b9e27 GS |
1116 | /* |
1117 | * Some devices might not have IOMMU table and group | |
1118 | * and we needn't detach them from the associated | |
1119 | * IOMMU groups | |
1120 | */ | |
1121 | if (!dev->iommu_group) { | |
1122 | pr_debug("iommu_tce: skipping device %s with no tbl\n", | |
1123 | dev_name(dev)); | |
1124 | return; | |
1125 | } | |
1126 | ||
4e13c1ac AK |
1127 | iommu_group_remove_device(dev); |
1128 | } | |
d905c5df | 1129 | EXPORT_SYMBOL_GPL(iommu_del_device); |
4e13c1ac | 1130 | |
4ad04e59 NA |
1131 | static int tce_iommu_bus_notifier(struct notifier_block *nb, |
1132 | unsigned long action, void *data) | |
1133 | { | |
1134 | struct device *dev = data; | |
1135 | ||
1136 | switch (action) { | |
1137 | case BUS_NOTIFY_ADD_DEVICE: | |
1138 | return iommu_add_device(dev); | |
1139 | case BUS_NOTIFY_DEL_DEVICE: | |
1140 | if (dev->iommu_group) | |
1141 | iommu_del_device(dev); | |
1142 | return 0; | |
1143 | default: | |
1144 | return 0; | |
1145 | } | |
1146 | } | |
1147 | ||
1148 | static struct notifier_block tce_iommu_bus_nb = { | |
1149 | .notifier_call = tce_iommu_bus_notifier, | |
1150 | }; | |
1151 | ||
1152 | int __init tce_iommu_bus_notifier_init(void) | |
1153 | { | |
1154 | bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb); | |
1155 | return 0; | |
1156 | } | |
4e13c1ac | 1157 | #endif /* CONFIG_IOMMU_API */ |