Commit | Line | Data |
---|---|---|
ba395927 KA |
1 | /* |
2 | * Copyright (c) 2006, Intel Corporation. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify it | |
5 | * under the terms and conditions of the GNU General Public License, | |
6 | * version 2, as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope it will be useful, but WITHOUT | |
9 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
10 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
11 | * more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public License along with | |
14 | * this program; if not, write to the Free Software Foundation, Inc., 59 Temple | |
15 | * Place - Suite 330, Boston, MA 02111-1307 USA. | |
16 | * | |
98bcef56 | 17 | * Copyright (C) 2006-2008 Intel Corporation |
18 | * Author: Ashok Raj <ashok.raj@intel.com> | |
19 | * Author: Shaohua Li <shaohua.li@intel.com> | |
20 | * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com> | |
5b6985ce | 21 | * Author: Fenghua Yu <fenghua.yu@intel.com> |
ba395927 KA |
22 | */ |
23 | ||
24 | #include <linux/init.h> | |
25 | #include <linux/bitmap.h> | |
5e0d2a6f | 26 | #include <linux/debugfs.h> |
ba395927 KA |
27 | #include <linux/slab.h> |
28 | #include <linux/irq.h> | |
29 | #include <linux/interrupt.h> | |
ba395927 KA |
30 | #include <linux/spinlock.h> |
31 | #include <linux/pci.h> | |
32 | #include <linux/dmar.h> | |
33 | #include <linux/dma-mapping.h> | |
34 | #include <linux/mempool.h> | |
5e0d2a6f | 35 | #include <linux/timer.h> |
38717946 | 36 | #include <linux/iova.h> |
5d450806 | 37 | #include <linux/iommu.h> |
38717946 | 38 | #include <linux/intel-iommu.h> |
ba395927 | 39 | #include <asm/cacheflush.h> |
46a7fa27 | 40 | #include <asm/iommu.h> |
ba395927 KA |
41 | #include "pci.h" |
42 | ||
5b6985ce FY |
43 | #define ROOT_SIZE VTD_PAGE_SIZE |
44 | #define CONTEXT_SIZE VTD_PAGE_SIZE | |
45 | ||
ba395927 KA |
46 | #define IS_GFX_DEVICE(pdev) ((pdev->class >> 16) == PCI_BASE_CLASS_DISPLAY) |
47 | #define IS_ISA_DEVICE(pdev) ((pdev->class >> 8) == PCI_CLASS_BRIDGE_ISA) | |
48 | ||
49 | #define IOAPIC_RANGE_START (0xfee00000) | |
50 | #define IOAPIC_RANGE_END (0xfeefffff) | |
51 | #define IOVA_START_ADDR (0x1000) | |
52 | ||
53 | #define DEFAULT_DOMAIN_ADDRESS_WIDTH 48 | |
54 | ||
ba395927 KA |
55 | #define DOMAIN_MAX_ADDR(gaw) ((((u64)1) << gaw) - 1) |
56 | ||
f27be03b MM |
57 | #define IOVA_PFN(addr) ((addr) >> PAGE_SHIFT) |
58 | #define DMA_32BIT_PFN IOVA_PFN(DMA_32BIT_MASK) | |
59 | #define DMA_64BIT_PFN IOVA_PFN(DMA_64BIT_MASK) | |
5e0d2a6f | 60 | |
d9630fe9 WH |
61 | /* global iommu list, set NULL for ignored DMAR units */ |
62 | static struct intel_iommu **g_iommus; | |
63 | ||
9af88143 DW |
64 | static int rwbf_quirk; |
65 | ||
46b08e1a MM |
66 | /* |
67 | * 0: Present | |
68 | * 1-11: Reserved | |
69 | * 12-63: Context Ptr (12 - (haw-1)) | |
70 | * 64-127: Reserved | |
71 | */ | |
72 | struct root_entry { | |
73 | u64 val; | |
74 | u64 rsvd1; | |
75 | }; | |
76 | #define ROOT_ENTRY_NR (VTD_PAGE_SIZE/sizeof(struct root_entry)) | |
77 | static inline bool root_present(struct root_entry *root) | |
78 | { | |
79 | return (root->val & 1); | |
80 | } | |
81 | static inline void set_root_present(struct root_entry *root) | |
82 | { | |
83 | root->val |= 1; | |
84 | } | |
85 | static inline void set_root_value(struct root_entry *root, unsigned long value) | |
86 | { | |
87 | root->val |= value & VTD_PAGE_MASK; | |
88 | } | |
89 | ||
90 | static inline struct context_entry * | |
91 | get_context_addr_from_root(struct root_entry *root) | |
92 | { | |
93 | return (struct context_entry *) | |
94 | (root_present(root)?phys_to_virt( | |
95 | root->val & VTD_PAGE_MASK) : | |
96 | NULL); | |
97 | } | |
98 | ||
7a8fc25e MM |
99 | /* |
100 | * low 64 bits: | |
101 | * 0: present | |
102 | * 1: fault processing disable | |
103 | * 2-3: translation type | |
104 | * 12-63: address space root | |
105 | * high 64 bits: | |
106 | * 0-2: address width | |
107 | * 3-6: aval | |
108 | * 8-23: domain id | |
109 | */ | |
110 | struct context_entry { | |
111 | u64 lo; | |
112 | u64 hi; | |
113 | }; | |
c07e7d21 MM |
114 | |
115 | static inline bool context_present(struct context_entry *context) | |
116 | { | |
117 | return (context->lo & 1); | |
118 | } | |
119 | static inline void context_set_present(struct context_entry *context) | |
120 | { | |
121 | context->lo |= 1; | |
122 | } | |
123 | ||
124 | static inline void context_set_fault_enable(struct context_entry *context) | |
125 | { | |
126 | context->lo &= (((u64)-1) << 2) | 1; | |
127 | } | |
128 | ||
7a8fc25e | 129 | #define CONTEXT_TT_MULTI_LEVEL 0 |
c07e7d21 MM |
130 | |
131 | static inline void context_set_translation_type(struct context_entry *context, | |
132 | unsigned long value) | |
133 | { | |
134 | context->lo &= (((u64)-1) << 4) | 3; | |
135 | context->lo |= (value & 3) << 2; | |
136 | } | |
137 | ||
138 | static inline void context_set_address_root(struct context_entry *context, | |
139 | unsigned long value) | |
140 | { | |
141 | context->lo |= value & VTD_PAGE_MASK; | |
142 | } | |
143 | ||
144 | static inline void context_set_address_width(struct context_entry *context, | |
145 | unsigned long value) | |
146 | { | |
147 | context->hi |= value & 7; | |
148 | } | |
149 | ||
150 | static inline void context_set_domain_id(struct context_entry *context, | |
151 | unsigned long value) | |
152 | { | |
153 | context->hi |= (value & ((1 << 16) - 1)) << 8; | |
154 | } | |
155 | ||
156 | static inline void context_clear_entry(struct context_entry *context) | |
157 | { | |
158 | context->lo = 0; | |
159 | context->hi = 0; | |
160 | } | |
7a8fc25e | 161 | |
622ba12a MM |
162 | /* |
163 | * 0: readable | |
164 | * 1: writable | |
165 | * 2-6: reserved | |
166 | * 7: super page | |
167 | * 8-11: available | |
168 | * 12-63: Host physcial address | |
169 | */ | |
170 | struct dma_pte { | |
171 | u64 val; | |
172 | }; | |
622ba12a | 173 | |
19c239ce MM |
174 | static inline void dma_clear_pte(struct dma_pte *pte) |
175 | { | |
176 | pte->val = 0; | |
177 | } | |
178 | ||
179 | static inline void dma_set_pte_readable(struct dma_pte *pte) | |
180 | { | |
181 | pte->val |= DMA_PTE_READ; | |
182 | } | |
183 | ||
184 | static inline void dma_set_pte_writable(struct dma_pte *pte) | |
185 | { | |
186 | pte->val |= DMA_PTE_WRITE; | |
187 | } | |
188 | ||
189 | static inline void dma_set_pte_prot(struct dma_pte *pte, unsigned long prot) | |
190 | { | |
191 | pte->val = (pte->val & ~3) | (prot & 3); | |
192 | } | |
193 | ||
194 | static inline u64 dma_pte_addr(struct dma_pte *pte) | |
195 | { | |
196 | return (pte->val & VTD_PAGE_MASK); | |
197 | } | |
198 | ||
199 | static inline void dma_set_pte_addr(struct dma_pte *pte, u64 addr) | |
200 | { | |
201 | pte->val |= (addr & VTD_PAGE_MASK); | |
202 | } | |
203 | ||
204 | static inline bool dma_pte_present(struct dma_pte *pte) | |
205 | { | |
206 | return (pte->val & 3) != 0; | |
207 | } | |
622ba12a | 208 | |
3b5410e7 | 209 | /* devices under the same p2p bridge are owned in one domain */ |
cdc7b837 | 210 | #define DOMAIN_FLAG_P2P_MULTIPLE_DEVICES (1 << 0) |
3b5410e7 | 211 | |
1ce28feb WH |
212 | /* domain represents a virtual machine, more than one devices |
213 | * across iommus may be owned in one domain, e.g. kvm guest. | |
214 | */ | |
215 | #define DOMAIN_FLAG_VIRTUAL_MACHINE (1 << 1) | |
216 | ||
99126f7c MM |
217 | struct dmar_domain { |
218 | int id; /* domain id */ | |
8c11e798 | 219 | unsigned long iommu_bmp; /* bitmap of iommus this domain uses*/ |
99126f7c MM |
220 | |
221 | struct list_head devices; /* all devices' list */ | |
222 | struct iova_domain iovad; /* iova's that belong to this domain */ | |
223 | ||
224 | struct dma_pte *pgd; /* virtual address */ | |
225 | spinlock_t mapping_lock; /* page table lock */ | |
226 | int gaw; /* max guest address width */ | |
227 | ||
228 | /* adjusted guest address width, 0 is level 2 30-bit */ | |
229 | int agaw; | |
230 | ||
3b5410e7 | 231 | int flags; /* flags to find out type of domain */ |
8e604097 WH |
232 | |
233 | int iommu_coherency;/* indicate coherency of iommu access */ | |
c7151a8d WH |
234 | int iommu_count; /* reference count of iommu */ |
235 | spinlock_t iommu_lock; /* protect iommu set in domain */ | |
fe40f1e0 | 236 | u64 max_addr; /* maximum mapped address */ |
99126f7c MM |
237 | }; |
238 | ||
a647dacb MM |
239 | /* PCI domain-device relationship */ |
240 | struct device_domain_info { | |
241 | struct list_head link; /* link to domain siblings */ | |
242 | struct list_head global; /* link to global list */ | |
243 | u8 bus; /* PCI bus numer */ | |
244 | u8 devfn; /* PCI devfn number */ | |
245 | struct pci_dev *dev; /* it's NULL for PCIE-to-PCI bridge */ | |
246 | struct dmar_domain *domain; /* pointer to domain */ | |
247 | }; | |
248 | ||
5e0d2a6f | 249 | static void flush_unmaps_timeout(unsigned long data); |
250 | ||
251 | DEFINE_TIMER(unmap_timer, flush_unmaps_timeout, 0, 0); | |
252 | ||
80b20dd8 | 253 | #define HIGH_WATER_MARK 250 |
254 | struct deferred_flush_tables { | |
255 | int next; | |
256 | struct iova *iova[HIGH_WATER_MARK]; | |
257 | struct dmar_domain *domain[HIGH_WATER_MARK]; | |
258 | }; | |
259 | ||
260 | static struct deferred_flush_tables *deferred_flush; | |
261 | ||
5e0d2a6f | 262 | /* bitmap for indexing intel_iommus */ |
5e0d2a6f | 263 | static int g_num_of_iommus; |
264 | ||
265 | static DEFINE_SPINLOCK(async_umap_flush_lock); | |
266 | static LIST_HEAD(unmaps_to_do); | |
267 | ||
268 | static int timer_on; | |
269 | static long list_size; | |
5e0d2a6f | 270 | |
ba395927 KA |
271 | static void domain_remove_dev_info(struct dmar_domain *domain); |
272 | ||
0cd5c3c8 KM |
273 | #ifdef CONFIG_DMAR_DEFAULT_ON |
274 | int dmar_disabled = 0; | |
275 | #else | |
276 | int dmar_disabled = 1; | |
277 | #endif /*CONFIG_DMAR_DEFAULT_ON*/ | |
278 | ||
ba395927 | 279 | static int __initdata dmar_map_gfx = 1; |
7d3b03ce | 280 | static int dmar_forcedac; |
5e0d2a6f | 281 | static int intel_iommu_strict; |
ba395927 KA |
282 | |
283 | #define DUMMY_DEVICE_DOMAIN_INFO ((struct device_domain_info *)(-1)) | |
284 | static DEFINE_SPINLOCK(device_domain_lock); | |
285 | static LIST_HEAD(device_domain_list); | |
286 | ||
a8bcbb0d JR |
287 | static struct iommu_ops intel_iommu_ops; |
288 | ||
ba395927 KA |
289 | static int __init intel_iommu_setup(char *str) |
290 | { | |
291 | if (!str) | |
292 | return -EINVAL; | |
293 | while (*str) { | |
0cd5c3c8 KM |
294 | if (!strncmp(str, "on", 2)) { |
295 | dmar_disabled = 0; | |
296 | printk(KERN_INFO "Intel-IOMMU: enabled\n"); | |
297 | } else if (!strncmp(str, "off", 3)) { | |
ba395927 | 298 | dmar_disabled = 1; |
0cd5c3c8 | 299 | printk(KERN_INFO "Intel-IOMMU: disabled\n"); |
ba395927 KA |
300 | } else if (!strncmp(str, "igfx_off", 8)) { |
301 | dmar_map_gfx = 0; | |
302 | printk(KERN_INFO | |
303 | "Intel-IOMMU: disable GFX device mapping\n"); | |
7d3b03ce | 304 | } else if (!strncmp(str, "forcedac", 8)) { |
5e0d2a6f | 305 | printk(KERN_INFO |
7d3b03ce KA |
306 | "Intel-IOMMU: Forcing DAC for PCI devices\n"); |
307 | dmar_forcedac = 1; | |
5e0d2a6f | 308 | } else if (!strncmp(str, "strict", 6)) { |
309 | printk(KERN_INFO | |
310 | "Intel-IOMMU: disable batched IOTLB flush\n"); | |
311 | intel_iommu_strict = 1; | |
ba395927 KA |
312 | } |
313 | ||
314 | str += strcspn(str, ","); | |
315 | while (*str == ',') | |
316 | str++; | |
317 | } | |
318 | return 0; | |
319 | } | |
320 | __setup("intel_iommu=", intel_iommu_setup); | |
321 | ||
322 | static struct kmem_cache *iommu_domain_cache; | |
323 | static struct kmem_cache *iommu_devinfo_cache; | |
324 | static struct kmem_cache *iommu_iova_cache; | |
325 | ||
eb3fa7cb KA |
326 | static inline void *iommu_kmem_cache_alloc(struct kmem_cache *cachep) |
327 | { | |
328 | unsigned int flags; | |
329 | void *vaddr; | |
330 | ||
331 | /* trying to avoid low memory issues */ | |
332 | flags = current->flags & PF_MEMALLOC; | |
333 | current->flags |= PF_MEMALLOC; | |
334 | vaddr = kmem_cache_alloc(cachep, GFP_ATOMIC); | |
335 | current->flags &= (~PF_MEMALLOC | flags); | |
336 | return vaddr; | |
337 | } | |
338 | ||
339 | ||
ba395927 KA |
340 | static inline void *alloc_pgtable_page(void) |
341 | { | |
eb3fa7cb KA |
342 | unsigned int flags; |
343 | void *vaddr; | |
344 | ||
345 | /* trying to avoid low memory issues */ | |
346 | flags = current->flags & PF_MEMALLOC; | |
347 | current->flags |= PF_MEMALLOC; | |
348 | vaddr = (void *)get_zeroed_page(GFP_ATOMIC); | |
349 | current->flags &= (~PF_MEMALLOC | flags); | |
350 | return vaddr; | |
ba395927 KA |
351 | } |
352 | ||
353 | static inline void free_pgtable_page(void *vaddr) | |
354 | { | |
355 | free_page((unsigned long)vaddr); | |
356 | } | |
357 | ||
358 | static inline void *alloc_domain_mem(void) | |
359 | { | |
eb3fa7cb | 360 | return iommu_kmem_cache_alloc(iommu_domain_cache); |
ba395927 KA |
361 | } |
362 | ||
38717946 | 363 | static void free_domain_mem(void *vaddr) |
ba395927 KA |
364 | { |
365 | kmem_cache_free(iommu_domain_cache, vaddr); | |
366 | } | |
367 | ||
368 | static inline void * alloc_devinfo_mem(void) | |
369 | { | |
eb3fa7cb | 370 | return iommu_kmem_cache_alloc(iommu_devinfo_cache); |
ba395927 KA |
371 | } |
372 | ||
373 | static inline void free_devinfo_mem(void *vaddr) | |
374 | { | |
375 | kmem_cache_free(iommu_devinfo_cache, vaddr); | |
376 | } | |
377 | ||
378 | struct iova *alloc_iova_mem(void) | |
379 | { | |
eb3fa7cb | 380 | return iommu_kmem_cache_alloc(iommu_iova_cache); |
ba395927 KA |
381 | } |
382 | ||
383 | void free_iova_mem(struct iova *iova) | |
384 | { | |
385 | kmem_cache_free(iommu_iova_cache, iova); | |
386 | } | |
387 | ||
1b573683 WH |
388 | |
389 | static inline int width_to_agaw(int width); | |
390 | ||
391 | /* calculate agaw for each iommu. | |
392 | * "SAGAW" may be different across iommus, use a default agaw, and | |
393 | * get a supported less agaw for iommus that don't support the default agaw. | |
394 | */ | |
395 | int iommu_calculate_agaw(struct intel_iommu *iommu) | |
396 | { | |
397 | unsigned long sagaw; | |
398 | int agaw = -1; | |
399 | ||
400 | sagaw = cap_sagaw(iommu->cap); | |
401 | for (agaw = width_to_agaw(DEFAULT_DOMAIN_ADDRESS_WIDTH); | |
402 | agaw >= 0; agaw--) { | |
403 | if (test_bit(agaw, &sagaw)) | |
404 | break; | |
405 | } | |
406 | ||
407 | return agaw; | |
408 | } | |
409 | ||
8c11e798 WH |
410 | /* in native case, each domain is related to only one iommu */ |
411 | static struct intel_iommu *domain_get_iommu(struct dmar_domain *domain) | |
412 | { | |
413 | int iommu_id; | |
414 | ||
1ce28feb WH |
415 | BUG_ON(domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE); |
416 | ||
8c11e798 WH |
417 | iommu_id = find_first_bit(&domain->iommu_bmp, g_num_of_iommus); |
418 | if (iommu_id < 0 || iommu_id >= g_num_of_iommus) | |
419 | return NULL; | |
420 | ||
421 | return g_iommus[iommu_id]; | |
422 | } | |
423 | ||
8e604097 WH |
424 | /* "Coherency" capability may be different across iommus */ |
425 | static void domain_update_iommu_coherency(struct dmar_domain *domain) | |
426 | { | |
427 | int i; | |
428 | ||
429 | domain->iommu_coherency = 1; | |
430 | ||
431 | i = find_first_bit(&domain->iommu_bmp, g_num_of_iommus); | |
432 | for (; i < g_num_of_iommus; ) { | |
433 | if (!ecap_coherent(g_iommus[i]->ecap)) { | |
434 | domain->iommu_coherency = 0; | |
435 | break; | |
436 | } | |
437 | i = find_next_bit(&domain->iommu_bmp, g_num_of_iommus, i+1); | |
438 | } | |
439 | } | |
440 | ||
c7151a8d WH |
441 | static struct intel_iommu *device_to_iommu(u8 bus, u8 devfn) |
442 | { | |
443 | struct dmar_drhd_unit *drhd = NULL; | |
444 | int i; | |
445 | ||
446 | for_each_drhd_unit(drhd) { | |
447 | if (drhd->ignored) | |
448 | continue; | |
449 | ||
450 | for (i = 0; i < drhd->devices_cnt; i++) | |
288e4877 DH |
451 | if (drhd->devices[i] && |
452 | drhd->devices[i]->bus->number == bus && | |
c7151a8d WH |
453 | drhd->devices[i]->devfn == devfn) |
454 | return drhd->iommu; | |
455 | ||
456 | if (drhd->include_all) | |
457 | return drhd->iommu; | |
458 | } | |
459 | ||
460 | return NULL; | |
461 | } | |
462 | ||
5331fe6f WH |
463 | static void domain_flush_cache(struct dmar_domain *domain, |
464 | void *addr, int size) | |
465 | { | |
466 | if (!domain->iommu_coherency) | |
467 | clflush_cache_range(addr, size); | |
468 | } | |
469 | ||
ba395927 KA |
470 | /* Gets context entry for a given bus and devfn */ |
471 | static struct context_entry * device_to_context_entry(struct intel_iommu *iommu, | |
472 | u8 bus, u8 devfn) | |
473 | { | |
474 | struct root_entry *root; | |
475 | struct context_entry *context; | |
476 | unsigned long phy_addr; | |
477 | unsigned long flags; | |
478 | ||
479 | spin_lock_irqsave(&iommu->lock, flags); | |
480 | root = &iommu->root_entry[bus]; | |
481 | context = get_context_addr_from_root(root); | |
482 | if (!context) { | |
483 | context = (struct context_entry *)alloc_pgtable_page(); | |
484 | if (!context) { | |
485 | spin_unlock_irqrestore(&iommu->lock, flags); | |
486 | return NULL; | |
487 | } | |
5b6985ce | 488 | __iommu_flush_cache(iommu, (void *)context, CONTEXT_SIZE); |
ba395927 KA |
489 | phy_addr = virt_to_phys((void *)context); |
490 | set_root_value(root, phy_addr); | |
491 | set_root_present(root); | |
492 | __iommu_flush_cache(iommu, root, sizeof(*root)); | |
493 | } | |
494 | spin_unlock_irqrestore(&iommu->lock, flags); | |
495 | return &context[devfn]; | |
496 | } | |
497 | ||
498 | static int device_context_mapped(struct intel_iommu *iommu, u8 bus, u8 devfn) | |
499 | { | |
500 | struct root_entry *root; | |
501 | struct context_entry *context; | |
502 | int ret; | |
503 | unsigned long flags; | |
504 | ||
505 | spin_lock_irqsave(&iommu->lock, flags); | |
506 | root = &iommu->root_entry[bus]; | |
507 | context = get_context_addr_from_root(root); | |
508 | if (!context) { | |
509 | ret = 0; | |
510 | goto out; | |
511 | } | |
c07e7d21 | 512 | ret = context_present(&context[devfn]); |
ba395927 KA |
513 | out: |
514 | spin_unlock_irqrestore(&iommu->lock, flags); | |
515 | return ret; | |
516 | } | |
517 | ||
518 | static void clear_context_table(struct intel_iommu *iommu, u8 bus, u8 devfn) | |
519 | { | |
520 | struct root_entry *root; | |
521 | struct context_entry *context; | |
522 | unsigned long flags; | |
523 | ||
524 | spin_lock_irqsave(&iommu->lock, flags); | |
525 | root = &iommu->root_entry[bus]; | |
526 | context = get_context_addr_from_root(root); | |
527 | if (context) { | |
c07e7d21 | 528 | context_clear_entry(&context[devfn]); |
ba395927 KA |
529 | __iommu_flush_cache(iommu, &context[devfn], \ |
530 | sizeof(*context)); | |
531 | } | |
532 | spin_unlock_irqrestore(&iommu->lock, flags); | |
533 | } | |
534 | ||
535 | static void free_context_table(struct intel_iommu *iommu) | |
536 | { | |
537 | struct root_entry *root; | |
538 | int i; | |
539 | unsigned long flags; | |
540 | struct context_entry *context; | |
541 | ||
542 | spin_lock_irqsave(&iommu->lock, flags); | |
543 | if (!iommu->root_entry) { | |
544 | goto out; | |
545 | } | |
546 | for (i = 0; i < ROOT_ENTRY_NR; i++) { | |
547 | root = &iommu->root_entry[i]; | |
548 | context = get_context_addr_from_root(root); | |
549 | if (context) | |
550 | free_pgtable_page(context); | |
551 | } | |
552 | free_pgtable_page(iommu->root_entry); | |
553 | iommu->root_entry = NULL; | |
554 | out: | |
555 | spin_unlock_irqrestore(&iommu->lock, flags); | |
556 | } | |
557 | ||
558 | /* page table handling */ | |
559 | #define LEVEL_STRIDE (9) | |
560 | #define LEVEL_MASK (((u64)1 << LEVEL_STRIDE) - 1) | |
561 | ||
562 | static inline int agaw_to_level(int agaw) | |
563 | { | |
564 | return agaw + 2; | |
565 | } | |
566 | ||
567 | static inline int agaw_to_width(int agaw) | |
568 | { | |
569 | return 30 + agaw * LEVEL_STRIDE; | |
570 | ||
571 | } | |
572 | ||
573 | static inline int width_to_agaw(int width) | |
574 | { | |
575 | return (width - 30) / LEVEL_STRIDE; | |
576 | } | |
577 | ||
578 | static inline unsigned int level_to_offset_bits(int level) | |
579 | { | |
580 | return (12 + (level - 1) * LEVEL_STRIDE); | |
581 | } | |
582 | ||
583 | static inline int address_level_offset(u64 addr, int level) | |
584 | { | |
585 | return ((addr >> level_to_offset_bits(level)) & LEVEL_MASK); | |
586 | } | |
587 | ||
588 | static inline u64 level_mask(int level) | |
589 | { | |
590 | return ((u64)-1 << level_to_offset_bits(level)); | |
591 | } | |
592 | ||
593 | static inline u64 level_size(int level) | |
594 | { | |
595 | return ((u64)1 << level_to_offset_bits(level)); | |
596 | } | |
597 | ||
598 | static inline u64 align_to_level(u64 addr, int level) | |
599 | { | |
600 | return ((addr + level_size(level) - 1) & level_mask(level)); | |
601 | } | |
602 | ||
603 | static struct dma_pte * addr_to_dma_pte(struct dmar_domain *domain, u64 addr) | |
604 | { | |
605 | int addr_width = agaw_to_width(domain->agaw); | |
606 | struct dma_pte *parent, *pte = NULL; | |
607 | int level = agaw_to_level(domain->agaw); | |
608 | int offset; | |
609 | unsigned long flags; | |
610 | ||
611 | BUG_ON(!domain->pgd); | |
612 | ||
613 | addr &= (((u64)1) << addr_width) - 1; | |
614 | parent = domain->pgd; | |
615 | ||
616 | spin_lock_irqsave(&domain->mapping_lock, flags); | |
617 | while (level > 0) { | |
618 | void *tmp_page; | |
619 | ||
620 | offset = address_level_offset(addr, level); | |
621 | pte = &parent[offset]; | |
622 | if (level == 1) | |
623 | break; | |
624 | ||
19c239ce | 625 | if (!dma_pte_present(pte)) { |
ba395927 KA |
626 | tmp_page = alloc_pgtable_page(); |
627 | ||
628 | if (!tmp_page) { | |
629 | spin_unlock_irqrestore(&domain->mapping_lock, | |
630 | flags); | |
631 | return NULL; | |
632 | } | |
5331fe6f | 633 | domain_flush_cache(domain, tmp_page, PAGE_SIZE); |
19c239ce | 634 | dma_set_pte_addr(pte, virt_to_phys(tmp_page)); |
ba395927 KA |
635 | /* |
636 | * high level table always sets r/w, last level page | |
637 | * table control read/write | |
638 | */ | |
19c239ce MM |
639 | dma_set_pte_readable(pte); |
640 | dma_set_pte_writable(pte); | |
5331fe6f | 641 | domain_flush_cache(domain, pte, sizeof(*pte)); |
ba395927 | 642 | } |
19c239ce | 643 | parent = phys_to_virt(dma_pte_addr(pte)); |
ba395927 KA |
644 | level--; |
645 | } | |
646 | ||
647 | spin_unlock_irqrestore(&domain->mapping_lock, flags); | |
648 | return pte; | |
649 | } | |
650 | ||
651 | /* return address's pte at specific level */ | |
652 | static struct dma_pte *dma_addr_level_pte(struct dmar_domain *domain, u64 addr, | |
653 | int level) | |
654 | { | |
655 | struct dma_pte *parent, *pte = NULL; | |
656 | int total = agaw_to_level(domain->agaw); | |
657 | int offset; | |
658 | ||
659 | parent = domain->pgd; | |
660 | while (level <= total) { | |
661 | offset = address_level_offset(addr, total); | |
662 | pte = &parent[offset]; | |
663 | if (level == total) | |
664 | return pte; | |
665 | ||
19c239ce | 666 | if (!dma_pte_present(pte)) |
ba395927 | 667 | break; |
19c239ce | 668 | parent = phys_to_virt(dma_pte_addr(pte)); |
ba395927 KA |
669 | total--; |
670 | } | |
671 | return NULL; | |
672 | } | |
673 | ||
674 | /* clear one page's page table */ | |
675 | static void dma_pte_clear_one(struct dmar_domain *domain, u64 addr) | |
676 | { | |
677 | struct dma_pte *pte = NULL; | |
678 | ||
679 | /* get last level pte */ | |
680 | pte = dma_addr_level_pte(domain, addr, 1); | |
681 | ||
682 | if (pte) { | |
19c239ce | 683 | dma_clear_pte(pte); |
5331fe6f | 684 | domain_flush_cache(domain, pte, sizeof(*pte)); |
ba395927 KA |
685 | } |
686 | } | |
687 | ||
688 | /* clear last level pte, a tlb flush should be followed */ | |
689 | static void dma_pte_clear_range(struct dmar_domain *domain, u64 start, u64 end) | |
690 | { | |
691 | int addr_width = agaw_to_width(domain->agaw); | |
692 | ||
693 | start &= (((u64)1) << addr_width) - 1; | |
694 | end &= (((u64)1) << addr_width) - 1; | |
695 | /* in case it's partial page */ | |
5b6985ce FY |
696 | start = PAGE_ALIGN(start); |
697 | end &= PAGE_MASK; | |
ba395927 KA |
698 | |
699 | /* we don't need lock here, nobody else touches the iova range */ | |
700 | while (start < end) { | |
701 | dma_pte_clear_one(domain, start); | |
5b6985ce | 702 | start += VTD_PAGE_SIZE; |
ba395927 KA |
703 | } |
704 | } | |
705 | ||
706 | /* free page table pages. last level pte should already be cleared */ | |
707 | static void dma_pte_free_pagetable(struct dmar_domain *domain, | |
708 | u64 start, u64 end) | |
709 | { | |
710 | int addr_width = agaw_to_width(domain->agaw); | |
711 | struct dma_pte *pte; | |
712 | int total = agaw_to_level(domain->agaw); | |
713 | int level; | |
714 | u64 tmp; | |
715 | ||
716 | start &= (((u64)1) << addr_width) - 1; | |
717 | end &= (((u64)1) << addr_width) - 1; | |
718 | ||
719 | /* we don't need lock here, nobody else touches the iova range */ | |
720 | level = 2; | |
721 | while (level <= total) { | |
722 | tmp = align_to_level(start, level); | |
723 | if (tmp >= end || (tmp + level_size(level) > end)) | |
724 | return; | |
725 | ||
726 | while (tmp < end) { | |
727 | pte = dma_addr_level_pte(domain, tmp, level); | |
728 | if (pte) { | |
729 | free_pgtable_page( | |
19c239ce MM |
730 | phys_to_virt(dma_pte_addr(pte))); |
731 | dma_clear_pte(pte); | |
5331fe6f | 732 | domain_flush_cache(domain, pte, sizeof(*pte)); |
ba395927 KA |
733 | } |
734 | tmp += level_size(level); | |
735 | } | |
736 | level++; | |
737 | } | |
738 | /* free pgd */ | |
739 | if (start == 0 && end >= ((((u64)1) << addr_width) - 1)) { | |
740 | free_pgtable_page(domain->pgd); | |
741 | domain->pgd = NULL; | |
742 | } | |
743 | } | |
744 | ||
745 | /* iommu handling */ | |
746 | static int iommu_alloc_root_entry(struct intel_iommu *iommu) | |
747 | { | |
748 | struct root_entry *root; | |
749 | unsigned long flags; | |
750 | ||
751 | root = (struct root_entry *)alloc_pgtable_page(); | |
752 | if (!root) | |
753 | return -ENOMEM; | |
754 | ||
5b6985ce | 755 | __iommu_flush_cache(iommu, root, ROOT_SIZE); |
ba395927 KA |
756 | |
757 | spin_lock_irqsave(&iommu->lock, flags); | |
758 | iommu->root_entry = root; | |
759 | spin_unlock_irqrestore(&iommu->lock, flags); | |
760 | ||
761 | return 0; | |
762 | } | |
763 | ||
ba395927 KA |
764 | static void iommu_set_root_entry(struct intel_iommu *iommu) |
765 | { | |
766 | void *addr; | |
767 | u32 cmd, sts; | |
768 | unsigned long flag; | |
769 | ||
770 | addr = iommu->root_entry; | |
771 | ||
772 | spin_lock_irqsave(&iommu->register_lock, flag); | |
773 | dmar_writeq(iommu->reg + DMAR_RTADDR_REG, virt_to_phys(addr)); | |
774 | ||
775 | cmd = iommu->gcmd | DMA_GCMD_SRTP; | |
776 | writel(cmd, iommu->reg + DMAR_GCMD_REG); | |
777 | ||
778 | /* Make sure hardware complete it */ | |
779 | IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, | |
780 | readl, (sts & DMA_GSTS_RTPS), sts); | |
781 | ||
782 | spin_unlock_irqrestore(&iommu->register_lock, flag); | |
783 | } | |
784 | ||
785 | static void iommu_flush_write_buffer(struct intel_iommu *iommu) | |
786 | { | |
787 | u32 val; | |
788 | unsigned long flag; | |
789 | ||
9af88143 | 790 | if (!rwbf_quirk && !cap_rwbf(iommu->cap)) |
ba395927 KA |
791 | return; |
792 | val = iommu->gcmd | DMA_GCMD_WBF; | |
793 | ||
794 | spin_lock_irqsave(&iommu->register_lock, flag); | |
795 | writel(val, iommu->reg + DMAR_GCMD_REG); | |
796 | ||
797 | /* Make sure hardware complete it */ | |
798 | IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, | |
799 | readl, (!(val & DMA_GSTS_WBFS)), val); | |
800 | ||
801 | spin_unlock_irqrestore(&iommu->register_lock, flag); | |
802 | } | |
803 | ||
804 | /* return value determine if we need a write buffer flush */ | |
805 | static int __iommu_flush_context(struct intel_iommu *iommu, | |
806 | u16 did, u16 source_id, u8 function_mask, u64 type, | |
807 | int non_present_entry_flush) | |
808 | { | |
809 | u64 val = 0; | |
810 | unsigned long flag; | |
811 | ||
812 | /* | |
813 | * In the non-present entry flush case, if hardware doesn't cache | |
814 | * non-present entry we do nothing and if hardware cache non-present | |
815 | * entry, we flush entries of domain 0 (the domain id is used to cache | |
816 | * any non-present entries) | |
817 | */ | |
818 | if (non_present_entry_flush) { | |
819 | if (!cap_caching_mode(iommu->cap)) | |
820 | return 1; | |
821 | else | |
822 | did = 0; | |
823 | } | |
824 | ||
825 | switch (type) { | |
826 | case DMA_CCMD_GLOBAL_INVL: | |
827 | val = DMA_CCMD_GLOBAL_INVL; | |
828 | break; | |
829 | case DMA_CCMD_DOMAIN_INVL: | |
830 | val = DMA_CCMD_DOMAIN_INVL|DMA_CCMD_DID(did); | |
831 | break; | |
832 | case DMA_CCMD_DEVICE_INVL: | |
833 | val = DMA_CCMD_DEVICE_INVL|DMA_CCMD_DID(did) | |
834 | | DMA_CCMD_SID(source_id) | DMA_CCMD_FM(function_mask); | |
835 | break; | |
836 | default: | |
837 | BUG(); | |
838 | } | |
839 | val |= DMA_CCMD_ICC; | |
840 | ||
841 | spin_lock_irqsave(&iommu->register_lock, flag); | |
842 | dmar_writeq(iommu->reg + DMAR_CCMD_REG, val); | |
843 | ||
844 | /* Make sure hardware complete it */ | |
845 | IOMMU_WAIT_OP(iommu, DMAR_CCMD_REG, | |
846 | dmar_readq, (!(val & DMA_CCMD_ICC)), val); | |
847 | ||
848 | spin_unlock_irqrestore(&iommu->register_lock, flag); | |
849 | ||
4d235ba6 | 850 | /* flush context entry will implicitly flush write buffer */ |
ba395927 KA |
851 | return 0; |
852 | } | |
853 | ||
ba395927 KA |
854 | /* return value determine if we need a write buffer flush */ |
855 | static int __iommu_flush_iotlb(struct intel_iommu *iommu, u16 did, | |
856 | u64 addr, unsigned int size_order, u64 type, | |
857 | int non_present_entry_flush) | |
858 | { | |
859 | int tlb_offset = ecap_iotlb_offset(iommu->ecap); | |
860 | u64 val = 0, val_iva = 0; | |
861 | unsigned long flag; | |
862 | ||
863 | /* | |
864 | * In the non-present entry flush case, if hardware doesn't cache | |
865 | * non-present entry we do nothing and if hardware cache non-present | |
866 | * entry, we flush entries of domain 0 (the domain id is used to cache | |
867 | * any non-present entries) | |
868 | */ | |
869 | if (non_present_entry_flush) { | |
870 | if (!cap_caching_mode(iommu->cap)) | |
871 | return 1; | |
872 | else | |
873 | did = 0; | |
874 | } | |
875 | ||
876 | switch (type) { | |
877 | case DMA_TLB_GLOBAL_FLUSH: | |
878 | /* global flush doesn't need set IVA_REG */ | |
879 | val = DMA_TLB_GLOBAL_FLUSH|DMA_TLB_IVT; | |
880 | break; | |
881 | case DMA_TLB_DSI_FLUSH: | |
882 | val = DMA_TLB_DSI_FLUSH|DMA_TLB_IVT|DMA_TLB_DID(did); | |
883 | break; | |
884 | case DMA_TLB_PSI_FLUSH: | |
885 | val = DMA_TLB_PSI_FLUSH|DMA_TLB_IVT|DMA_TLB_DID(did); | |
886 | /* Note: always flush non-leaf currently */ | |
887 | val_iva = size_order | addr; | |
888 | break; | |
889 | default: | |
890 | BUG(); | |
891 | } | |
892 | /* Note: set drain read/write */ | |
893 | #if 0 | |
894 | /* | |
895 | * This is probably to be super secure.. Looks like we can | |
896 | * ignore it without any impact. | |
897 | */ | |
898 | if (cap_read_drain(iommu->cap)) | |
899 | val |= DMA_TLB_READ_DRAIN; | |
900 | #endif | |
901 | if (cap_write_drain(iommu->cap)) | |
902 | val |= DMA_TLB_WRITE_DRAIN; | |
903 | ||
904 | spin_lock_irqsave(&iommu->register_lock, flag); | |
905 | /* Note: Only uses first TLB reg currently */ | |
906 | if (val_iva) | |
907 | dmar_writeq(iommu->reg + tlb_offset, val_iva); | |
908 | dmar_writeq(iommu->reg + tlb_offset + 8, val); | |
909 | ||
910 | /* Make sure hardware complete it */ | |
911 | IOMMU_WAIT_OP(iommu, tlb_offset + 8, | |
912 | dmar_readq, (!(val & DMA_TLB_IVT)), val); | |
913 | ||
914 | spin_unlock_irqrestore(&iommu->register_lock, flag); | |
915 | ||
916 | /* check IOTLB invalidation granularity */ | |
917 | if (DMA_TLB_IAIG(val) == 0) | |
918 | printk(KERN_ERR"IOMMU: flush IOTLB failed\n"); | |
919 | if (DMA_TLB_IAIG(val) != DMA_TLB_IIRG(type)) | |
920 | pr_debug("IOMMU: tlb flush request %Lx, actual %Lx\n", | |
5b6985ce FY |
921 | (unsigned long long)DMA_TLB_IIRG(type), |
922 | (unsigned long long)DMA_TLB_IAIG(val)); | |
4d235ba6 | 923 | /* flush iotlb entry will implicitly flush write buffer */ |
ba395927 KA |
924 | return 0; |
925 | } | |
926 | ||
ba395927 KA |
927 | static int iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did, |
928 | u64 addr, unsigned int pages, int non_present_entry_flush) | |
929 | { | |
f76aec76 | 930 | unsigned int mask; |
ba395927 | 931 | |
5b6985ce | 932 | BUG_ON(addr & (~VTD_PAGE_MASK)); |
ba395927 KA |
933 | BUG_ON(pages == 0); |
934 | ||
935 | /* Fallback to domain selective flush if no PSI support */ | |
936 | if (!cap_pgsel_inv(iommu->cap)) | |
a77b67d4 YS |
937 | return iommu->flush.flush_iotlb(iommu, did, 0, 0, |
938 | DMA_TLB_DSI_FLUSH, | |
939 | non_present_entry_flush); | |
ba395927 KA |
940 | |
941 | /* | |
942 | * PSI requires page size to be 2 ^ x, and the base address is naturally | |
943 | * aligned to the size | |
944 | */ | |
f76aec76 | 945 | mask = ilog2(__roundup_pow_of_two(pages)); |
ba395927 | 946 | /* Fallback to domain selective flush if size is too big */ |
f76aec76 | 947 | if (mask > cap_max_amask_val(iommu->cap)) |
a77b67d4 YS |
948 | return iommu->flush.flush_iotlb(iommu, did, 0, 0, |
949 | DMA_TLB_DSI_FLUSH, non_present_entry_flush); | |
ba395927 | 950 | |
a77b67d4 YS |
951 | return iommu->flush.flush_iotlb(iommu, did, addr, mask, |
952 | DMA_TLB_PSI_FLUSH, | |
953 | non_present_entry_flush); | |
ba395927 KA |
954 | } |
955 | ||
f8bab735 | 956 | static void iommu_disable_protect_mem_regions(struct intel_iommu *iommu) |
957 | { | |
958 | u32 pmen; | |
959 | unsigned long flags; | |
960 | ||
961 | spin_lock_irqsave(&iommu->register_lock, flags); | |
962 | pmen = readl(iommu->reg + DMAR_PMEN_REG); | |
963 | pmen &= ~DMA_PMEN_EPM; | |
964 | writel(pmen, iommu->reg + DMAR_PMEN_REG); | |
965 | ||
966 | /* wait for the protected region status bit to clear */ | |
967 | IOMMU_WAIT_OP(iommu, DMAR_PMEN_REG, | |
968 | readl, !(pmen & DMA_PMEN_PRS), pmen); | |
969 | ||
970 | spin_unlock_irqrestore(&iommu->register_lock, flags); | |
971 | } | |
972 | ||
ba395927 KA |
973 | static int iommu_enable_translation(struct intel_iommu *iommu) |
974 | { | |
975 | u32 sts; | |
976 | unsigned long flags; | |
977 | ||
978 | spin_lock_irqsave(&iommu->register_lock, flags); | |
979 | writel(iommu->gcmd|DMA_GCMD_TE, iommu->reg + DMAR_GCMD_REG); | |
980 | ||
981 | /* Make sure hardware complete it */ | |
982 | IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, | |
983 | readl, (sts & DMA_GSTS_TES), sts); | |
984 | ||
985 | iommu->gcmd |= DMA_GCMD_TE; | |
986 | spin_unlock_irqrestore(&iommu->register_lock, flags); | |
987 | return 0; | |
988 | } | |
989 | ||
990 | static int iommu_disable_translation(struct intel_iommu *iommu) | |
991 | { | |
992 | u32 sts; | |
993 | unsigned long flag; | |
994 | ||
995 | spin_lock_irqsave(&iommu->register_lock, flag); | |
996 | iommu->gcmd &= ~DMA_GCMD_TE; | |
997 | writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG); | |
998 | ||
999 | /* Make sure hardware complete it */ | |
1000 | IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, | |
1001 | readl, (!(sts & DMA_GSTS_TES)), sts); | |
1002 | ||
1003 | spin_unlock_irqrestore(&iommu->register_lock, flag); | |
1004 | return 0; | |
1005 | } | |
1006 | ||
3460a6d9 | 1007 | |
ba395927 KA |
1008 | static int iommu_init_domains(struct intel_iommu *iommu) |
1009 | { | |
1010 | unsigned long ndomains; | |
1011 | unsigned long nlongs; | |
1012 | ||
1013 | ndomains = cap_ndoms(iommu->cap); | |
1014 | pr_debug("Number of Domains supportd <%ld>\n", ndomains); | |
1015 | nlongs = BITS_TO_LONGS(ndomains); | |
1016 | ||
1017 | /* TBD: there might be 64K domains, | |
1018 | * consider other allocation for future chip | |
1019 | */ | |
1020 | iommu->domain_ids = kcalloc(nlongs, sizeof(unsigned long), GFP_KERNEL); | |
1021 | if (!iommu->domain_ids) { | |
1022 | printk(KERN_ERR "Allocating domain id array failed\n"); | |
1023 | return -ENOMEM; | |
1024 | } | |
1025 | iommu->domains = kcalloc(ndomains, sizeof(struct dmar_domain *), | |
1026 | GFP_KERNEL); | |
1027 | if (!iommu->domains) { | |
1028 | printk(KERN_ERR "Allocating domain array failed\n"); | |
1029 | kfree(iommu->domain_ids); | |
1030 | return -ENOMEM; | |
1031 | } | |
1032 | ||
e61d98d8 SS |
1033 | spin_lock_init(&iommu->lock); |
1034 | ||
ba395927 KA |
1035 | /* |
1036 | * if Caching mode is set, then invalid translations are tagged | |
1037 | * with domainid 0. Hence we need to pre-allocate it. | |
1038 | */ | |
1039 | if (cap_caching_mode(iommu->cap)) | |
1040 | set_bit(0, iommu->domain_ids); | |
1041 | return 0; | |
1042 | } | |
ba395927 | 1043 | |
ba395927 KA |
1044 | |
1045 | static void domain_exit(struct dmar_domain *domain); | |
5e98c4b1 | 1046 | static void vm_domain_exit(struct dmar_domain *domain); |
e61d98d8 SS |
1047 | |
1048 | void free_dmar_iommu(struct intel_iommu *iommu) | |
ba395927 KA |
1049 | { |
1050 | struct dmar_domain *domain; | |
1051 | int i; | |
c7151a8d | 1052 | unsigned long flags; |
ba395927 | 1053 | |
ba395927 KA |
1054 | i = find_first_bit(iommu->domain_ids, cap_ndoms(iommu->cap)); |
1055 | for (; i < cap_ndoms(iommu->cap); ) { | |
1056 | domain = iommu->domains[i]; | |
1057 | clear_bit(i, iommu->domain_ids); | |
c7151a8d WH |
1058 | |
1059 | spin_lock_irqsave(&domain->iommu_lock, flags); | |
5e98c4b1 WH |
1060 | if (--domain->iommu_count == 0) { |
1061 | if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) | |
1062 | vm_domain_exit(domain); | |
1063 | else | |
1064 | domain_exit(domain); | |
1065 | } | |
c7151a8d WH |
1066 | spin_unlock_irqrestore(&domain->iommu_lock, flags); |
1067 | ||
ba395927 KA |
1068 | i = find_next_bit(iommu->domain_ids, |
1069 | cap_ndoms(iommu->cap), i+1); | |
1070 | } | |
1071 | ||
1072 | if (iommu->gcmd & DMA_GCMD_TE) | |
1073 | iommu_disable_translation(iommu); | |
1074 | ||
1075 | if (iommu->irq) { | |
1076 | set_irq_data(iommu->irq, NULL); | |
1077 | /* This will mask the irq */ | |
1078 | free_irq(iommu->irq, iommu); | |
1079 | destroy_irq(iommu->irq); | |
1080 | } | |
1081 | ||
1082 | kfree(iommu->domains); | |
1083 | kfree(iommu->domain_ids); | |
1084 | ||
d9630fe9 WH |
1085 | g_iommus[iommu->seq_id] = NULL; |
1086 | ||
1087 | /* if all iommus are freed, free g_iommus */ | |
1088 | for (i = 0; i < g_num_of_iommus; i++) { | |
1089 | if (g_iommus[i]) | |
1090 | break; | |
1091 | } | |
1092 | ||
1093 | if (i == g_num_of_iommus) | |
1094 | kfree(g_iommus); | |
1095 | ||
ba395927 KA |
1096 | /* free context mapping */ |
1097 | free_context_table(iommu); | |
ba395927 KA |
1098 | } |
1099 | ||
1100 | static struct dmar_domain * iommu_alloc_domain(struct intel_iommu *iommu) | |
1101 | { | |
1102 | unsigned long num; | |
1103 | unsigned long ndomains; | |
1104 | struct dmar_domain *domain; | |
1105 | unsigned long flags; | |
1106 | ||
1107 | domain = alloc_domain_mem(); | |
1108 | if (!domain) | |
1109 | return NULL; | |
1110 | ||
1111 | ndomains = cap_ndoms(iommu->cap); | |
1112 | ||
1113 | spin_lock_irqsave(&iommu->lock, flags); | |
1114 | num = find_first_zero_bit(iommu->domain_ids, ndomains); | |
1115 | if (num >= ndomains) { | |
1116 | spin_unlock_irqrestore(&iommu->lock, flags); | |
1117 | free_domain_mem(domain); | |
1118 | printk(KERN_ERR "IOMMU: no free domain ids\n"); | |
1119 | return NULL; | |
1120 | } | |
1121 | ||
1122 | set_bit(num, iommu->domain_ids); | |
1123 | domain->id = num; | |
8c11e798 WH |
1124 | memset(&domain->iommu_bmp, 0, sizeof(unsigned long)); |
1125 | set_bit(iommu->seq_id, &domain->iommu_bmp); | |
d71a2f33 | 1126 | domain->flags = 0; |
ba395927 KA |
1127 | iommu->domains[num] = domain; |
1128 | spin_unlock_irqrestore(&iommu->lock, flags); | |
1129 | ||
1130 | return domain; | |
1131 | } | |
1132 | ||
1133 | static void iommu_free_domain(struct dmar_domain *domain) | |
1134 | { | |
1135 | unsigned long flags; | |
8c11e798 WH |
1136 | struct intel_iommu *iommu; |
1137 | ||
1138 | iommu = domain_get_iommu(domain); | |
ba395927 | 1139 | |
8c11e798 WH |
1140 | spin_lock_irqsave(&iommu->lock, flags); |
1141 | clear_bit(domain->id, iommu->domain_ids); | |
1142 | spin_unlock_irqrestore(&iommu->lock, flags); | |
ba395927 KA |
1143 | } |
1144 | ||
1145 | static struct iova_domain reserved_iova_list; | |
8a443df4 MG |
1146 | static struct lock_class_key reserved_alloc_key; |
1147 | static struct lock_class_key reserved_rbtree_key; | |
ba395927 KA |
1148 | |
1149 | static void dmar_init_reserved_ranges(void) | |
1150 | { | |
1151 | struct pci_dev *pdev = NULL; | |
1152 | struct iova *iova; | |
1153 | int i; | |
1154 | u64 addr, size; | |
1155 | ||
f661197e | 1156 | init_iova_domain(&reserved_iova_list, DMA_32BIT_PFN); |
ba395927 | 1157 | |
8a443df4 MG |
1158 | lockdep_set_class(&reserved_iova_list.iova_alloc_lock, |
1159 | &reserved_alloc_key); | |
1160 | lockdep_set_class(&reserved_iova_list.iova_rbtree_lock, | |
1161 | &reserved_rbtree_key); | |
1162 | ||
ba395927 KA |
1163 | /* IOAPIC ranges shouldn't be accessed by DMA */ |
1164 | iova = reserve_iova(&reserved_iova_list, IOVA_PFN(IOAPIC_RANGE_START), | |
1165 | IOVA_PFN(IOAPIC_RANGE_END)); | |
1166 | if (!iova) | |
1167 | printk(KERN_ERR "Reserve IOAPIC range failed\n"); | |
1168 | ||
1169 | /* Reserve all PCI MMIO to avoid peer-to-peer access */ | |
1170 | for_each_pci_dev(pdev) { | |
1171 | struct resource *r; | |
1172 | ||
1173 | for (i = 0; i < PCI_NUM_RESOURCES; i++) { | |
1174 | r = &pdev->resource[i]; | |
1175 | if (!r->flags || !(r->flags & IORESOURCE_MEM)) | |
1176 | continue; | |
1177 | addr = r->start; | |
5b6985ce | 1178 | addr &= PAGE_MASK; |
ba395927 | 1179 | size = r->end - addr; |
5b6985ce | 1180 | size = PAGE_ALIGN(size); |
ba395927 KA |
1181 | iova = reserve_iova(&reserved_iova_list, IOVA_PFN(addr), |
1182 | IOVA_PFN(size + addr) - 1); | |
1183 | if (!iova) | |
1184 | printk(KERN_ERR "Reserve iova failed\n"); | |
1185 | } | |
1186 | } | |
1187 | ||
1188 | } | |
1189 | ||
1190 | static void domain_reserve_special_ranges(struct dmar_domain *domain) | |
1191 | { | |
1192 | copy_reserved_iova(&reserved_iova_list, &domain->iovad); | |
1193 | } | |
1194 | ||
1195 | static inline int guestwidth_to_adjustwidth(int gaw) | |
1196 | { | |
1197 | int agaw; | |
1198 | int r = (gaw - 12) % 9; | |
1199 | ||
1200 | if (r == 0) | |
1201 | agaw = gaw; | |
1202 | else | |
1203 | agaw = gaw + 9 - r; | |
1204 | if (agaw > 64) | |
1205 | agaw = 64; | |
1206 | return agaw; | |
1207 | } | |
1208 | ||
1209 | static int domain_init(struct dmar_domain *domain, int guest_width) | |
1210 | { | |
1211 | struct intel_iommu *iommu; | |
1212 | int adjust_width, agaw; | |
1213 | unsigned long sagaw; | |
1214 | ||
f661197e | 1215 | init_iova_domain(&domain->iovad, DMA_32BIT_PFN); |
ba395927 | 1216 | spin_lock_init(&domain->mapping_lock); |
c7151a8d | 1217 | spin_lock_init(&domain->iommu_lock); |
ba395927 KA |
1218 | |
1219 | domain_reserve_special_ranges(domain); | |
1220 | ||
1221 | /* calculate AGAW */ | |
8c11e798 | 1222 | iommu = domain_get_iommu(domain); |
ba395927 KA |
1223 | if (guest_width > cap_mgaw(iommu->cap)) |
1224 | guest_width = cap_mgaw(iommu->cap); | |
1225 | domain->gaw = guest_width; | |
1226 | adjust_width = guestwidth_to_adjustwidth(guest_width); | |
1227 | agaw = width_to_agaw(adjust_width); | |
1228 | sagaw = cap_sagaw(iommu->cap); | |
1229 | if (!test_bit(agaw, &sagaw)) { | |
1230 | /* hardware doesn't support it, choose a bigger one */ | |
1231 | pr_debug("IOMMU: hardware doesn't support agaw %d\n", agaw); | |
1232 | agaw = find_next_bit(&sagaw, 5, agaw); | |
1233 | if (agaw >= 5) | |
1234 | return -ENODEV; | |
1235 | } | |
1236 | domain->agaw = agaw; | |
1237 | INIT_LIST_HEAD(&domain->devices); | |
1238 | ||
8e604097 WH |
1239 | if (ecap_coherent(iommu->ecap)) |
1240 | domain->iommu_coherency = 1; | |
1241 | else | |
1242 | domain->iommu_coherency = 0; | |
1243 | ||
c7151a8d WH |
1244 | domain->iommu_count = 1; |
1245 | ||
ba395927 KA |
1246 | /* always allocate the top pgd */ |
1247 | domain->pgd = (struct dma_pte *)alloc_pgtable_page(); | |
1248 | if (!domain->pgd) | |
1249 | return -ENOMEM; | |
5b6985ce | 1250 | __iommu_flush_cache(iommu, domain->pgd, PAGE_SIZE); |
ba395927 KA |
1251 | return 0; |
1252 | } | |
1253 | ||
1254 | static void domain_exit(struct dmar_domain *domain) | |
1255 | { | |
1256 | u64 end; | |
1257 | ||
1258 | /* Domain 0 is reserved, so dont process it */ | |
1259 | if (!domain) | |
1260 | return; | |
1261 | ||
1262 | domain_remove_dev_info(domain); | |
1263 | /* destroy iovas */ | |
1264 | put_iova_domain(&domain->iovad); | |
1265 | end = DOMAIN_MAX_ADDR(domain->gaw); | |
5b6985ce | 1266 | end = end & (~PAGE_MASK); |
ba395927 KA |
1267 | |
1268 | /* clear ptes */ | |
1269 | dma_pte_clear_range(domain, 0, end); | |
1270 | ||
1271 | /* free page tables */ | |
1272 | dma_pte_free_pagetable(domain, 0, end); | |
1273 | ||
1274 | iommu_free_domain(domain); | |
1275 | free_domain_mem(domain); | |
1276 | } | |
1277 | ||
1278 | static int domain_context_mapping_one(struct dmar_domain *domain, | |
1279 | u8 bus, u8 devfn) | |
1280 | { | |
1281 | struct context_entry *context; | |
ba395927 | 1282 | unsigned long flags; |
5331fe6f | 1283 | struct intel_iommu *iommu; |
ea6606b0 WH |
1284 | struct dma_pte *pgd; |
1285 | unsigned long num; | |
1286 | unsigned long ndomains; | |
1287 | int id; | |
1288 | int agaw; | |
ba395927 KA |
1289 | |
1290 | pr_debug("Set context mapping for %02x:%02x.%d\n", | |
1291 | bus, PCI_SLOT(devfn), PCI_FUNC(devfn)); | |
1292 | BUG_ON(!domain->pgd); | |
5331fe6f WH |
1293 | |
1294 | iommu = device_to_iommu(bus, devfn); | |
1295 | if (!iommu) | |
1296 | return -ENODEV; | |
1297 | ||
ba395927 KA |
1298 | context = device_to_context_entry(iommu, bus, devfn); |
1299 | if (!context) | |
1300 | return -ENOMEM; | |
1301 | spin_lock_irqsave(&iommu->lock, flags); | |
c07e7d21 | 1302 | if (context_present(context)) { |
ba395927 KA |
1303 | spin_unlock_irqrestore(&iommu->lock, flags); |
1304 | return 0; | |
1305 | } | |
1306 | ||
ea6606b0 WH |
1307 | id = domain->id; |
1308 | pgd = domain->pgd; | |
1309 | ||
1310 | if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) { | |
1311 | int found = 0; | |
1312 | ||
1313 | /* find an available domain id for this device in iommu */ | |
1314 | ndomains = cap_ndoms(iommu->cap); | |
1315 | num = find_first_bit(iommu->domain_ids, ndomains); | |
1316 | for (; num < ndomains; ) { | |
1317 | if (iommu->domains[num] == domain) { | |
1318 | id = num; | |
1319 | found = 1; | |
1320 | break; | |
1321 | } | |
1322 | num = find_next_bit(iommu->domain_ids, | |
1323 | cap_ndoms(iommu->cap), num+1); | |
1324 | } | |
1325 | ||
1326 | if (found == 0) { | |
1327 | num = find_first_zero_bit(iommu->domain_ids, ndomains); | |
1328 | if (num >= ndomains) { | |
1329 | spin_unlock_irqrestore(&iommu->lock, flags); | |
1330 | printk(KERN_ERR "IOMMU: no free domain ids\n"); | |
1331 | return -EFAULT; | |
1332 | } | |
1333 | ||
1334 | set_bit(num, iommu->domain_ids); | |
1335 | iommu->domains[num] = domain; | |
1336 | id = num; | |
1337 | } | |
1338 | ||
1339 | /* Skip top levels of page tables for | |
1340 | * iommu which has less agaw than default. | |
1341 | */ | |
1342 | for (agaw = domain->agaw; agaw != iommu->agaw; agaw--) { | |
1343 | pgd = phys_to_virt(dma_pte_addr(pgd)); | |
1344 | if (!dma_pte_present(pgd)) { | |
1345 | spin_unlock_irqrestore(&iommu->lock, flags); | |
1346 | return -ENOMEM; | |
1347 | } | |
1348 | } | |
1349 | } | |
1350 | ||
1351 | context_set_domain_id(context, id); | |
1352 | context_set_address_width(context, iommu->agaw); | |
1353 | context_set_address_root(context, virt_to_phys(pgd)); | |
c07e7d21 MM |
1354 | context_set_translation_type(context, CONTEXT_TT_MULTI_LEVEL); |
1355 | context_set_fault_enable(context); | |
1356 | context_set_present(context); | |
5331fe6f | 1357 | domain_flush_cache(domain, context, sizeof(*context)); |
ba395927 KA |
1358 | |
1359 | /* it's a non-present to present mapping */ | |
a77b67d4 YS |
1360 | if (iommu->flush.flush_context(iommu, domain->id, |
1361 | (((u16)bus) << 8) | devfn, DMA_CCMD_MASK_NOBIT, | |
1362 | DMA_CCMD_DEVICE_INVL, 1)) | |
ba395927 KA |
1363 | iommu_flush_write_buffer(iommu); |
1364 | else | |
a77b67d4 YS |
1365 | iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_DSI_FLUSH, 0); |
1366 | ||
ba395927 | 1367 | spin_unlock_irqrestore(&iommu->lock, flags); |
c7151a8d WH |
1368 | |
1369 | spin_lock_irqsave(&domain->iommu_lock, flags); | |
1370 | if (!test_and_set_bit(iommu->seq_id, &domain->iommu_bmp)) { | |
1371 | domain->iommu_count++; | |
1372 | domain_update_iommu_coherency(domain); | |
1373 | } | |
1374 | spin_unlock_irqrestore(&domain->iommu_lock, flags); | |
ba395927 KA |
1375 | return 0; |
1376 | } | |
1377 | ||
1378 | static int | |
1379 | domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev) | |
1380 | { | |
1381 | int ret; | |
1382 | struct pci_dev *tmp, *parent; | |
1383 | ||
1384 | ret = domain_context_mapping_one(domain, pdev->bus->number, | |
1385 | pdev->devfn); | |
1386 | if (ret) | |
1387 | return ret; | |
1388 | ||
1389 | /* dependent device mapping */ | |
1390 | tmp = pci_find_upstream_pcie_bridge(pdev); | |
1391 | if (!tmp) | |
1392 | return 0; | |
1393 | /* Secondary interface's bus number and devfn 0 */ | |
1394 | parent = pdev->bus->self; | |
1395 | while (parent != tmp) { | |
1396 | ret = domain_context_mapping_one(domain, parent->bus->number, | |
1397 | parent->devfn); | |
1398 | if (ret) | |
1399 | return ret; | |
1400 | parent = parent->bus->self; | |
1401 | } | |
1402 | if (tmp->is_pcie) /* this is a PCIE-to-PCI bridge */ | |
1403 | return domain_context_mapping_one(domain, | |
1404 | tmp->subordinate->number, 0); | |
1405 | else /* this is a legacy PCI bridge */ | |
1406 | return domain_context_mapping_one(domain, | |
1407 | tmp->bus->number, tmp->devfn); | |
1408 | } | |
1409 | ||
5331fe6f | 1410 | static int domain_context_mapped(struct pci_dev *pdev) |
ba395927 KA |
1411 | { |
1412 | int ret; | |
1413 | struct pci_dev *tmp, *parent; | |
5331fe6f WH |
1414 | struct intel_iommu *iommu; |
1415 | ||
1416 | iommu = device_to_iommu(pdev->bus->number, pdev->devfn); | |
1417 | if (!iommu) | |
1418 | return -ENODEV; | |
ba395927 | 1419 | |
8c11e798 | 1420 | ret = device_context_mapped(iommu, |
ba395927 KA |
1421 | pdev->bus->number, pdev->devfn); |
1422 | if (!ret) | |
1423 | return ret; | |
1424 | /* dependent device mapping */ | |
1425 | tmp = pci_find_upstream_pcie_bridge(pdev); | |
1426 | if (!tmp) | |
1427 | return ret; | |
1428 | /* Secondary interface's bus number and devfn 0 */ | |
1429 | parent = pdev->bus->self; | |
1430 | while (parent != tmp) { | |
8c11e798 | 1431 | ret = device_context_mapped(iommu, parent->bus->number, |
ba395927 KA |
1432 | parent->devfn); |
1433 | if (!ret) | |
1434 | return ret; | |
1435 | parent = parent->bus->self; | |
1436 | } | |
1437 | if (tmp->is_pcie) | |
8c11e798 | 1438 | return device_context_mapped(iommu, |
ba395927 KA |
1439 | tmp->subordinate->number, 0); |
1440 | else | |
8c11e798 | 1441 | return device_context_mapped(iommu, |
ba395927 KA |
1442 | tmp->bus->number, tmp->devfn); |
1443 | } | |
1444 | ||
1445 | static int | |
1446 | domain_page_mapping(struct dmar_domain *domain, dma_addr_t iova, | |
1447 | u64 hpa, size_t size, int prot) | |
1448 | { | |
1449 | u64 start_pfn, end_pfn; | |
1450 | struct dma_pte *pte; | |
1451 | int index; | |
5b6985ce FY |
1452 | int addr_width = agaw_to_width(domain->agaw); |
1453 | ||
1454 | hpa &= (((u64)1) << addr_width) - 1; | |
ba395927 KA |
1455 | |
1456 | if ((prot & (DMA_PTE_READ|DMA_PTE_WRITE)) == 0) | |
1457 | return -EINVAL; | |
5b6985ce FY |
1458 | iova &= PAGE_MASK; |
1459 | start_pfn = ((u64)hpa) >> VTD_PAGE_SHIFT; | |
1460 | end_pfn = (VTD_PAGE_ALIGN(((u64)hpa) + size)) >> VTD_PAGE_SHIFT; | |
ba395927 KA |
1461 | index = 0; |
1462 | while (start_pfn < end_pfn) { | |
5b6985ce | 1463 | pte = addr_to_dma_pte(domain, iova + VTD_PAGE_SIZE * index); |
ba395927 KA |
1464 | if (!pte) |
1465 | return -ENOMEM; | |
1466 | /* We don't need lock here, nobody else | |
1467 | * touches the iova range | |
1468 | */ | |
19c239ce MM |
1469 | BUG_ON(dma_pte_addr(pte)); |
1470 | dma_set_pte_addr(pte, start_pfn << VTD_PAGE_SHIFT); | |
1471 | dma_set_pte_prot(pte, prot); | |
5331fe6f | 1472 | domain_flush_cache(domain, pte, sizeof(*pte)); |
ba395927 KA |
1473 | start_pfn++; |
1474 | index++; | |
1475 | } | |
1476 | return 0; | |
1477 | } | |
1478 | ||
c7151a8d | 1479 | static void iommu_detach_dev(struct intel_iommu *iommu, u8 bus, u8 devfn) |
ba395927 | 1480 | { |
c7151a8d WH |
1481 | if (!iommu) |
1482 | return; | |
8c11e798 WH |
1483 | |
1484 | clear_context_table(iommu, bus, devfn); | |
1485 | iommu->flush.flush_context(iommu, 0, 0, 0, | |
a77b67d4 | 1486 | DMA_CCMD_GLOBAL_INVL, 0); |
8c11e798 | 1487 | iommu->flush.flush_iotlb(iommu, 0, 0, 0, |
a77b67d4 | 1488 | DMA_TLB_GLOBAL_FLUSH, 0); |
ba395927 KA |
1489 | } |
1490 | ||
1491 | static void domain_remove_dev_info(struct dmar_domain *domain) | |
1492 | { | |
1493 | struct device_domain_info *info; | |
1494 | unsigned long flags; | |
c7151a8d | 1495 | struct intel_iommu *iommu; |
ba395927 KA |
1496 | |
1497 | spin_lock_irqsave(&device_domain_lock, flags); | |
1498 | while (!list_empty(&domain->devices)) { | |
1499 | info = list_entry(domain->devices.next, | |
1500 | struct device_domain_info, link); | |
1501 | list_del(&info->link); | |
1502 | list_del(&info->global); | |
1503 | if (info->dev) | |
358dd8ac | 1504 | info->dev->dev.archdata.iommu = NULL; |
ba395927 KA |
1505 | spin_unlock_irqrestore(&device_domain_lock, flags); |
1506 | ||
c7151a8d WH |
1507 | iommu = device_to_iommu(info->bus, info->devfn); |
1508 | iommu_detach_dev(iommu, info->bus, info->devfn); | |
ba395927 KA |
1509 | free_devinfo_mem(info); |
1510 | ||
1511 | spin_lock_irqsave(&device_domain_lock, flags); | |
1512 | } | |
1513 | spin_unlock_irqrestore(&device_domain_lock, flags); | |
1514 | } | |
1515 | ||
1516 | /* | |
1517 | * find_domain | |
358dd8ac | 1518 | * Note: we use struct pci_dev->dev.archdata.iommu stores the info |
ba395927 | 1519 | */ |
38717946 | 1520 | static struct dmar_domain * |
ba395927 KA |
1521 | find_domain(struct pci_dev *pdev) |
1522 | { | |
1523 | struct device_domain_info *info; | |
1524 | ||
1525 | /* No lock here, assumes no domain exit in normal case */ | |
358dd8ac | 1526 | info = pdev->dev.archdata.iommu; |
ba395927 KA |
1527 | if (info) |
1528 | return info->domain; | |
1529 | return NULL; | |
1530 | } | |
1531 | ||
ba395927 KA |
1532 | /* domain is initialized */ |
1533 | static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw) | |
1534 | { | |
1535 | struct dmar_domain *domain, *found = NULL; | |
1536 | struct intel_iommu *iommu; | |
1537 | struct dmar_drhd_unit *drhd; | |
1538 | struct device_domain_info *info, *tmp; | |
1539 | struct pci_dev *dev_tmp; | |
1540 | unsigned long flags; | |
1541 | int bus = 0, devfn = 0; | |
1542 | ||
1543 | domain = find_domain(pdev); | |
1544 | if (domain) | |
1545 | return domain; | |
1546 | ||
1547 | dev_tmp = pci_find_upstream_pcie_bridge(pdev); | |
1548 | if (dev_tmp) { | |
1549 | if (dev_tmp->is_pcie) { | |
1550 | bus = dev_tmp->subordinate->number; | |
1551 | devfn = 0; | |
1552 | } else { | |
1553 | bus = dev_tmp->bus->number; | |
1554 | devfn = dev_tmp->devfn; | |
1555 | } | |
1556 | spin_lock_irqsave(&device_domain_lock, flags); | |
1557 | list_for_each_entry(info, &device_domain_list, global) { | |
1558 | if (info->bus == bus && info->devfn == devfn) { | |
1559 | found = info->domain; | |
1560 | break; | |
1561 | } | |
1562 | } | |
1563 | spin_unlock_irqrestore(&device_domain_lock, flags); | |
1564 | /* pcie-pci bridge already has a domain, uses it */ | |
1565 | if (found) { | |
1566 | domain = found; | |
1567 | goto found_domain; | |
1568 | } | |
1569 | } | |
1570 | ||
1571 | /* Allocate new domain for the device */ | |
1572 | drhd = dmar_find_matched_drhd_unit(pdev); | |
1573 | if (!drhd) { | |
1574 | printk(KERN_ERR "IOMMU: can't find DMAR for device %s\n", | |
1575 | pci_name(pdev)); | |
1576 | return NULL; | |
1577 | } | |
1578 | iommu = drhd->iommu; | |
1579 | ||
1580 | domain = iommu_alloc_domain(iommu); | |
1581 | if (!domain) | |
1582 | goto error; | |
1583 | ||
1584 | if (domain_init(domain, gaw)) { | |
1585 | domain_exit(domain); | |
1586 | goto error; | |
1587 | } | |
1588 | ||
1589 | /* register pcie-to-pci device */ | |
1590 | if (dev_tmp) { | |
1591 | info = alloc_devinfo_mem(); | |
1592 | if (!info) { | |
1593 | domain_exit(domain); | |
1594 | goto error; | |
1595 | } | |
1596 | info->bus = bus; | |
1597 | info->devfn = devfn; | |
1598 | info->dev = NULL; | |
1599 | info->domain = domain; | |
1600 | /* This domain is shared by devices under p2p bridge */ | |
3b5410e7 | 1601 | domain->flags |= DOMAIN_FLAG_P2P_MULTIPLE_DEVICES; |
ba395927 KA |
1602 | |
1603 | /* pcie-to-pci bridge already has a domain, uses it */ | |
1604 | found = NULL; | |
1605 | spin_lock_irqsave(&device_domain_lock, flags); | |
1606 | list_for_each_entry(tmp, &device_domain_list, global) { | |
1607 | if (tmp->bus == bus && tmp->devfn == devfn) { | |
1608 | found = tmp->domain; | |
1609 | break; | |
1610 | } | |
1611 | } | |
1612 | if (found) { | |
1613 | free_devinfo_mem(info); | |
1614 | domain_exit(domain); | |
1615 | domain = found; | |
1616 | } else { | |
1617 | list_add(&info->link, &domain->devices); | |
1618 | list_add(&info->global, &device_domain_list); | |
1619 | } | |
1620 | spin_unlock_irqrestore(&device_domain_lock, flags); | |
1621 | } | |
1622 | ||
1623 | found_domain: | |
1624 | info = alloc_devinfo_mem(); | |
1625 | if (!info) | |
1626 | goto error; | |
1627 | info->bus = pdev->bus->number; | |
1628 | info->devfn = pdev->devfn; | |
1629 | info->dev = pdev; | |
1630 | info->domain = domain; | |
1631 | spin_lock_irqsave(&device_domain_lock, flags); | |
1632 | /* somebody is fast */ | |
1633 | found = find_domain(pdev); | |
1634 | if (found != NULL) { | |
1635 | spin_unlock_irqrestore(&device_domain_lock, flags); | |
1636 | if (found != domain) { | |
1637 | domain_exit(domain); | |
1638 | domain = found; | |
1639 | } | |
1640 | free_devinfo_mem(info); | |
1641 | return domain; | |
1642 | } | |
1643 | list_add(&info->link, &domain->devices); | |
1644 | list_add(&info->global, &device_domain_list); | |
358dd8ac | 1645 | pdev->dev.archdata.iommu = info; |
ba395927 KA |
1646 | spin_unlock_irqrestore(&device_domain_lock, flags); |
1647 | return domain; | |
1648 | error: | |
1649 | /* recheck it here, maybe others set it */ | |
1650 | return find_domain(pdev); | |
1651 | } | |
1652 | ||
5b6985ce FY |
1653 | static int iommu_prepare_identity_map(struct pci_dev *pdev, |
1654 | unsigned long long start, | |
1655 | unsigned long long end) | |
ba395927 KA |
1656 | { |
1657 | struct dmar_domain *domain; | |
1658 | unsigned long size; | |
5b6985ce | 1659 | unsigned long long base; |
ba395927 KA |
1660 | int ret; |
1661 | ||
1662 | printk(KERN_INFO | |
1663 | "IOMMU: Setting identity map for device %s [0x%Lx - 0x%Lx]\n", | |
1664 | pci_name(pdev), start, end); | |
1665 | /* page table init */ | |
1666 | domain = get_domain_for_dev(pdev, DEFAULT_DOMAIN_ADDRESS_WIDTH); | |
1667 | if (!domain) | |
1668 | return -ENOMEM; | |
1669 | ||
1670 | /* The address might not be aligned */ | |
5b6985ce | 1671 | base = start & PAGE_MASK; |
ba395927 | 1672 | size = end - base; |
5b6985ce | 1673 | size = PAGE_ALIGN(size); |
ba395927 KA |
1674 | if (!reserve_iova(&domain->iovad, IOVA_PFN(base), |
1675 | IOVA_PFN(base + size) - 1)) { | |
1676 | printk(KERN_ERR "IOMMU: reserve iova failed\n"); | |
1677 | ret = -ENOMEM; | |
1678 | goto error; | |
1679 | } | |
1680 | ||
1681 | pr_debug("Mapping reserved region %lx@%llx for %s\n", | |
1682 | size, base, pci_name(pdev)); | |
1683 | /* | |
1684 | * RMRR range might have overlap with physical memory range, | |
1685 | * clear it first | |
1686 | */ | |
1687 | dma_pte_clear_range(domain, base, base + size); | |
1688 | ||
1689 | ret = domain_page_mapping(domain, base, base, size, | |
1690 | DMA_PTE_READ|DMA_PTE_WRITE); | |
1691 | if (ret) | |
1692 | goto error; | |
1693 | ||
1694 | /* context entry init */ | |
1695 | ret = domain_context_mapping(domain, pdev); | |
1696 | if (!ret) | |
1697 | return 0; | |
1698 | error: | |
1699 | domain_exit(domain); | |
1700 | return ret; | |
1701 | ||
1702 | } | |
1703 | ||
1704 | static inline int iommu_prepare_rmrr_dev(struct dmar_rmrr_unit *rmrr, | |
1705 | struct pci_dev *pdev) | |
1706 | { | |
358dd8ac | 1707 | if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO) |
ba395927 KA |
1708 | return 0; |
1709 | return iommu_prepare_identity_map(pdev, rmrr->base_address, | |
1710 | rmrr->end_address + 1); | |
1711 | } | |
1712 | ||
e820482c | 1713 | #ifdef CONFIG_DMAR_GFX_WA |
d52d53b8 YL |
1714 | struct iommu_prepare_data { |
1715 | struct pci_dev *pdev; | |
1716 | int ret; | |
1717 | }; | |
1718 | ||
1719 | static int __init iommu_prepare_work_fn(unsigned long start_pfn, | |
1720 | unsigned long end_pfn, void *datax) | |
1721 | { | |
1722 | struct iommu_prepare_data *data; | |
1723 | ||
1724 | data = (struct iommu_prepare_data *)datax; | |
1725 | ||
1726 | data->ret = iommu_prepare_identity_map(data->pdev, | |
1727 | start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT); | |
1728 | return data->ret; | |
1729 | ||
1730 | } | |
1731 | ||
1732 | static int __init iommu_prepare_with_active_regions(struct pci_dev *pdev) | |
1733 | { | |
1734 | int nid; | |
1735 | struct iommu_prepare_data data; | |
1736 | ||
1737 | data.pdev = pdev; | |
1738 | data.ret = 0; | |
1739 | ||
1740 | for_each_online_node(nid) { | |
1741 | work_with_active_regions(nid, iommu_prepare_work_fn, &data); | |
1742 | if (data.ret) | |
1743 | return data.ret; | |
1744 | } | |
1745 | return data.ret; | |
1746 | } | |
1747 | ||
e820482c KA |
1748 | static void __init iommu_prepare_gfx_mapping(void) |
1749 | { | |
1750 | struct pci_dev *pdev = NULL; | |
e820482c KA |
1751 | int ret; |
1752 | ||
1753 | for_each_pci_dev(pdev) { | |
358dd8ac | 1754 | if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO || |
e820482c KA |
1755 | !IS_GFX_DEVICE(pdev)) |
1756 | continue; | |
1757 | printk(KERN_INFO "IOMMU: gfx device %s 1-1 mapping\n", | |
1758 | pci_name(pdev)); | |
d52d53b8 YL |
1759 | ret = iommu_prepare_with_active_regions(pdev); |
1760 | if (ret) | |
1761 | printk(KERN_ERR "IOMMU: mapping reserved region failed\n"); | |
e820482c KA |
1762 | } |
1763 | } | |
2abd7e16 MM |
1764 | #else /* !CONFIG_DMAR_GFX_WA */ |
1765 | static inline void iommu_prepare_gfx_mapping(void) | |
1766 | { | |
1767 | return; | |
1768 | } | |
e820482c KA |
1769 | #endif |
1770 | ||
49a0429e KA |
1771 | #ifdef CONFIG_DMAR_FLOPPY_WA |
1772 | static inline void iommu_prepare_isa(void) | |
1773 | { | |
1774 | struct pci_dev *pdev; | |
1775 | int ret; | |
1776 | ||
1777 | pdev = pci_get_class(PCI_CLASS_BRIDGE_ISA << 8, NULL); | |
1778 | if (!pdev) | |
1779 | return; | |
1780 | ||
1781 | printk(KERN_INFO "IOMMU: Prepare 0-16M unity mapping for LPC\n"); | |
1782 | ret = iommu_prepare_identity_map(pdev, 0, 16*1024*1024); | |
1783 | ||
1784 | if (ret) | |
1c35b8e5 | 1785 | printk(KERN_ERR "IOMMU: Failed to create 0-64M identity map, " |
49a0429e KA |
1786 | "floppy might not work\n"); |
1787 | ||
1788 | } | |
1789 | #else | |
1790 | static inline void iommu_prepare_isa(void) | |
1791 | { | |
1792 | return; | |
1793 | } | |
1794 | #endif /* !CONFIG_DMAR_FLPY_WA */ | |
1795 | ||
519a0549 | 1796 | static int __init init_dmars(void) |
ba395927 KA |
1797 | { |
1798 | struct dmar_drhd_unit *drhd; | |
1799 | struct dmar_rmrr_unit *rmrr; | |
1800 | struct pci_dev *pdev; | |
1801 | struct intel_iommu *iommu; | |
9d783ba0 | 1802 | int i, ret; |
ba395927 KA |
1803 | |
1804 | /* | |
1805 | * for each drhd | |
1806 | * allocate root | |
1807 | * initialize and program root entry to not present | |
1808 | * endfor | |
1809 | */ | |
1810 | for_each_drhd_unit(drhd) { | |
5e0d2a6f | 1811 | g_num_of_iommus++; |
1812 | /* | |
1813 | * lock not needed as this is only incremented in the single | |
1814 | * threaded kernel __init code path all other access are read | |
1815 | * only | |
1816 | */ | |
1817 | } | |
1818 | ||
d9630fe9 WH |
1819 | g_iommus = kcalloc(g_num_of_iommus, sizeof(struct intel_iommu *), |
1820 | GFP_KERNEL); | |
1821 | if (!g_iommus) { | |
1822 | printk(KERN_ERR "Allocating global iommu array failed\n"); | |
1823 | ret = -ENOMEM; | |
1824 | goto error; | |
1825 | } | |
1826 | ||
80b20dd8 | 1827 | deferred_flush = kzalloc(g_num_of_iommus * |
1828 | sizeof(struct deferred_flush_tables), GFP_KERNEL); | |
1829 | if (!deferred_flush) { | |
d9630fe9 | 1830 | kfree(g_iommus); |
5e0d2a6f | 1831 | ret = -ENOMEM; |
1832 | goto error; | |
1833 | } | |
1834 | ||
5e0d2a6f | 1835 | for_each_drhd_unit(drhd) { |
1836 | if (drhd->ignored) | |
1837 | continue; | |
1886e8a9 SS |
1838 | |
1839 | iommu = drhd->iommu; | |
d9630fe9 | 1840 | g_iommus[iommu->seq_id] = iommu; |
ba395927 | 1841 | |
e61d98d8 SS |
1842 | ret = iommu_init_domains(iommu); |
1843 | if (ret) | |
1844 | goto error; | |
1845 | ||
ba395927 KA |
1846 | /* |
1847 | * TBD: | |
1848 | * we could share the same root & context tables | |
1849 | * amoung all IOMMU's. Need to Split it later. | |
1850 | */ | |
1851 | ret = iommu_alloc_root_entry(iommu); | |
1852 | if (ret) { | |
1853 | printk(KERN_ERR "IOMMU: allocate root entry failed\n"); | |
1854 | goto error; | |
1855 | } | |
1856 | } | |
1857 | ||
1531a6a6 SS |
1858 | /* |
1859 | * Start from the sane iommu hardware state. | |
1860 | */ | |
a77b67d4 YS |
1861 | for_each_drhd_unit(drhd) { |
1862 | if (drhd->ignored) | |
1863 | continue; | |
1864 | ||
1865 | iommu = drhd->iommu; | |
1531a6a6 SS |
1866 | |
1867 | /* | |
1868 | * If the queued invalidation is already initialized by us | |
1869 | * (for example, while enabling interrupt-remapping) then | |
1870 | * we got the things already rolling from a sane state. | |
1871 | */ | |
1872 | if (iommu->qi) | |
1873 | continue; | |
1874 | ||
1875 | /* | |
1876 | * Clear any previous faults. | |
1877 | */ | |
1878 | dmar_fault(-1, iommu); | |
1879 | /* | |
1880 | * Disable queued invalidation if supported and already enabled | |
1881 | * before OS handover. | |
1882 | */ | |
1883 | dmar_disable_qi(iommu); | |
1884 | } | |
1885 | ||
1886 | for_each_drhd_unit(drhd) { | |
1887 | if (drhd->ignored) | |
1888 | continue; | |
1889 | ||
1890 | iommu = drhd->iommu; | |
1891 | ||
a77b67d4 YS |
1892 | if (dmar_enable_qi(iommu)) { |
1893 | /* | |
1894 | * Queued Invalidate not enabled, use Register Based | |
1895 | * Invalidate | |
1896 | */ | |
1897 | iommu->flush.flush_context = __iommu_flush_context; | |
1898 | iommu->flush.flush_iotlb = __iommu_flush_iotlb; | |
1899 | printk(KERN_INFO "IOMMU 0x%Lx: using Register based " | |
b4e0f9eb FT |
1900 | "invalidation\n", |
1901 | (unsigned long long)drhd->reg_base_addr); | |
a77b67d4 YS |
1902 | } else { |
1903 | iommu->flush.flush_context = qi_flush_context; | |
1904 | iommu->flush.flush_iotlb = qi_flush_iotlb; | |
1905 | printk(KERN_INFO "IOMMU 0x%Lx: using Queued " | |
b4e0f9eb FT |
1906 | "invalidation\n", |
1907 | (unsigned long long)drhd->reg_base_addr); | |
a77b67d4 YS |
1908 | } |
1909 | } | |
1910 | ||
ba395927 KA |
1911 | /* |
1912 | * For each rmrr | |
1913 | * for each dev attached to rmrr | |
1914 | * do | |
1915 | * locate drhd for dev, alloc domain for dev | |
1916 | * allocate free domain | |
1917 | * allocate page table entries for rmrr | |
1918 | * if context not allocated for bus | |
1919 | * allocate and init context | |
1920 | * set present in root table for this bus | |
1921 | * init context with domain, translation etc | |
1922 | * endfor | |
1923 | * endfor | |
1924 | */ | |
1925 | for_each_rmrr_units(rmrr) { | |
ba395927 KA |
1926 | for (i = 0; i < rmrr->devices_cnt; i++) { |
1927 | pdev = rmrr->devices[i]; | |
1928 | /* some BIOS lists non-exist devices in DMAR table */ | |
1929 | if (!pdev) | |
1930 | continue; | |
1931 | ret = iommu_prepare_rmrr_dev(rmrr, pdev); | |
1932 | if (ret) | |
1933 | printk(KERN_ERR | |
1934 | "IOMMU: mapping reserved region failed\n"); | |
1935 | } | |
1936 | } | |
1937 | ||
e820482c KA |
1938 | iommu_prepare_gfx_mapping(); |
1939 | ||
49a0429e KA |
1940 | iommu_prepare_isa(); |
1941 | ||
ba395927 KA |
1942 | /* |
1943 | * for each drhd | |
1944 | * enable fault log | |
1945 | * global invalidate context cache | |
1946 | * global invalidate iotlb | |
1947 | * enable translation | |
1948 | */ | |
1949 | for_each_drhd_unit(drhd) { | |
1950 | if (drhd->ignored) | |
1951 | continue; | |
1952 | iommu = drhd->iommu; | |
ba395927 KA |
1953 | |
1954 | iommu_flush_write_buffer(iommu); | |
1955 | ||
3460a6d9 KA |
1956 | ret = dmar_set_interrupt(iommu); |
1957 | if (ret) | |
1958 | goto error; | |
1959 | ||
ba395927 KA |
1960 | iommu_set_root_entry(iommu); |
1961 | ||
a77b67d4 YS |
1962 | iommu->flush.flush_context(iommu, 0, 0, 0, DMA_CCMD_GLOBAL_INVL, |
1963 | 0); | |
1964 | iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH, | |
1965 | 0); | |
f8bab735 | 1966 | iommu_disable_protect_mem_regions(iommu); |
1967 | ||
ba395927 KA |
1968 | ret = iommu_enable_translation(iommu); |
1969 | if (ret) | |
1970 | goto error; | |
1971 | } | |
1972 | ||
1973 | return 0; | |
1974 | error: | |
1975 | for_each_drhd_unit(drhd) { | |
1976 | if (drhd->ignored) | |
1977 | continue; | |
1978 | iommu = drhd->iommu; | |
1979 | free_iommu(iommu); | |
1980 | } | |
d9630fe9 | 1981 | kfree(g_iommus); |
ba395927 KA |
1982 | return ret; |
1983 | } | |
1984 | ||
1985 | static inline u64 aligned_size(u64 host_addr, size_t size) | |
1986 | { | |
1987 | u64 addr; | |
5b6985ce FY |
1988 | addr = (host_addr & (~PAGE_MASK)) + size; |
1989 | return PAGE_ALIGN(addr); | |
ba395927 KA |
1990 | } |
1991 | ||
1992 | struct iova * | |
f76aec76 | 1993 | iommu_alloc_iova(struct dmar_domain *domain, size_t size, u64 end) |
ba395927 | 1994 | { |
ba395927 KA |
1995 | struct iova *piova; |
1996 | ||
1997 | /* Make sure it's in range */ | |
ba395927 | 1998 | end = min_t(u64, DOMAIN_MAX_ADDR(domain->gaw), end); |
f76aec76 | 1999 | if (!size || (IOVA_START_ADDR + size > end)) |
ba395927 KA |
2000 | return NULL; |
2001 | ||
2002 | piova = alloc_iova(&domain->iovad, | |
5b6985ce | 2003 | size >> PAGE_SHIFT, IOVA_PFN(end), 1); |
ba395927 KA |
2004 | return piova; |
2005 | } | |
2006 | ||
f76aec76 KA |
2007 | static struct iova * |
2008 | __intel_alloc_iova(struct device *dev, struct dmar_domain *domain, | |
bb9e6d65 | 2009 | size_t size, u64 dma_mask) |
ba395927 | 2010 | { |
ba395927 | 2011 | struct pci_dev *pdev = to_pci_dev(dev); |
ba395927 | 2012 | struct iova *iova = NULL; |
ba395927 | 2013 | |
bb9e6d65 FT |
2014 | if (dma_mask <= DMA_32BIT_MASK || dmar_forcedac) |
2015 | iova = iommu_alloc_iova(domain, size, dma_mask); | |
2016 | else { | |
ba395927 KA |
2017 | /* |
2018 | * First try to allocate an io virtual address in | |
2019 | * DMA_32BIT_MASK and if that fails then try allocating | |
3609801e | 2020 | * from higher range |
ba395927 | 2021 | */ |
f76aec76 | 2022 | iova = iommu_alloc_iova(domain, size, DMA_32BIT_MASK); |
ba395927 | 2023 | if (!iova) |
bb9e6d65 | 2024 | iova = iommu_alloc_iova(domain, size, dma_mask); |
ba395927 KA |
2025 | } |
2026 | ||
2027 | if (!iova) { | |
2028 | printk(KERN_ERR"Allocating iova for %s failed", pci_name(pdev)); | |
f76aec76 KA |
2029 | return NULL; |
2030 | } | |
2031 | ||
2032 | return iova; | |
2033 | } | |
2034 | ||
2035 | static struct dmar_domain * | |
2036 | get_valid_domain_for_dev(struct pci_dev *pdev) | |
2037 | { | |
2038 | struct dmar_domain *domain; | |
2039 | int ret; | |
2040 | ||
2041 | domain = get_domain_for_dev(pdev, | |
2042 | DEFAULT_DOMAIN_ADDRESS_WIDTH); | |
2043 | if (!domain) { | |
2044 | printk(KERN_ERR | |
2045 | "Allocating domain for %s failed", pci_name(pdev)); | |
4fe05bbc | 2046 | return NULL; |
ba395927 KA |
2047 | } |
2048 | ||
2049 | /* make sure context mapping is ok */ | |
5331fe6f | 2050 | if (unlikely(!domain_context_mapped(pdev))) { |
ba395927 | 2051 | ret = domain_context_mapping(domain, pdev); |
f76aec76 KA |
2052 | if (ret) { |
2053 | printk(KERN_ERR | |
2054 | "Domain context map for %s failed", | |
2055 | pci_name(pdev)); | |
4fe05bbc | 2056 | return NULL; |
f76aec76 | 2057 | } |
ba395927 KA |
2058 | } |
2059 | ||
f76aec76 KA |
2060 | return domain; |
2061 | } | |
2062 | ||
bb9e6d65 FT |
2063 | static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr, |
2064 | size_t size, int dir, u64 dma_mask) | |
f76aec76 KA |
2065 | { |
2066 | struct pci_dev *pdev = to_pci_dev(hwdev); | |
f76aec76 | 2067 | struct dmar_domain *domain; |
5b6985ce | 2068 | phys_addr_t start_paddr; |
f76aec76 KA |
2069 | struct iova *iova; |
2070 | int prot = 0; | |
6865f0d1 | 2071 | int ret; |
8c11e798 | 2072 | struct intel_iommu *iommu; |
f76aec76 KA |
2073 | |
2074 | BUG_ON(dir == DMA_NONE); | |
358dd8ac | 2075 | if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO) |
6865f0d1 | 2076 | return paddr; |
f76aec76 KA |
2077 | |
2078 | domain = get_valid_domain_for_dev(pdev); | |
2079 | if (!domain) | |
2080 | return 0; | |
2081 | ||
8c11e798 | 2082 | iommu = domain_get_iommu(domain); |
6865f0d1 | 2083 | size = aligned_size((u64)paddr, size); |
f76aec76 | 2084 | |
bb9e6d65 | 2085 | iova = __intel_alloc_iova(hwdev, domain, size, pdev->dma_mask); |
f76aec76 KA |
2086 | if (!iova) |
2087 | goto error; | |
2088 | ||
5b6985ce | 2089 | start_paddr = (phys_addr_t)iova->pfn_lo << PAGE_SHIFT; |
f76aec76 | 2090 | |
ba395927 KA |
2091 | /* |
2092 | * Check if DMAR supports zero-length reads on write only | |
2093 | * mappings.. | |
2094 | */ | |
2095 | if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL || \ | |
8c11e798 | 2096 | !cap_zlr(iommu->cap)) |
ba395927 KA |
2097 | prot |= DMA_PTE_READ; |
2098 | if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) | |
2099 | prot |= DMA_PTE_WRITE; | |
2100 | /* | |
6865f0d1 | 2101 | * paddr - (paddr + size) might be partial page, we should map the whole |
ba395927 | 2102 | * page. Note: if two part of one page are separately mapped, we |
6865f0d1 | 2103 | * might have two guest_addr mapping to the same host paddr, but this |
ba395927 KA |
2104 | * is not a big problem |
2105 | */ | |
6865f0d1 | 2106 | ret = domain_page_mapping(domain, start_paddr, |
5b6985ce | 2107 | ((u64)paddr) & PAGE_MASK, size, prot); |
ba395927 KA |
2108 | if (ret) |
2109 | goto error; | |
2110 | ||
f76aec76 | 2111 | /* it's a non-present to present mapping */ |
8c11e798 | 2112 | ret = iommu_flush_iotlb_psi(iommu, domain->id, |
5b6985ce | 2113 | start_paddr, size >> VTD_PAGE_SHIFT, 1); |
f76aec76 | 2114 | if (ret) |
8c11e798 | 2115 | iommu_flush_write_buffer(iommu); |
f76aec76 | 2116 | |
5b6985ce | 2117 | return start_paddr + ((u64)paddr & (~PAGE_MASK)); |
ba395927 | 2118 | |
ba395927 | 2119 | error: |
f76aec76 KA |
2120 | if (iova) |
2121 | __free_iova(&domain->iovad, iova); | |
ba395927 | 2122 | printk(KERN_ERR"Device %s request: %lx@%llx dir %d --- failed\n", |
5b6985ce | 2123 | pci_name(pdev), size, (unsigned long long)paddr, dir); |
ba395927 KA |
2124 | return 0; |
2125 | } | |
2126 | ||
ffbbef5c FT |
2127 | static dma_addr_t intel_map_page(struct device *dev, struct page *page, |
2128 | unsigned long offset, size_t size, | |
2129 | enum dma_data_direction dir, | |
2130 | struct dma_attrs *attrs) | |
bb9e6d65 | 2131 | { |
ffbbef5c FT |
2132 | return __intel_map_single(dev, page_to_phys(page) + offset, size, |
2133 | dir, to_pci_dev(dev)->dma_mask); | |
bb9e6d65 FT |
2134 | } |
2135 | ||
5e0d2a6f | 2136 | static void flush_unmaps(void) |
2137 | { | |
80b20dd8 | 2138 | int i, j; |
5e0d2a6f | 2139 | |
5e0d2a6f | 2140 | timer_on = 0; |
2141 | ||
2142 | /* just flush them all */ | |
2143 | for (i = 0; i < g_num_of_iommus; i++) { | |
a2bb8459 WH |
2144 | struct intel_iommu *iommu = g_iommus[i]; |
2145 | if (!iommu) | |
2146 | continue; | |
c42d9f32 | 2147 | |
a2bb8459 | 2148 | if (deferred_flush[i].next) { |
a77b67d4 YS |
2149 | iommu->flush.flush_iotlb(iommu, 0, 0, 0, |
2150 | DMA_TLB_GLOBAL_FLUSH, 0); | |
80b20dd8 | 2151 | for (j = 0; j < deferred_flush[i].next; j++) { |
2152 | __free_iova(&deferred_flush[i].domain[j]->iovad, | |
2153 | deferred_flush[i].iova[j]); | |
2154 | } | |
2155 | deferred_flush[i].next = 0; | |
2156 | } | |
5e0d2a6f | 2157 | } |
2158 | ||
5e0d2a6f | 2159 | list_size = 0; |
5e0d2a6f | 2160 | } |
2161 | ||
2162 | static void flush_unmaps_timeout(unsigned long data) | |
2163 | { | |
80b20dd8 | 2164 | unsigned long flags; |
2165 | ||
2166 | spin_lock_irqsave(&async_umap_flush_lock, flags); | |
5e0d2a6f | 2167 | flush_unmaps(); |
80b20dd8 | 2168 | spin_unlock_irqrestore(&async_umap_flush_lock, flags); |
5e0d2a6f | 2169 | } |
2170 | ||
2171 | static void add_unmap(struct dmar_domain *dom, struct iova *iova) | |
2172 | { | |
2173 | unsigned long flags; | |
80b20dd8 | 2174 | int next, iommu_id; |
8c11e798 | 2175 | struct intel_iommu *iommu; |
5e0d2a6f | 2176 | |
2177 | spin_lock_irqsave(&async_umap_flush_lock, flags); | |
80b20dd8 | 2178 | if (list_size == HIGH_WATER_MARK) |
2179 | flush_unmaps(); | |
2180 | ||
8c11e798 WH |
2181 | iommu = domain_get_iommu(dom); |
2182 | iommu_id = iommu->seq_id; | |
c42d9f32 | 2183 | |
80b20dd8 | 2184 | next = deferred_flush[iommu_id].next; |
2185 | deferred_flush[iommu_id].domain[next] = dom; | |
2186 | deferred_flush[iommu_id].iova[next] = iova; | |
2187 | deferred_flush[iommu_id].next++; | |
5e0d2a6f | 2188 | |
2189 | if (!timer_on) { | |
2190 | mod_timer(&unmap_timer, jiffies + msecs_to_jiffies(10)); | |
2191 | timer_on = 1; | |
2192 | } | |
2193 | list_size++; | |
2194 | spin_unlock_irqrestore(&async_umap_flush_lock, flags); | |
2195 | } | |
2196 | ||
ffbbef5c FT |
2197 | static void intel_unmap_page(struct device *dev, dma_addr_t dev_addr, |
2198 | size_t size, enum dma_data_direction dir, | |
2199 | struct dma_attrs *attrs) | |
ba395927 | 2200 | { |
ba395927 | 2201 | struct pci_dev *pdev = to_pci_dev(dev); |
f76aec76 KA |
2202 | struct dmar_domain *domain; |
2203 | unsigned long start_addr; | |
ba395927 | 2204 | struct iova *iova; |
8c11e798 | 2205 | struct intel_iommu *iommu; |
ba395927 | 2206 | |
358dd8ac | 2207 | if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO) |
f76aec76 | 2208 | return; |
ba395927 KA |
2209 | domain = find_domain(pdev); |
2210 | BUG_ON(!domain); | |
2211 | ||
8c11e798 WH |
2212 | iommu = domain_get_iommu(domain); |
2213 | ||
ba395927 | 2214 | iova = find_iova(&domain->iovad, IOVA_PFN(dev_addr)); |
f76aec76 | 2215 | if (!iova) |
ba395927 | 2216 | return; |
ba395927 | 2217 | |
5b6985ce | 2218 | start_addr = iova->pfn_lo << PAGE_SHIFT; |
f76aec76 | 2219 | size = aligned_size((u64)dev_addr, size); |
ba395927 | 2220 | |
f76aec76 | 2221 | pr_debug("Device %s unmapping: %lx@%llx\n", |
5b6985ce | 2222 | pci_name(pdev), size, (unsigned long long)start_addr); |
ba395927 | 2223 | |
f76aec76 KA |
2224 | /* clear the whole page */ |
2225 | dma_pte_clear_range(domain, start_addr, start_addr + size); | |
2226 | /* free page tables */ | |
2227 | dma_pte_free_pagetable(domain, start_addr, start_addr + size); | |
5e0d2a6f | 2228 | if (intel_iommu_strict) { |
8c11e798 | 2229 | if (iommu_flush_iotlb_psi(iommu, |
5b6985ce | 2230 | domain->id, start_addr, size >> VTD_PAGE_SHIFT, 0)) |
8c11e798 | 2231 | iommu_flush_write_buffer(iommu); |
5e0d2a6f | 2232 | /* free iova */ |
2233 | __free_iova(&domain->iovad, iova); | |
2234 | } else { | |
2235 | add_unmap(domain, iova); | |
2236 | /* | |
2237 | * queue up the release of the unmap to save the 1/6th of the | |
2238 | * cpu used up by the iotlb flush operation... | |
2239 | */ | |
5e0d2a6f | 2240 | } |
ba395927 KA |
2241 | } |
2242 | ||
d7ab5c46 FT |
2243 | static void intel_unmap_single(struct device *dev, dma_addr_t dev_addr, size_t size, |
2244 | int dir) | |
ffbbef5c FT |
2245 | { |
2246 | intel_unmap_page(dev, dev_addr, size, dir, NULL); | |
2247 | } | |
2248 | ||
d7ab5c46 FT |
2249 | static void *intel_alloc_coherent(struct device *hwdev, size_t size, |
2250 | dma_addr_t *dma_handle, gfp_t flags) | |
ba395927 KA |
2251 | { |
2252 | void *vaddr; | |
2253 | int order; | |
2254 | ||
5b6985ce | 2255 | size = PAGE_ALIGN(size); |
ba395927 KA |
2256 | order = get_order(size); |
2257 | flags &= ~(GFP_DMA | GFP_DMA32); | |
2258 | ||
2259 | vaddr = (void *)__get_free_pages(flags, order); | |
2260 | if (!vaddr) | |
2261 | return NULL; | |
2262 | memset(vaddr, 0, size); | |
2263 | ||
bb9e6d65 FT |
2264 | *dma_handle = __intel_map_single(hwdev, virt_to_bus(vaddr), size, |
2265 | DMA_BIDIRECTIONAL, | |
2266 | hwdev->coherent_dma_mask); | |
ba395927 KA |
2267 | if (*dma_handle) |
2268 | return vaddr; | |
2269 | free_pages((unsigned long)vaddr, order); | |
2270 | return NULL; | |
2271 | } | |
2272 | ||
d7ab5c46 FT |
2273 | static void intel_free_coherent(struct device *hwdev, size_t size, void *vaddr, |
2274 | dma_addr_t dma_handle) | |
ba395927 KA |
2275 | { |
2276 | int order; | |
2277 | ||
5b6985ce | 2278 | size = PAGE_ALIGN(size); |
ba395927 KA |
2279 | order = get_order(size); |
2280 | ||
2281 | intel_unmap_single(hwdev, dma_handle, size, DMA_BIDIRECTIONAL); | |
2282 | free_pages((unsigned long)vaddr, order); | |
2283 | } | |
2284 | ||
12d4d40e | 2285 | #define SG_ENT_VIRT_ADDRESS(sg) (sg_virt((sg))) |
5b6985ce | 2286 | |
d7ab5c46 FT |
2287 | static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist, |
2288 | int nelems, enum dma_data_direction dir, | |
2289 | struct dma_attrs *attrs) | |
ba395927 KA |
2290 | { |
2291 | int i; | |
2292 | struct pci_dev *pdev = to_pci_dev(hwdev); | |
2293 | struct dmar_domain *domain; | |
f76aec76 KA |
2294 | unsigned long start_addr; |
2295 | struct iova *iova; | |
2296 | size_t size = 0; | |
2297 | void *addr; | |
c03ab37c | 2298 | struct scatterlist *sg; |
8c11e798 | 2299 | struct intel_iommu *iommu; |
ba395927 | 2300 | |
358dd8ac | 2301 | if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO) |
ba395927 KA |
2302 | return; |
2303 | ||
2304 | domain = find_domain(pdev); | |
8c11e798 WH |
2305 | BUG_ON(!domain); |
2306 | ||
2307 | iommu = domain_get_iommu(domain); | |
ba395927 | 2308 | |
c03ab37c | 2309 | iova = find_iova(&domain->iovad, IOVA_PFN(sglist[0].dma_address)); |
f76aec76 KA |
2310 | if (!iova) |
2311 | return; | |
c03ab37c | 2312 | for_each_sg(sglist, sg, nelems, i) { |
f76aec76 KA |
2313 | addr = SG_ENT_VIRT_ADDRESS(sg); |
2314 | size += aligned_size((u64)addr, sg->length); | |
2315 | } | |
2316 | ||
5b6985ce | 2317 | start_addr = iova->pfn_lo << PAGE_SHIFT; |
f76aec76 KA |
2318 | |
2319 | /* clear the whole page */ | |
2320 | dma_pte_clear_range(domain, start_addr, start_addr + size); | |
2321 | /* free page tables */ | |
2322 | dma_pte_free_pagetable(domain, start_addr, start_addr + size); | |
2323 | ||
8c11e798 | 2324 | if (iommu_flush_iotlb_psi(iommu, domain->id, start_addr, |
5b6985ce | 2325 | size >> VTD_PAGE_SHIFT, 0)) |
8c11e798 | 2326 | iommu_flush_write_buffer(iommu); |
f76aec76 KA |
2327 | |
2328 | /* free iova */ | |
2329 | __free_iova(&domain->iovad, iova); | |
ba395927 KA |
2330 | } |
2331 | ||
ba395927 | 2332 | static int intel_nontranslate_map_sg(struct device *hddev, |
c03ab37c | 2333 | struct scatterlist *sglist, int nelems, int dir) |
ba395927 KA |
2334 | { |
2335 | int i; | |
c03ab37c | 2336 | struct scatterlist *sg; |
ba395927 | 2337 | |
c03ab37c | 2338 | for_each_sg(sglist, sg, nelems, i) { |
12d4d40e | 2339 | BUG_ON(!sg_page(sg)); |
c03ab37c FT |
2340 | sg->dma_address = virt_to_bus(SG_ENT_VIRT_ADDRESS(sg)); |
2341 | sg->dma_length = sg->length; | |
ba395927 KA |
2342 | } |
2343 | return nelems; | |
2344 | } | |
2345 | ||
d7ab5c46 FT |
2346 | static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems, |
2347 | enum dma_data_direction dir, struct dma_attrs *attrs) | |
ba395927 KA |
2348 | { |
2349 | void *addr; | |
2350 | int i; | |
ba395927 KA |
2351 | struct pci_dev *pdev = to_pci_dev(hwdev); |
2352 | struct dmar_domain *domain; | |
f76aec76 KA |
2353 | size_t size = 0; |
2354 | int prot = 0; | |
2355 | size_t offset = 0; | |
2356 | struct iova *iova = NULL; | |
2357 | int ret; | |
c03ab37c | 2358 | struct scatterlist *sg; |
f76aec76 | 2359 | unsigned long start_addr; |
8c11e798 | 2360 | struct intel_iommu *iommu; |
ba395927 KA |
2361 | |
2362 | BUG_ON(dir == DMA_NONE); | |
358dd8ac | 2363 | if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO) |
c03ab37c | 2364 | return intel_nontranslate_map_sg(hwdev, sglist, nelems, dir); |
ba395927 | 2365 | |
f76aec76 KA |
2366 | domain = get_valid_domain_for_dev(pdev); |
2367 | if (!domain) | |
2368 | return 0; | |
2369 | ||
8c11e798 WH |
2370 | iommu = domain_get_iommu(domain); |
2371 | ||
c03ab37c | 2372 | for_each_sg(sglist, sg, nelems, i) { |
ba395927 | 2373 | addr = SG_ENT_VIRT_ADDRESS(sg); |
f76aec76 KA |
2374 | addr = (void *)virt_to_phys(addr); |
2375 | size += aligned_size((u64)addr, sg->length); | |
2376 | } | |
2377 | ||
bb9e6d65 | 2378 | iova = __intel_alloc_iova(hwdev, domain, size, pdev->dma_mask); |
f76aec76 | 2379 | if (!iova) { |
c03ab37c | 2380 | sglist->dma_length = 0; |
f76aec76 KA |
2381 | return 0; |
2382 | } | |
2383 | ||
2384 | /* | |
2385 | * Check if DMAR supports zero-length reads on write only | |
2386 | * mappings.. | |
2387 | */ | |
2388 | if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL || \ | |
8c11e798 | 2389 | !cap_zlr(iommu->cap)) |
f76aec76 KA |
2390 | prot |= DMA_PTE_READ; |
2391 | if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) | |
2392 | prot |= DMA_PTE_WRITE; | |
2393 | ||
5b6985ce | 2394 | start_addr = iova->pfn_lo << PAGE_SHIFT; |
f76aec76 | 2395 | offset = 0; |
c03ab37c | 2396 | for_each_sg(sglist, sg, nelems, i) { |
f76aec76 KA |
2397 | addr = SG_ENT_VIRT_ADDRESS(sg); |
2398 | addr = (void *)virt_to_phys(addr); | |
2399 | size = aligned_size((u64)addr, sg->length); | |
2400 | ret = domain_page_mapping(domain, start_addr + offset, | |
5b6985ce | 2401 | ((u64)addr) & PAGE_MASK, |
f76aec76 KA |
2402 | size, prot); |
2403 | if (ret) { | |
2404 | /* clear the page */ | |
2405 | dma_pte_clear_range(domain, start_addr, | |
2406 | start_addr + offset); | |
2407 | /* free page tables */ | |
2408 | dma_pte_free_pagetable(domain, start_addr, | |
2409 | start_addr + offset); | |
2410 | /* free iova */ | |
2411 | __free_iova(&domain->iovad, iova); | |
ba395927 KA |
2412 | return 0; |
2413 | } | |
f76aec76 | 2414 | sg->dma_address = start_addr + offset + |
5b6985ce | 2415 | ((u64)addr & (~PAGE_MASK)); |
ba395927 | 2416 | sg->dma_length = sg->length; |
f76aec76 | 2417 | offset += size; |
ba395927 KA |
2418 | } |
2419 | ||
ba395927 | 2420 | /* it's a non-present to present mapping */ |
8c11e798 | 2421 | if (iommu_flush_iotlb_psi(iommu, domain->id, |
5b6985ce | 2422 | start_addr, offset >> VTD_PAGE_SHIFT, 1)) |
8c11e798 | 2423 | iommu_flush_write_buffer(iommu); |
ba395927 KA |
2424 | return nelems; |
2425 | } | |
2426 | ||
dfb805e8 FT |
2427 | static int intel_mapping_error(struct device *dev, dma_addr_t dma_addr) |
2428 | { | |
2429 | return !dma_addr; | |
2430 | } | |
2431 | ||
160c1d8e | 2432 | struct dma_map_ops intel_dma_ops = { |
ba395927 KA |
2433 | .alloc_coherent = intel_alloc_coherent, |
2434 | .free_coherent = intel_free_coherent, | |
ba395927 KA |
2435 | .map_sg = intel_map_sg, |
2436 | .unmap_sg = intel_unmap_sg, | |
ffbbef5c FT |
2437 | .map_page = intel_map_page, |
2438 | .unmap_page = intel_unmap_page, | |
dfb805e8 | 2439 | .mapping_error = intel_mapping_error, |
ba395927 KA |
2440 | }; |
2441 | ||
2442 | static inline int iommu_domain_cache_init(void) | |
2443 | { | |
2444 | int ret = 0; | |
2445 | ||
2446 | iommu_domain_cache = kmem_cache_create("iommu_domain", | |
2447 | sizeof(struct dmar_domain), | |
2448 | 0, | |
2449 | SLAB_HWCACHE_ALIGN, | |
2450 | ||
2451 | NULL); | |
2452 | if (!iommu_domain_cache) { | |
2453 | printk(KERN_ERR "Couldn't create iommu_domain cache\n"); | |
2454 | ret = -ENOMEM; | |
2455 | } | |
2456 | ||
2457 | return ret; | |
2458 | } | |
2459 | ||
2460 | static inline int iommu_devinfo_cache_init(void) | |
2461 | { | |
2462 | int ret = 0; | |
2463 | ||
2464 | iommu_devinfo_cache = kmem_cache_create("iommu_devinfo", | |
2465 | sizeof(struct device_domain_info), | |
2466 | 0, | |
2467 | SLAB_HWCACHE_ALIGN, | |
ba395927 KA |
2468 | NULL); |
2469 | if (!iommu_devinfo_cache) { | |
2470 | printk(KERN_ERR "Couldn't create devinfo cache\n"); | |
2471 | ret = -ENOMEM; | |
2472 | } | |
2473 | ||
2474 | return ret; | |
2475 | } | |
2476 | ||
2477 | static inline int iommu_iova_cache_init(void) | |
2478 | { | |
2479 | int ret = 0; | |
2480 | ||
2481 | iommu_iova_cache = kmem_cache_create("iommu_iova", | |
2482 | sizeof(struct iova), | |
2483 | 0, | |
2484 | SLAB_HWCACHE_ALIGN, | |
ba395927 KA |
2485 | NULL); |
2486 | if (!iommu_iova_cache) { | |
2487 | printk(KERN_ERR "Couldn't create iova cache\n"); | |
2488 | ret = -ENOMEM; | |
2489 | } | |
2490 | ||
2491 | return ret; | |
2492 | } | |
2493 | ||
2494 | static int __init iommu_init_mempool(void) | |
2495 | { | |
2496 | int ret; | |
2497 | ret = iommu_iova_cache_init(); | |
2498 | if (ret) | |
2499 | return ret; | |
2500 | ||
2501 | ret = iommu_domain_cache_init(); | |
2502 | if (ret) | |
2503 | goto domain_error; | |
2504 | ||
2505 | ret = iommu_devinfo_cache_init(); | |
2506 | if (!ret) | |
2507 | return ret; | |
2508 | ||
2509 | kmem_cache_destroy(iommu_domain_cache); | |
2510 | domain_error: | |
2511 | kmem_cache_destroy(iommu_iova_cache); | |
2512 | ||
2513 | return -ENOMEM; | |
2514 | } | |
2515 | ||
2516 | static void __init iommu_exit_mempool(void) | |
2517 | { | |
2518 | kmem_cache_destroy(iommu_devinfo_cache); | |
2519 | kmem_cache_destroy(iommu_domain_cache); | |
2520 | kmem_cache_destroy(iommu_iova_cache); | |
2521 | ||
2522 | } | |
2523 | ||
ba395927 KA |
2524 | static void __init init_no_remapping_devices(void) |
2525 | { | |
2526 | struct dmar_drhd_unit *drhd; | |
2527 | ||
2528 | for_each_drhd_unit(drhd) { | |
2529 | if (!drhd->include_all) { | |
2530 | int i; | |
2531 | for (i = 0; i < drhd->devices_cnt; i++) | |
2532 | if (drhd->devices[i] != NULL) | |
2533 | break; | |
2534 | /* ignore DMAR unit if no pci devices exist */ | |
2535 | if (i == drhd->devices_cnt) | |
2536 | drhd->ignored = 1; | |
2537 | } | |
2538 | } | |
2539 | ||
2540 | if (dmar_map_gfx) | |
2541 | return; | |
2542 | ||
2543 | for_each_drhd_unit(drhd) { | |
2544 | int i; | |
2545 | if (drhd->ignored || drhd->include_all) | |
2546 | continue; | |
2547 | ||
2548 | for (i = 0; i < drhd->devices_cnt; i++) | |
2549 | if (drhd->devices[i] && | |
2550 | !IS_GFX_DEVICE(drhd->devices[i])) | |
2551 | break; | |
2552 | ||
2553 | if (i < drhd->devices_cnt) | |
2554 | continue; | |
2555 | ||
2556 | /* bypass IOMMU if it is just for gfx devices */ | |
2557 | drhd->ignored = 1; | |
2558 | for (i = 0; i < drhd->devices_cnt; i++) { | |
2559 | if (!drhd->devices[i]) | |
2560 | continue; | |
358dd8ac | 2561 | drhd->devices[i]->dev.archdata.iommu = DUMMY_DEVICE_DOMAIN_INFO; |
ba395927 KA |
2562 | } |
2563 | } | |
2564 | } | |
2565 | ||
2566 | int __init intel_iommu_init(void) | |
2567 | { | |
2568 | int ret = 0; | |
2569 | ||
ba395927 KA |
2570 | if (dmar_table_init()) |
2571 | return -ENODEV; | |
2572 | ||
1886e8a9 SS |
2573 | if (dmar_dev_scope_init()) |
2574 | return -ENODEV; | |
2575 | ||
2ae21010 SS |
2576 | /* |
2577 | * Check the need for DMA-remapping initialization now. | |
2578 | * Above initialization will also be used by Interrupt-remapping. | |
2579 | */ | |
2580 | if (no_iommu || swiotlb || dmar_disabled) | |
2581 | return -ENODEV; | |
2582 | ||
ba395927 KA |
2583 | iommu_init_mempool(); |
2584 | dmar_init_reserved_ranges(); | |
2585 | ||
2586 | init_no_remapping_devices(); | |
2587 | ||
2588 | ret = init_dmars(); | |
2589 | if (ret) { | |
2590 | printk(KERN_ERR "IOMMU: dmar init failed\n"); | |
2591 | put_iova_domain(&reserved_iova_list); | |
2592 | iommu_exit_mempool(); | |
2593 | return ret; | |
2594 | } | |
2595 | printk(KERN_INFO | |
2596 | "PCI-DMA: Intel(R) Virtualization Technology for Directed I/O\n"); | |
2597 | ||
5e0d2a6f | 2598 | init_timer(&unmap_timer); |
ba395927 KA |
2599 | force_iommu = 1; |
2600 | dma_ops = &intel_dma_ops; | |
a8bcbb0d JR |
2601 | |
2602 | register_iommu(&intel_iommu_ops); | |
2603 | ||
ba395927 KA |
2604 | return 0; |
2605 | } | |
e820482c | 2606 | |
c7151a8d WH |
2607 | static int vm_domain_add_dev_info(struct dmar_domain *domain, |
2608 | struct pci_dev *pdev) | |
2609 | { | |
2610 | struct device_domain_info *info; | |
2611 | unsigned long flags; | |
2612 | ||
2613 | info = alloc_devinfo_mem(); | |
2614 | if (!info) | |
2615 | return -ENOMEM; | |
2616 | ||
2617 | info->bus = pdev->bus->number; | |
2618 | info->devfn = pdev->devfn; | |
2619 | info->dev = pdev; | |
2620 | info->domain = domain; | |
2621 | ||
2622 | spin_lock_irqsave(&device_domain_lock, flags); | |
2623 | list_add(&info->link, &domain->devices); | |
2624 | list_add(&info->global, &device_domain_list); | |
2625 | pdev->dev.archdata.iommu = info; | |
2626 | spin_unlock_irqrestore(&device_domain_lock, flags); | |
2627 | ||
2628 | return 0; | |
2629 | } | |
2630 | ||
2631 | static void vm_domain_remove_one_dev_info(struct dmar_domain *domain, | |
2632 | struct pci_dev *pdev) | |
2633 | { | |
2634 | struct device_domain_info *info; | |
2635 | struct intel_iommu *iommu; | |
2636 | unsigned long flags; | |
2637 | int found = 0; | |
2638 | struct list_head *entry, *tmp; | |
2639 | ||
2640 | iommu = device_to_iommu(pdev->bus->number, pdev->devfn); | |
2641 | if (!iommu) | |
2642 | return; | |
2643 | ||
2644 | spin_lock_irqsave(&device_domain_lock, flags); | |
2645 | list_for_each_safe(entry, tmp, &domain->devices) { | |
2646 | info = list_entry(entry, struct device_domain_info, link); | |
2647 | if (info->bus == pdev->bus->number && | |
2648 | info->devfn == pdev->devfn) { | |
2649 | list_del(&info->link); | |
2650 | list_del(&info->global); | |
2651 | if (info->dev) | |
2652 | info->dev->dev.archdata.iommu = NULL; | |
2653 | spin_unlock_irqrestore(&device_domain_lock, flags); | |
2654 | ||
2655 | iommu_detach_dev(iommu, info->bus, info->devfn); | |
2656 | free_devinfo_mem(info); | |
2657 | ||
2658 | spin_lock_irqsave(&device_domain_lock, flags); | |
2659 | ||
2660 | if (found) | |
2661 | break; | |
2662 | else | |
2663 | continue; | |
2664 | } | |
2665 | ||
2666 | /* if there is no other devices under the same iommu | |
2667 | * owned by this domain, clear this iommu in iommu_bmp | |
2668 | * update iommu count and coherency | |
2669 | */ | |
2670 | if (device_to_iommu(info->bus, info->devfn) == iommu) | |
2671 | found = 1; | |
2672 | } | |
2673 | ||
2674 | if (found == 0) { | |
2675 | unsigned long tmp_flags; | |
2676 | spin_lock_irqsave(&domain->iommu_lock, tmp_flags); | |
2677 | clear_bit(iommu->seq_id, &domain->iommu_bmp); | |
2678 | domain->iommu_count--; | |
2679 | domain_update_iommu_coherency(domain); | |
2680 | spin_unlock_irqrestore(&domain->iommu_lock, tmp_flags); | |
2681 | } | |
2682 | ||
2683 | spin_unlock_irqrestore(&device_domain_lock, flags); | |
2684 | } | |
2685 | ||
2686 | static void vm_domain_remove_all_dev_info(struct dmar_domain *domain) | |
2687 | { | |
2688 | struct device_domain_info *info; | |
2689 | struct intel_iommu *iommu; | |
2690 | unsigned long flags1, flags2; | |
2691 | ||
2692 | spin_lock_irqsave(&device_domain_lock, flags1); | |
2693 | while (!list_empty(&domain->devices)) { | |
2694 | info = list_entry(domain->devices.next, | |
2695 | struct device_domain_info, link); | |
2696 | list_del(&info->link); | |
2697 | list_del(&info->global); | |
2698 | if (info->dev) | |
2699 | info->dev->dev.archdata.iommu = NULL; | |
2700 | ||
2701 | spin_unlock_irqrestore(&device_domain_lock, flags1); | |
2702 | ||
2703 | iommu = device_to_iommu(info->bus, info->devfn); | |
2704 | iommu_detach_dev(iommu, info->bus, info->devfn); | |
2705 | ||
2706 | /* clear this iommu in iommu_bmp, update iommu count | |
2707 | * and coherency | |
2708 | */ | |
2709 | spin_lock_irqsave(&domain->iommu_lock, flags2); | |
2710 | if (test_and_clear_bit(iommu->seq_id, | |
2711 | &domain->iommu_bmp)) { | |
2712 | domain->iommu_count--; | |
2713 | domain_update_iommu_coherency(domain); | |
2714 | } | |
2715 | spin_unlock_irqrestore(&domain->iommu_lock, flags2); | |
2716 | ||
2717 | free_devinfo_mem(info); | |
2718 | spin_lock_irqsave(&device_domain_lock, flags1); | |
2719 | } | |
2720 | spin_unlock_irqrestore(&device_domain_lock, flags1); | |
2721 | } | |
2722 | ||
5e98c4b1 WH |
2723 | /* domain id for virtual machine, it won't be set in context */ |
2724 | static unsigned long vm_domid; | |
2725 | ||
fe40f1e0 WH |
2726 | static int vm_domain_min_agaw(struct dmar_domain *domain) |
2727 | { | |
2728 | int i; | |
2729 | int min_agaw = domain->agaw; | |
2730 | ||
2731 | i = find_first_bit(&domain->iommu_bmp, g_num_of_iommus); | |
2732 | for (; i < g_num_of_iommus; ) { | |
2733 | if (min_agaw > g_iommus[i]->agaw) | |
2734 | min_agaw = g_iommus[i]->agaw; | |
2735 | ||
2736 | i = find_next_bit(&domain->iommu_bmp, g_num_of_iommus, i+1); | |
2737 | } | |
2738 | ||
2739 | return min_agaw; | |
2740 | } | |
2741 | ||
5e98c4b1 WH |
2742 | static struct dmar_domain *iommu_alloc_vm_domain(void) |
2743 | { | |
2744 | struct dmar_domain *domain; | |
2745 | ||
2746 | domain = alloc_domain_mem(); | |
2747 | if (!domain) | |
2748 | return NULL; | |
2749 | ||
2750 | domain->id = vm_domid++; | |
2751 | memset(&domain->iommu_bmp, 0, sizeof(unsigned long)); | |
2752 | domain->flags = DOMAIN_FLAG_VIRTUAL_MACHINE; | |
2753 | ||
2754 | return domain; | |
2755 | } | |
2756 | ||
2757 | static int vm_domain_init(struct dmar_domain *domain, int guest_width) | |
2758 | { | |
2759 | int adjust_width; | |
2760 | ||
2761 | init_iova_domain(&domain->iovad, DMA_32BIT_PFN); | |
2762 | spin_lock_init(&domain->mapping_lock); | |
2763 | spin_lock_init(&domain->iommu_lock); | |
2764 | ||
2765 | domain_reserve_special_ranges(domain); | |
2766 | ||
2767 | /* calculate AGAW */ | |
2768 | domain->gaw = guest_width; | |
2769 | adjust_width = guestwidth_to_adjustwidth(guest_width); | |
2770 | domain->agaw = width_to_agaw(adjust_width); | |
2771 | ||
2772 | INIT_LIST_HEAD(&domain->devices); | |
2773 | ||
2774 | domain->iommu_count = 0; | |
2775 | domain->iommu_coherency = 0; | |
fe40f1e0 | 2776 | domain->max_addr = 0; |
5e98c4b1 WH |
2777 | |
2778 | /* always allocate the top pgd */ | |
2779 | domain->pgd = (struct dma_pte *)alloc_pgtable_page(); | |
2780 | if (!domain->pgd) | |
2781 | return -ENOMEM; | |
2782 | domain_flush_cache(domain, domain->pgd, PAGE_SIZE); | |
2783 | return 0; | |
2784 | } | |
2785 | ||
2786 | static void iommu_free_vm_domain(struct dmar_domain *domain) | |
2787 | { | |
2788 | unsigned long flags; | |
2789 | struct dmar_drhd_unit *drhd; | |
2790 | struct intel_iommu *iommu; | |
2791 | unsigned long i; | |
2792 | unsigned long ndomains; | |
2793 | ||
2794 | for_each_drhd_unit(drhd) { | |
2795 | if (drhd->ignored) | |
2796 | continue; | |
2797 | iommu = drhd->iommu; | |
2798 | ||
2799 | ndomains = cap_ndoms(iommu->cap); | |
2800 | i = find_first_bit(iommu->domain_ids, ndomains); | |
2801 | for (; i < ndomains; ) { | |
2802 | if (iommu->domains[i] == domain) { | |
2803 | spin_lock_irqsave(&iommu->lock, flags); | |
2804 | clear_bit(i, iommu->domain_ids); | |
2805 | iommu->domains[i] = NULL; | |
2806 | spin_unlock_irqrestore(&iommu->lock, flags); | |
2807 | break; | |
2808 | } | |
2809 | i = find_next_bit(iommu->domain_ids, ndomains, i+1); | |
2810 | } | |
2811 | } | |
2812 | } | |
2813 | ||
2814 | static void vm_domain_exit(struct dmar_domain *domain) | |
2815 | { | |
2816 | u64 end; | |
2817 | ||
2818 | /* Domain 0 is reserved, so dont process it */ | |
2819 | if (!domain) | |
2820 | return; | |
2821 | ||
2822 | vm_domain_remove_all_dev_info(domain); | |
2823 | /* destroy iovas */ | |
2824 | put_iova_domain(&domain->iovad); | |
2825 | end = DOMAIN_MAX_ADDR(domain->gaw); | |
2826 | end = end & (~VTD_PAGE_MASK); | |
2827 | ||
2828 | /* clear ptes */ | |
2829 | dma_pte_clear_range(domain, 0, end); | |
2830 | ||
2831 | /* free page tables */ | |
2832 | dma_pte_free_pagetable(domain, 0, end); | |
2833 | ||
2834 | iommu_free_vm_domain(domain); | |
2835 | free_domain_mem(domain); | |
2836 | } | |
2837 | ||
5d450806 | 2838 | static int intel_iommu_domain_init(struct iommu_domain *domain) |
38717946 | 2839 | { |
5d450806 | 2840 | struct dmar_domain *dmar_domain; |
38717946 | 2841 | |
5d450806 JR |
2842 | dmar_domain = iommu_alloc_vm_domain(); |
2843 | if (!dmar_domain) { | |
38717946 | 2844 | printk(KERN_ERR |
5d450806 JR |
2845 | "intel_iommu_domain_init: dmar_domain == NULL\n"); |
2846 | return -ENOMEM; | |
38717946 | 2847 | } |
5d450806 | 2848 | if (vm_domain_init(dmar_domain, DEFAULT_DOMAIN_ADDRESS_WIDTH)) { |
38717946 | 2849 | printk(KERN_ERR |
5d450806 JR |
2850 | "intel_iommu_domain_init() failed\n"); |
2851 | vm_domain_exit(dmar_domain); | |
2852 | return -ENOMEM; | |
38717946 | 2853 | } |
5d450806 | 2854 | domain->priv = dmar_domain; |
faa3d6f5 | 2855 | |
5d450806 | 2856 | return 0; |
38717946 | 2857 | } |
38717946 | 2858 | |
5d450806 | 2859 | static void intel_iommu_domain_destroy(struct iommu_domain *domain) |
38717946 | 2860 | { |
5d450806 JR |
2861 | struct dmar_domain *dmar_domain = domain->priv; |
2862 | ||
2863 | domain->priv = NULL; | |
2864 | vm_domain_exit(dmar_domain); | |
38717946 | 2865 | } |
38717946 | 2866 | |
4c5478c9 JR |
2867 | static int intel_iommu_attach_device(struct iommu_domain *domain, |
2868 | struct device *dev) | |
38717946 | 2869 | { |
4c5478c9 JR |
2870 | struct dmar_domain *dmar_domain = domain->priv; |
2871 | struct pci_dev *pdev = to_pci_dev(dev); | |
fe40f1e0 WH |
2872 | struct intel_iommu *iommu; |
2873 | int addr_width; | |
2874 | u64 end; | |
faa3d6f5 WH |
2875 | int ret; |
2876 | ||
2877 | /* normally pdev is not mapped */ | |
2878 | if (unlikely(domain_context_mapped(pdev))) { | |
2879 | struct dmar_domain *old_domain; | |
2880 | ||
2881 | old_domain = find_domain(pdev); | |
2882 | if (old_domain) { | |
4c5478c9 | 2883 | if (dmar_domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) |
faa3d6f5 WH |
2884 | vm_domain_remove_one_dev_info(old_domain, pdev); |
2885 | else | |
2886 | domain_remove_dev_info(old_domain); | |
2887 | } | |
2888 | } | |
2889 | ||
fe40f1e0 WH |
2890 | iommu = device_to_iommu(pdev->bus->number, pdev->devfn); |
2891 | if (!iommu) | |
2892 | return -ENODEV; | |
2893 | ||
2894 | /* check if this iommu agaw is sufficient for max mapped address */ | |
2895 | addr_width = agaw_to_width(iommu->agaw); | |
2896 | end = DOMAIN_MAX_ADDR(addr_width); | |
2897 | end = end & VTD_PAGE_MASK; | |
4c5478c9 | 2898 | if (end < dmar_domain->max_addr) { |
fe40f1e0 WH |
2899 | printk(KERN_ERR "%s: iommu agaw (%d) is not " |
2900 | "sufficient for the mapped address (%llx)\n", | |
4c5478c9 | 2901 | __func__, iommu->agaw, dmar_domain->max_addr); |
fe40f1e0 WH |
2902 | return -EFAULT; |
2903 | } | |
2904 | ||
4c5478c9 | 2905 | ret = domain_context_mapping(dmar_domain, pdev); |
faa3d6f5 WH |
2906 | if (ret) |
2907 | return ret; | |
2908 | ||
4c5478c9 | 2909 | ret = vm_domain_add_dev_info(dmar_domain, pdev); |
faa3d6f5 | 2910 | return ret; |
38717946 | 2911 | } |
38717946 | 2912 | |
4c5478c9 JR |
2913 | static void intel_iommu_detach_device(struct iommu_domain *domain, |
2914 | struct device *dev) | |
38717946 | 2915 | { |
4c5478c9 JR |
2916 | struct dmar_domain *dmar_domain = domain->priv; |
2917 | struct pci_dev *pdev = to_pci_dev(dev); | |
2918 | ||
2919 | vm_domain_remove_one_dev_info(dmar_domain, pdev); | |
faa3d6f5 | 2920 | } |
c7151a8d | 2921 | |
dde57a21 JR |
2922 | static int intel_iommu_map_range(struct iommu_domain *domain, |
2923 | unsigned long iova, phys_addr_t hpa, | |
2924 | size_t size, int iommu_prot) | |
faa3d6f5 | 2925 | { |
dde57a21 | 2926 | struct dmar_domain *dmar_domain = domain->priv; |
fe40f1e0 WH |
2927 | u64 max_addr; |
2928 | int addr_width; | |
dde57a21 | 2929 | int prot = 0; |
faa3d6f5 | 2930 | int ret; |
fe40f1e0 | 2931 | |
dde57a21 JR |
2932 | if (iommu_prot & IOMMU_READ) |
2933 | prot |= DMA_PTE_READ; | |
2934 | if (iommu_prot & IOMMU_WRITE) | |
2935 | prot |= DMA_PTE_WRITE; | |
2936 | ||
fe40f1e0 | 2937 | max_addr = (iova & VTD_PAGE_MASK) + VTD_PAGE_ALIGN(size); |
dde57a21 | 2938 | if (dmar_domain->max_addr < max_addr) { |
fe40f1e0 WH |
2939 | int min_agaw; |
2940 | u64 end; | |
2941 | ||
2942 | /* check if minimum agaw is sufficient for mapped address */ | |
dde57a21 | 2943 | min_agaw = vm_domain_min_agaw(dmar_domain); |
fe40f1e0 WH |
2944 | addr_width = agaw_to_width(min_agaw); |
2945 | end = DOMAIN_MAX_ADDR(addr_width); | |
2946 | end = end & VTD_PAGE_MASK; | |
2947 | if (end < max_addr) { | |
2948 | printk(KERN_ERR "%s: iommu agaw (%d) is not " | |
2949 | "sufficient for the mapped address (%llx)\n", | |
2950 | __func__, min_agaw, max_addr); | |
2951 | return -EFAULT; | |
2952 | } | |
dde57a21 | 2953 | dmar_domain->max_addr = max_addr; |
fe40f1e0 WH |
2954 | } |
2955 | ||
dde57a21 | 2956 | ret = domain_page_mapping(dmar_domain, iova, hpa, size, prot); |
faa3d6f5 | 2957 | return ret; |
38717946 | 2958 | } |
38717946 | 2959 | |
dde57a21 JR |
2960 | static void intel_iommu_unmap_range(struct iommu_domain *domain, |
2961 | unsigned long iova, size_t size) | |
38717946 | 2962 | { |
dde57a21 | 2963 | struct dmar_domain *dmar_domain = domain->priv; |
faa3d6f5 WH |
2964 | dma_addr_t base; |
2965 | ||
2966 | /* The address might not be aligned */ | |
2967 | base = iova & VTD_PAGE_MASK; | |
2968 | size = VTD_PAGE_ALIGN(size); | |
dde57a21 | 2969 | dma_pte_clear_range(dmar_domain, base, base + size); |
fe40f1e0 | 2970 | |
dde57a21 JR |
2971 | if (dmar_domain->max_addr == base + size) |
2972 | dmar_domain->max_addr = base; | |
38717946 | 2973 | } |
38717946 | 2974 | |
d14d6577 JR |
2975 | static phys_addr_t intel_iommu_iova_to_phys(struct iommu_domain *domain, |
2976 | unsigned long iova) | |
38717946 | 2977 | { |
d14d6577 | 2978 | struct dmar_domain *dmar_domain = domain->priv; |
38717946 | 2979 | struct dma_pte *pte; |
faa3d6f5 | 2980 | u64 phys = 0; |
38717946 | 2981 | |
d14d6577 | 2982 | pte = addr_to_dma_pte(dmar_domain, iova); |
38717946 | 2983 | if (pte) |
faa3d6f5 | 2984 | phys = dma_pte_addr(pte); |
38717946 | 2985 | |
faa3d6f5 | 2986 | return phys; |
38717946 | 2987 | } |
a8bcbb0d JR |
2988 | |
2989 | static struct iommu_ops intel_iommu_ops = { | |
2990 | .domain_init = intel_iommu_domain_init, | |
2991 | .domain_destroy = intel_iommu_domain_destroy, | |
2992 | .attach_dev = intel_iommu_attach_device, | |
2993 | .detach_dev = intel_iommu_detach_device, | |
2994 | .map = intel_iommu_map_range, | |
2995 | .unmap = intel_iommu_unmap_range, | |
2996 | .iova_to_phys = intel_iommu_iova_to_phys, | |
2997 | }; | |
9af88143 DW |
2998 | |
2999 | static void __devinit quirk_iommu_rwbf(struct pci_dev *dev) | |
3000 | { | |
3001 | /* | |
3002 | * Mobile 4 Series Chipset neglects to set RWBF capability, | |
3003 | * but needs it: | |
3004 | */ | |
3005 | printk(KERN_INFO "DMAR: Forcing write-buffer flush capability\n"); | |
3006 | rwbf_quirk = 1; | |
3007 | } | |
3008 | ||
3009 | DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x2a40, quirk_iommu_rwbf); |