?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ACPI / Wakeup: Enable button GPEs unconditionally during initialization
[linux/fpc-iii.git] / drivers / pci / intel-iommu.c
blob4789f8e8bf7ad91cae0b0a1043349ac96ea5fa64
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 *
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>
21 * Author: Fenghua Yu <fenghua.yu@intel.com>
22 */
23
24 #include <linux/init.h>
25 #include <linux/bitmap.h>
26 #include <linux/debugfs.h>
27 #include <linux/slab.h>
28 #include <linux/irq.h>
29 #include <linux/interrupt.h>
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>
35 #include <linux/timer.h>
36 #include <linux/iova.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/sysdev.h>
40 #include <linux/tboot.h>
41 #include <linux/dmi.h>
42 #include <asm/cacheflush.h>
43 #include <asm/iommu.h>
44 #include "pci.h"
45
46 #define ROOT_SIZE VTD_PAGE_SIZE
47 #define CONTEXT_SIZE VTD_PAGE_SIZE
48
49 #define IS_GFX_DEVICE(pdev) ((pdev->class >> 16) == PCI_BASE_CLASS_DISPLAY)
50 #define IS_ISA_DEVICE(pdev) ((pdev->class >> 8) == PCI_CLASS_BRIDGE_ISA)
51 #define IS_AZALIA(pdev) ((pdev)->vendor == 0x8086 && (pdev)->device == 0x3a3e)
52
53 #define IOAPIC_RANGE_START (0xfee00000)
54 #define IOAPIC_RANGE_END (0xfeefffff)
55 #define IOVA_START_ADDR (0x1000)
56
57 #define DEFAULT_DOMAIN_ADDRESS_WIDTH 48
58
59 #define MAX_AGAW_WIDTH 64
60
61 #define __DOMAIN_MAX_PFN(gaw) ((((uint64_t)1) << (gaw-VTD_PAGE_SHIFT)) - 1)
62 #define __DOMAIN_MAX_ADDR(gaw) ((((uint64_t)1) << gaw) - 1)
63
64 /* We limit DOMAIN_MAX_PFN to fit in an unsigned long, and DOMAIN_MAX_ADDR
65 to match. That way, we can use 'unsigned long' for PFNs with impunity. */
66 #define DOMAIN_MAX_PFN(gaw) ((unsigned long) min_t(uint64_t, \
67 __DOMAIN_MAX_PFN(gaw), (unsigned long)-1))
68 #define DOMAIN_MAX_ADDR(gaw) (((uint64_t)__DOMAIN_MAX_PFN(gaw)) << VTD_PAGE_SHIFT)
69
70 #define IOVA_PFN(addr) ((addr) >> PAGE_SHIFT)
71 #define DMA_32BIT_PFN IOVA_PFN(DMA_BIT_MASK(32))
72 #define DMA_64BIT_PFN IOVA_PFN(DMA_BIT_MASK(64))
73
74 /* page table handling */
75 #define LEVEL_STRIDE (9)
76 #define LEVEL_MASK (((u64)1 << LEVEL_STRIDE) - 1)
77
78 static inline int agaw_to_level(int agaw)
79 {
80 return agaw + 2;
81 }
82
83 static inline int agaw_to_width(int agaw)
84 {
85 return 30 + agaw * LEVEL_STRIDE;
86 }
87
88 static inline int width_to_agaw(int width)
89 {
90 return (width - 30) / LEVEL_STRIDE;
91 }
92
93 static inline unsigned int level_to_offset_bits(int level)
94 {
95 return (level - 1) * LEVEL_STRIDE;
96 }
97
98 static inline int pfn_level_offset(unsigned long pfn, int level)
99 {
100 return (pfn >> level_to_offset_bits(level)) & LEVEL_MASK;
101 }
102
103 static inline unsigned long level_mask(int level)
104 {
105 return -1UL << level_to_offset_bits(level);
106 }
107
108 static inline unsigned long level_size(int level)
109 {
110 return 1UL << level_to_offset_bits(level);
111 }
112
113 static inline unsigned long align_to_level(unsigned long pfn, int level)
114 {
115 return (pfn + level_size(level) - 1) & level_mask(level);
116 }
117
118 /* VT-d pages must always be _smaller_ than MM pages. Otherwise things
119 are never going to work. */
120 static inline unsigned long dma_to_mm_pfn(unsigned long dma_pfn)
121 {
122 return dma_pfn >> (PAGE_SHIFT - VTD_PAGE_SHIFT);
123 }
124
125 static inline unsigned long mm_to_dma_pfn(unsigned long mm_pfn)
126 {
127 return mm_pfn << (PAGE_SHIFT - VTD_PAGE_SHIFT);
128 }
129 static inline unsigned long page_to_dma_pfn(struct page *pg)
130 {
131 return mm_to_dma_pfn(page_to_pfn(pg));
132 }
133 static inline unsigned long virt_to_dma_pfn(void *p)
134 {
135 return page_to_dma_pfn(virt_to_page(p));
136 }
137
138 /* global iommu list, set NULL for ignored DMAR units */
139 static struct intel_iommu **g_iommus;
140
141 static void __init check_tylersburg_isoch(void);
142 static int rwbf_quirk;
143
144 /*
145 * 0: Present
146 * 1-11: Reserved
147 * 12-63: Context Ptr (12 - (haw-1))
148 * 64-127: Reserved
149 */
150 struct root_entry {
151 u64 val;
152 u64 rsvd1;
153 };
154 #define ROOT_ENTRY_NR (VTD_PAGE_SIZE/sizeof(struct root_entry))
155 static inline bool root_present(struct root_entry *root)
156 {
157 return (root->val & 1);
158 }
159 static inline void set_root_present(struct root_entry *root)
160 {
161 root->val |= 1;
162 }
163 static inline void set_root_value(struct root_entry *root, unsigned long value)
164 {
165 root->val |= value & VTD_PAGE_MASK;
166 }
167
168 static inline struct context_entry *
169 get_context_addr_from_root(struct root_entry *root)
170 {
171 return (struct context_entry *)
172 (root_present(root)?phys_to_virt(
173 root->val & VTD_PAGE_MASK) :
174 NULL);
175 }
176
177 /*
178 * low 64 bits:
179 * 0: present
180 * 1: fault processing disable
181 * 2-3: translation type
182 * 12-63: address space root
183 * high 64 bits:
184 * 0-2: address width
185 * 3-6: aval
186 * 8-23: domain id
187 */
188 struct context_entry {
189 u64 lo;
190 u64 hi;
191 };
192
193 static inline bool context_present(struct context_entry *context)
194 {
195 return (context->lo & 1);
196 }
197 static inline void context_set_present(struct context_entry *context)
198 {
199 context->lo |= 1;
200 }
201
202 static inline void context_set_fault_enable(struct context_entry *context)
203 {
204 context->lo &= (((u64)-1) << 2) | 1;
205 }
206
207 static inline void context_set_translation_type(struct context_entry *context,
208 unsigned long value)
209 {
210 context->lo &= (((u64)-1) << 4) | 3;
211 context->lo |= (value & 3) << 2;
212 }
213
214 static inline void context_set_address_root(struct context_entry *context,
215 unsigned long value)
216 {
217 context->lo |= value & VTD_PAGE_MASK;
218 }
219
220 static inline void context_set_address_width(struct context_entry *context,
221 unsigned long value)
222 {
223 context->hi |= value & 7;
224 }
225
226 static inline void context_set_domain_id(struct context_entry *context,
227 unsigned long value)
228 {
229 context->hi |= (value & ((1 << 16) - 1)) << 8;
230 }
231
232 static inline void context_clear_entry(struct context_entry *context)
233 {
234 context->lo = 0;
235 context->hi = 0;
236 }
237
238 /*
239 * 0: readable
240 * 1: writable
241 * 2-6: reserved
242 * 7: super page
243 * 8-10: available
244 * 11: snoop behavior
245 * 12-63: Host physcial address
246 */
247 struct dma_pte {
248 u64 val;
249 };
250
251 static inline void dma_clear_pte(struct dma_pte *pte)
252 {
253 pte->val = 0;
254 }
255
256 static inline void dma_set_pte_readable(struct dma_pte *pte)
257 {
258 pte->val |= DMA_PTE_READ;
259 }
260
261 static inline void dma_set_pte_writable(struct dma_pte *pte)
262 {
263 pte->val |= DMA_PTE_WRITE;
264 }
265
266 static inline void dma_set_pte_snp(struct dma_pte *pte)
267 {
268 pte->val |= DMA_PTE_SNP;
269 }
270
271 static inline void dma_set_pte_prot(struct dma_pte *pte, unsigned long prot)
272 {
273 pte->val = (pte->val & ~3) | (prot & 3);
274 }
275
276 static inline u64 dma_pte_addr(struct dma_pte *pte)
277 {
278 #ifdef CONFIG_64BIT
279 return pte->val & VTD_PAGE_MASK;
280 #else
281 /* Must have a full atomic 64-bit read */
282 return __cmpxchg64(&pte->val, 0ULL, 0ULL) & VTD_PAGE_MASK;
283 #endif
284 }
285
286 static inline void dma_set_pte_pfn(struct dma_pte *pte, unsigned long pfn)
287 {
288 pte->val |= (uint64_t)pfn << VTD_PAGE_SHIFT;
289 }
290
291 static inline bool dma_pte_present(struct dma_pte *pte)
292 {
293 return (pte->val & 3) != 0;
294 }
295
296 static inline int first_pte_in_page(struct dma_pte *pte)
297 {
298 return !((unsigned long)pte & ~VTD_PAGE_MASK);
299 }
300
301 /*
302 * This domain is a statically identity mapping domain.
303 * 1. This domain creats a static 1:1 mapping to all usable memory.
304 * 2. It maps to each iommu if successful.
305 * 3. Each iommu mapps to this domain if successful.
306 */
307 static struct dmar_domain *si_domain;
308 static int hw_pass_through = 1;
309
310 /* devices under the same p2p bridge are owned in one domain */
311 #define DOMAIN_FLAG_P2P_MULTIPLE_DEVICES (1 << 0)
312
313 /* domain represents a virtual machine, more than one devices
314 * across iommus may be owned in one domain, e.g. kvm guest.
315 */
316 #define DOMAIN_FLAG_VIRTUAL_MACHINE (1 << 1)
317
318 /* si_domain contains mulitple devices */
319 #define DOMAIN_FLAG_STATIC_IDENTITY (1 << 2)
320
321 struct dmar_domain {
322 int id; /* domain id */
323 int nid; /* node id */
324 unsigned long iommu_bmp; /* bitmap of iommus this domain uses*/
325
326 struct list_head devices; /* all devices' list */
327 struct iova_domain iovad; /* iova's that belong to this domain */
328
329 struct dma_pte *pgd; /* virtual address */
330 int gaw; /* max guest address width */
331
332 /* adjusted guest address width, 0 is level 2 30-bit */
333 int agaw;
334
335 int flags; /* flags to find out type of domain */
336
337 int iommu_coherency;/* indicate coherency of iommu access */
338 int iommu_snooping; /* indicate snooping control feature*/
339 int iommu_count; /* reference count of iommu */
340 spinlock_t iommu_lock; /* protect iommu set in domain */
341 u64 max_addr; /* maximum mapped address */
342 };
343
344 /* PCI domain-device relationship */
345 struct device_domain_info {
346 struct list_head link; /* link to domain siblings */
347 struct list_head global; /* link to global list */
348 int segment; /* PCI domain */
349 u8 bus; /* PCI bus number */
350 u8 devfn; /* PCI devfn number */
351 struct pci_dev *dev; /* it's NULL for PCIe-to-PCI bridge */
352 struct intel_iommu *iommu; /* IOMMU used by this device */
353 struct dmar_domain *domain; /* pointer to domain */
354 };
355
356 static void flush_unmaps_timeout(unsigned long data);
357
358 DEFINE_TIMER(unmap_timer, flush_unmaps_timeout, 0, 0);
359
360 #define HIGH_WATER_MARK 250
361 struct deferred_flush_tables {
362 int next;
363 struct iova *iova[HIGH_WATER_MARK];
364 struct dmar_domain *domain[HIGH_WATER_MARK];
365 };
366
367 static struct deferred_flush_tables *deferred_flush;
368
369 /* bitmap for indexing intel_iommus */
370 static int g_num_of_iommus;
371
372 static DEFINE_SPINLOCK(async_umap_flush_lock);
373 static LIST_HEAD(unmaps_to_do);
374
375 static int timer_on;
376 static long list_size;
377
378 static void domain_remove_dev_info(struct dmar_domain *domain);
379
380 #ifdef CONFIG_DMAR_DEFAULT_ON
381 int dmar_disabled = 0;
382 #else
383 int dmar_disabled = 1;
384 #endif /*CONFIG_DMAR_DEFAULT_ON*/
385
386 static int dmar_map_gfx = 1;
387 static int dmar_forcedac;
388 static int intel_iommu_strict;
389
390 #define DUMMY_DEVICE_DOMAIN_INFO ((struct device_domain_info *)(-1))
391 static DEFINE_SPINLOCK(device_domain_lock);
392 static LIST_HEAD(device_domain_list);
393
394 static struct iommu_ops intel_iommu_ops;
395
396 static int __init intel_iommu_setup(char *str)
397 {
398 if (!str)
399 return -EINVAL;
400 while (*str) {
401 if (!strncmp(str, "on", 2)) {
402 dmar_disabled = 0;
403 printk(KERN_INFO "Intel-IOMMU: enabled\n");
404 } else if (!strncmp(str, "off", 3)) {
405 dmar_disabled = 1;
406 printk(KERN_INFO "Intel-IOMMU: disabled\n");
407 } else if (!strncmp(str, "igfx_off", 8)) {
408 dmar_map_gfx = 0;
409 printk(KERN_INFO
410 "Intel-IOMMU: disable GFX device mapping\n");
411 } else if (!strncmp(str, "forcedac", 8)) {
412 printk(KERN_INFO
413 "Intel-IOMMU: Forcing DAC for PCI devices\n");
414 dmar_forcedac = 1;
415 } else if (!strncmp(str, "strict", 6)) {
416 printk(KERN_INFO
417 "Intel-IOMMU: disable batched IOTLB flush\n");
418 intel_iommu_strict = 1;
419 }
420
421 str += strcspn(str, ",");
422 while (*str == ',')
423 str++;
424 }
425 return 0;
426 }
427 __setup("intel_iommu=", intel_iommu_setup);
428
429 static struct kmem_cache *iommu_domain_cache;
430 static struct kmem_cache *iommu_devinfo_cache;
431 static struct kmem_cache *iommu_iova_cache;
432
433 static inline void *alloc_pgtable_page(int node)
434 {
435 struct page *page;
436 void *vaddr = NULL;
437
438 page = alloc_pages_node(node, GFP_ATOMIC | __GFP_ZERO, 0);
439 if (page)
440 vaddr = page_address(page);
441 return vaddr;
442 }
443
444 static inline void free_pgtable_page(void *vaddr)
445 {
446 free_page((unsigned long)vaddr);
447 }
448
449 static inline void *alloc_domain_mem(void)
450 {
451 return kmem_cache_alloc(iommu_domain_cache, GFP_ATOMIC);
452 }
453
454 static void free_domain_mem(void *vaddr)
455 {
456 kmem_cache_free(iommu_domain_cache, vaddr);
457 }
458
459 static inline void * alloc_devinfo_mem(void)
460 {
461 return kmem_cache_alloc(iommu_devinfo_cache, GFP_ATOMIC);
462 }
463
464 static inline void free_devinfo_mem(void *vaddr)
465 {
466 kmem_cache_free(iommu_devinfo_cache, vaddr);
467 }
468
469 struct iova *alloc_iova_mem(void)
470 {
471 return kmem_cache_alloc(iommu_iova_cache, GFP_ATOMIC);
472 }
473
474 void free_iova_mem(struct iova *iova)
475 {
476 kmem_cache_free(iommu_iova_cache, iova);
477 }
478
479
480 static int __iommu_calculate_agaw(struct intel_iommu *iommu, int max_gaw)
481 {
482 unsigned long sagaw;
483 int agaw = -1;
484
485 sagaw = cap_sagaw(iommu->cap);
486 for (agaw = width_to_agaw(max_gaw);
487 agaw >= 0; agaw--) {
488 if (test_bit(agaw, &sagaw))
489 break;
490 }
491
492 return agaw;
493 }
494
495 /*
496 * Calculate max SAGAW for each iommu.
497 */
498 int iommu_calculate_max_sagaw(struct intel_iommu *iommu)
499 {
500 return __iommu_calculate_agaw(iommu, MAX_AGAW_WIDTH);
501 }
502
503 /*
504 * calculate agaw for each iommu.
505 * "SAGAW" may be different across iommus, use a default agaw, and
506 * get a supported less agaw for iommus that don't support the default agaw.
507 */
508 int iommu_calculate_agaw(struct intel_iommu *iommu)
509 {
510 return __iommu_calculate_agaw(iommu, DEFAULT_DOMAIN_ADDRESS_WIDTH);
511 }
512
513 /* This functionin only returns single iommu in a domain */
514 static struct intel_iommu *domain_get_iommu(struct dmar_domain *domain)
515 {
516 int iommu_id;
517
518 /* si_domain and vm domain should not get here. */
519 BUG_ON(domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE);
520 BUG_ON(domain->flags & DOMAIN_FLAG_STATIC_IDENTITY);
521
522 iommu_id = find_first_bit(&domain->iommu_bmp, g_num_of_iommus);
523 if (iommu_id < 0 || iommu_id >= g_num_of_iommus)
524 return NULL;
525
526 return g_iommus[iommu_id];
527 }
528
529 static void domain_update_iommu_coherency(struct dmar_domain *domain)
530 {
531 int i;
532
533 domain->iommu_coherency = 1;
534
535 for_each_set_bit(i, &domain->iommu_bmp, g_num_of_iommus) {
536 if (!ecap_coherent(g_iommus[i]->ecap)) {
537 domain->iommu_coherency = 0;
538 break;
539 }
540 }
541 }
542
543 static void domain_update_iommu_snooping(struct dmar_domain *domain)
544 {
545 int i;
546
547 domain->iommu_snooping = 1;
548
549 for_each_set_bit(i, &domain->iommu_bmp, g_num_of_iommus) {
550 if (!ecap_sc_support(g_iommus[i]->ecap)) {
551 domain->iommu_snooping = 0;
552 break;
553 }
554 }
555 }
556
557 /* Some capabilities may be different across iommus */
558 static void domain_update_iommu_cap(struct dmar_domain *domain)
559 {
560 domain_update_iommu_coherency(domain);
561 domain_update_iommu_snooping(domain);
562 }
563
564 static struct intel_iommu *device_to_iommu(int segment, u8 bus, u8 devfn)
565 {
566 struct dmar_drhd_unit *drhd = NULL;
567 int i;
568
569 for_each_drhd_unit(drhd) {
570 if (drhd->ignored)
571 continue;
572 if (segment != drhd->segment)
573 continue;
574
575 for (i = 0; i < drhd->devices_cnt; i++) {
576 if (drhd->devices[i] &&
577 drhd->devices[i]->bus->number == bus &&
578 drhd->devices[i]->devfn == devfn)
579 return drhd->iommu;
580 if (drhd->devices[i] &&
581 drhd->devices[i]->subordinate &&
582 drhd->devices[i]->subordinate->number <= bus &&
583 drhd->devices[i]->subordinate->subordinate >= bus)
584 return drhd->iommu;
585 }
586
587 if (drhd->include_all)
588 return drhd->iommu;
589 }
590
591 return NULL;
592 }
593
594 static void domain_flush_cache(struct dmar_domain *domain,
595 void *addr, int size)
596 {
597 if (!domain->iommu_coherency)
598 clflush_cache_range(addr, size);
599 }
600
601 /* Gets context entry for a given bus and devfn */
602 static struct context_entry * device_to_context_entry(struct intel_iommu *iommu,
603 u8 bus, u8 devfn)
604 {
605 struct root_entry *root;
606 struct context_entry *context;
607 unsigned long phy_addr;
608 unsigned long flags;
609
610 spin_lock_irqsave(&iommu->lock, flags);
611 root = &iommu->root_entry[bus];
612 context = get_context_addr_from_root(root);
613 if (!context) {
614 context = (struct context_entry *)
615 alloc_pgtable_page(iommu->node);
616 if (!context) {
617 spin_unlock_irqrestore(&iommu->lock, flags);
618 return NULL;
619 }
620 __iommu_flush_cache(iommu, (void *)context, CONTEXT_SIZE);
621 phy_addr = virt_to_phys((void *)context);
622 set_root_value(root, phy_addr);
623 set_root_present(root);
624 __iommu_flush_cache(iommu, root, sizeof(*root));
625 }
626 spin_unlock_irqrestore(&iommu->lock, flags);
627 return &context[devfn];
628 }
629
630 static int device_context_mapped(struct intel_iommu *iommu, u8 bus, u8 devfn)
631 {
632 struct root_entry *root;
633 struct context_entry *context;
634 int ret;
635 unsigned long flags;
636
637 spin_lock_irqsave(&iommu->lock, flags);
638 root = &iommu->root_entry[bus];
639 context = get_context_addr_from_root(root);
640 if (!context) {
641 ret = 0;
642 goto out;
643 }
644 ret = context_present(&context[devfn]);
645 out:
646 spin_unlock_irqrestore(&iommu->lock, flags);
647 return ret;
648 }
649
650 static void clear_context_table(struct intel_iommu *iommu, u8 bus, u8 devfn)
651 {
652 struct root_entry *root;
653 struct context_entry *context;
654 unsigned long flags;
655
656 spin_lock_irqsave(&iommu->lock, flags);
657 root = &iommu->root_entry[bus];
658 context = get_context_addr_from_root(root);
659 if (context) {
660 context_clear_entry(&context[devfn]);
661 __iommu_flush_cache(iommu, &context[devfn], \
662 sizeof(*context));
663 }
664 spin_unlock_irqrestore(&iommu->lock, flags);
665 }
666
667 static void free_context_table(struct intel_iommu *iommu)
668 {
669 struct root_entry *root;
670 int i;
671 unsigned long flags;
672 struct context_entry *context;
673
674 spin_lock_irqsave(&iommu->lock, flags);
675 if (!iommu->root_entry) {
676 goto out;
677 }
678 for (i = 0; i < ROOT_ENTRY_NR; i++) {
679 root = &iommu->root_entry[i];
680 context = get_context_addr_from_root(root);
681 if (context)
682 free_pgtable_page(context);
683 }
684 free_pgtable_page(iommu->root_entry);
685 iommu->root_entry = NULL;
686 out:
687 spin_unlock_irqrestore(&iommu->lock, flags);
688 }
689
690 static struct dma_pte *pfn_to_dma_pte(struct dmar_domain *domain,
691 unsigned long pfn)
692 {
693 int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
694 struct dma_pte *parent, *pte = NULL;
695 int level = agaw_to_level(domain->agaw);
696 int offset;
697
698 BUG_ON(!domain->pgd);
699 BUG_ON(addr_width < BITS_PER_LONG && pfn >> addr_width);
700 parent = domain->pgd;
701
702 while (level > 0) {
703 void *tmp_page;
704
705 offset = pfn_level_offset(pfn, level);
706 pte = &parent[offset];
707 if (level == 1)
708 break;
709
710 if (!dma_pte_present(pte)) {
711 uint64_t pteval;
712
713 tmp_page = alloc_pgtable_page(domain->nid);
714
715 if (!tmp_page)
716 return NULL;
717
718 domain_flush_cache(domain, tmp_page, VTD_PAGE_SIZE);
719 pteval = ((uint64_t)virt_to_dma_pfn(tmp_page) << VTD_PAGE_SHIFT) | DMA_PTE_READ | DMA_PTE_WRITE;
720 if (cmpxchg64(&pte->val, 0ULL, pteval)) {
721 /* Someone else set it while we were thinking; use theirs. */
722 free_pgtable_page(tmp_page);
723 } else {
724 dma_pte_addr(pte);
725 domain_flush_cache(domain, pte, sizeof(*pte));
726 }
727 }
728 parent = phys_to_virt(dma_pte_addr(pte));
729 level--;
730 }
731
732 return pte;
733 }
734
735 /* return address's pte at specific level */
736 static struct dma_pte *dma_pfn_level_pte(struct dmar_domain *domain,
737 unsigned long pfn,
738 int level)
739 {
740 struct dma_pte *parent, *pte = NULL;
741 int total = agaw_to_level(domain->agaw);
742 int offset;
743
744 parent = domain->pgd;
745 while (level <= total) {
746 offset = pfn_level_offset(pfn, total);
747 pte = &parent[offset];
748 if (level == total)
749 return pte;
750
751 if (!dma_pte_present(pte))
752 break;
753 parent = phys_to_virt(dma_pte_addr(pte));
754 total--;
755 }
756 return NULL;
757 }
758
759 /* clear last level pte, a tlb flush should be followed */
760 static void dma_pte_clear_range(struct dmar_domain *domain,
761 unsigned long start_pfn,
762 unsigned long last_pfn)
763 {
764 int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
765 struct dma_pte *first_pte, *pte;
766
767 BUG_ON(addr_width < BITS_PER_LONG && start_pfn >> addr_width);
768 BUG_ON(addr_width < BITS_PER_LONG && last_pfn >> addr_width);
769 BUG_ON(start_pfn > last_pfn);
770
771 /* we don't need lock here; nobody else touches the iova range */
772 do {
773 first_pte = pte = dma_pfn_level_pte(domain, start_pfn, 1);
774 if (!pte) {
775 start_pfn = align_to_level(start_pfn + 1, 2);
776 continue;
777 }
778 do {
779 dma_clear_pte(pte);
780 start_pfn++;
781 pte++;
782 } while (start_pfn <= last_pfn && !first_pte_in_page(pte));
783
784 domain_flush_cache(domain, first_pte,
785 (void *)pte - (void *)first_pte);
786
787 } while (start_pfn && start_pfn <= last_pfn);
788 }
789
790 /* free page table pages. last level pte should already be cleared */
791 static void dma_pte_free_pagetable(struct dmar_domain *domain,
792 unsigned long start_pfn,
793 unsigned long last_pfn)
794 {
795 int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
796 struct dma_pte *first_pte, *pte;
797 int total = agaw_to_level(domain->agaw);
798 int level;
799 unsigned long tmp;
800
801 BUG_ON(addr_width < BITS_PER_LONG && start_pfn >> addr_width);
802 BUG_ON(addr_width < BITS_PER_LONG && last_pfn >> addr_width);
803 BUG_ON(start_pfn > last_pfn);
804
805 /* We don't need lock here; nobody else touches the iova range */
806 level = 2;
807 while (level <= total) {
808 tmp = align_to_level(start_pfn, level);
809
810 /* If we can't even clear one PTE at this level, we're done */
811 if (tmp + level_size(level) - 1 > last_pfn)
812 return;
813
814 do {
815 first_pte = pte = dma_pfn_level_pte(domain, tmp, level);
816 if (!pte) {
817 tmp = align_to_level(tmp + 1, level + 1);
818 continue;
819 }
820 do {
821 if (dma_pte_present(pte)) {
822 free_pgtable_page(phys_to_virt(dma_pte_addr(pte)));
823 dma_clear_pte(pte);
824 }
825 pte++;
826 tmp += level_size(level);
827 } while (!first_pte_in_page(pte) &&
828 tmp + level_size(level) - 1 <= last_pfn);
829
830 domain_flush_cache(domain, first_pte,
831 (void *)pte - (void *)first_pte);
832
833 } while (tmp && tmp + level_size(level) - 1 <= last_pfn);
834 level++;
835 }
836 /* free pgd */
837 if (start_pfn == 0 && last_pfn == DOMAIN_MAX_PFN(domain->gaw)) {
838 free_pgtable_page(domain->pgd);
839 domain->pgd = NULL;
840 }
841 }
842
843 /* iommu handling */
844 static int iommu_alloc_root_entry(struct intel_iommu *iommu)
845 {
846 struct root_entry *root;
847 unsigned long flags;
848
849 root = (struct root_entry *)alloc_pgtable_page(iommu->node);
850 if (!root)
851 return -ENOMEM;
852
853 __iommu_flush_cache(iommu, root, ROOT_SIZE);
854
855 spin_lock_irqsave(&iommu->lock, flags);
856 iommu->root_entry = root;
857 spin_unlock_irqrestore(&iommu->lock, flags);
858
859 return 0;
860 }
861
862 static void iommu_set_root_entry(struct intel_iommu *iommu)
863 {
864 void *addr;
865 u32 sts;
866 unsigned long flag;
867
868 addr = iommu->root_entry;
869
870 spin_lock_irqsave(&iommu->register_lock, flag);
871 dmar_writeq(iommu->reg + DMAR_RTADDR_REG, virt_to_phys(addr));
872
873 writel(iommu->gcmd | DMA_GCMD_SRTP, iommu->reg + DMAR_GCMD_REG);
874
875 /* Make sure hardware complete it */
876 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
877 readl, (sts & DMA_GSTS_RTPS), sts);
878
879 spin_unlock_irqrestore(&iommu->register_lock, flag);
880 }
881
882 static void iommu_flush_write_buffer(struct intel_iommu *iommu)
883 {
884 u32 val;
885 unsigned long flag;
886
887 if (!rwbf_quirk && !cap_rwbf(iommu->cap))
888 return;
889
890 spin_lock_irqsave(&iommu->register_lock, flag);
891 writel(iommu->gcmd | DMA_GCMD_WBF, iommu->reg + DMAR_GCMD_REG);
892
893 /* Make sure hardware complete it */
894 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
895 readl, (!(val & DMA_GSTS_WBFS)), val);
896
897 spin_unlock_irqrestore(&iommu->register_lock, flag);
898 }
899
900 /* return value determine if we need a write buffer flush */
901 static void __iommu_flush_context(struct intel_iommu *iommu,
902 u16 did, u16 source_id, u8 function_mask,
903 u64 type)
904 {
905 u64 val = 0;
906 unsigned long flag;
907
908 switch (type) {
909 case DMA_CCMD_GLOBAL_INVL:
910 val = DMA_CCMD_GLOBAL_INVL;
911 break;
912 case DMA_CCMD_DOMAIN_INVL:
913 val = DMA_CCMD_DOMAIN_INVL|DMA_CCMD_DID(did);
914 break;
915 case DMA_CCMD_DEVICE_INVL:
916 val = DMA_CCMD_DEVICE_INVL|DMA_CCMD_DID(did)
917 | DMA_CCMD_SID(source_id) | DMA_CCMD_FM(function_mask);
918 break;
919 default:
920 BUG();
921 }
922 val |= DMA_CCMD_ICC;
923
924 spin_lock_irqsave(&iommu->register_lock, flag);
925 dmar_writeq(iommu->reg + DMAR_CCMD_REG, val);
926
927 /* Make sure hardware complete it */
928 IOMMU_WAIT_OP(iommu, DMAR_CCMD_REG,
929 dmar_readq, (!(val & DMA_CCMD_ICC)), val);
930
931 spin_unlock_irqrestore(&iommu->register_lock, flag);
932 }
933
934 /* return value determine if we need a write buffer flush */
935 static void __iommu_flush_iotlb(struct intel_iommu *iommu, u16 did,
936 u64 addr, unsigned int size_order, u64 type)
937 {
938 int tlb_offset = ecap_iotlb_offset(iommu->ecap);
939 u64 val = 0, val_iva = 0;
940 unsigned long flag;
941
942 switch (type) {
943 case DMA_TLB_GLOBAL_FLUSH:
944 /* global flush doesn't need set IVA_REG */
945 val = DMA_TLB_GLOBAL_FLUSH|DMA_TLB_IVT;
946 break;
947 case DMA_TLB_DSI_FLUSH:
948 val = DMA_TLB_DSI_FLUSH|DMA_TLB_IVT|DMA_TLB_DID(did);
949 break;
950 case DMA_TLB_PSI_FLUSH:
951 val = DMA_TLB_PSI_FLUSH|DMA_TLB_IVT|DMA_TLB_DID(did);
952 /* Note: always flush non-leaf currently */
953 val_iva = size_order | addr;
954 break;
955 default:
956 BUG();
957 }
958 /* Note: set drain read/write */
959 #if 0
960 /*
961 * This is probably to be super secure.. Looks like we can
962 * ignore it without any impact.
963 */
964 if (cap_read_drain(iommu->cap))
965 val |= DMA_TLB_READ_DRAIN;
966 #endif
967 if (cap_write_drain(iommu->cap))
968 val |= DMA_TLB_WRITE_DRAIN;
969
970 spin_lock_irqsave(&iommu->register_lock, flag);
971 /* Note: Only uses first TLB reg currently */
972 if (val_iva)
973 dmar_writeq(iommu->reg + tlb_offset, val_iva);
974 dmar_writeq(iommu->reg + tlb_offset + 8, val);
975
976 /* Make sure hardware complete it */
977 IOMMU_WAIT_OP(iommu, tlb_offset + 8,
978 dmar_readq, (!(val & DMA_TLB_IVT)), val);
979
980 spin_unlock_irqrestore(&iommu->register_lock, flag);
981
982 /* check IOTLB invalidation granularity */
983 if (DMA_TLB_IAIG(val) == 0)
984 printk(KERN_ERR"IOMMU: flush IOTLB failed\n");
985 if (DMA_TLB_IAIG(val) != DMA_TLB_IIRG(type))
986 pr_debug("IOMMU: tlb flush request %Lx, actual %Lx\n",
987 (unsigned long long)DMA_TLB_IIRG(type),
988 (unsigned long long)DMA_TLB_IAIG(val));
989 }
990
991 static struct device_domain_info *iommu_support_dev_iotlb(
992 struct dmar_domain *domain, int segment, u8 bus, u8 devfn)
993 {
994 int found = 0;
995 unsigned long flags;
996 struct device_domain_info *info;
997 struct intel_iommu *iommu = device_to_iommu(segment, bus, devfn);
998
999 if (!ecap_dev_iotlb_support(iommu->ecap))
1000 return NULL;
1001
1002 if (!iommu->qi)
1003 return NULL;
1004
1005 spin_lock_irqsave(&device_domain_lock, flags);
1006 list_for_each_entry(info, &domain->devices, link)
1007 if (info->bus == bus && info->devfn == devfn) {
1008 found = 1;
1009 break;
1010 }
1011 spin_unlock_irqrestore(&device_domain_lock, flags);
1012
1013 if (!found || !info->dev)
1014 return NULL;
1015
1016 if (!pci_find_ext_capability(info->dev, PCI_EXT_CAP_ID_ATS))
1017 return NULL;
1018
1019 if (!dmar_find_matched_atsr_unit(info->dev))
1020 return NULL;
1021
1022 info->iommu = iommu;
1023
1024 return info;
1025 }
1026
1027 static void iommu_enable_dev_iotlb(struct device_domain_info *info)
1028 {
1029 if (!info)
1030 return;
1031
1032 pci_enable_ats(info->dev, VTD_PAGE_SHIFT);
1033 }
1034
1035 static void iommu_disable_dev_iotlb(struct device_domain_info *info)
1036 {
1037 if (!info->dev || !pci_ats_enabled(info->dev))
1038 return;
1039
1040 pci_disable_ats(info->dev);
1041 }
1042
1043 static void iommu_flush_dev_iotlb(struct dmar_domain *domain,
1044 u64 addr, unsigned mask)
1045 {
1046 u16 sid, qdep;
1047 unsigned long flags;
1048 struct device_domain_info *info;
1049
1050 spin_lock_irqsave(&device_domain_lock, flags);
1051 list_for_each_entry(info, &domain->devices, link) {
1052 if (!info->dev || !pci_ats_enabled(info->dev))
1053 continue;
1054
1055 sid = info->bus << 8 | info->devfn;
1056 qdep = pci_ats_queue_depth(info->dev);
1057 qi_flush_dev_iotlb(info->iommu, sid, qdep, addr, mask);
1058 }
1059 spin_unlock_irqrestore(&device_domain_lock, flags);
1060 }
1061
1062 static void iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did,
1063 unsigned long pfn, unsigned int pages, int map)
1064 {
1065 unsigned int mask = ilog2(__roundup_pow_of_two(pages));
1066 uint64_t addr = (uint64_t)pfn << VTD_PAGE_SHIFT;
1067
1068 BUG_ON(pages == 0);
1069
1070 /*
1071 * Fallback to domain selective flush if no PSI support or the size is
1072 * too big.
1073 * PSI requires page size to be 2 ^ x, and the base address is naturally
1074 * aligned to the size
1075 */
1076 if (!cap_pgsel_inv(iommu->cap) || mask > cap_max_amask_val(iommu->cap))
1077 iommu->flush.flush_iotlb(iommu, did, 0, 0,
1078 DMA_TLB_DSI_FLUSH);
1079 else
1080 iommu->flush.flush_iotlb(iommu, did, addr, mask,
1081 DMA_TLB_PSI_FLUSH);
1082
1083 /*
1084 * In caching mode, changes of pages from non-present to present require
1085 * flush. However, device IOTLB doesn't need to be flushed in this case.
1086 */
1087 if (!cap_caching_mode(iommu->cap) || !map)
1088 iommu_flush_dev_iotlb(iommu->domains[did], addr, mask);
1089 }
1090
1091 static void iommu_disable_protect_mem_regions(struct intel_iommu *iommu)
1092 {
1093 u32 pmen;
1094 unsigned long flags;
1095
1096 spin_lock_irqsave(&iommu->register_lock, flags);
1097 pmen = readl(iommu->reg + DMAR_PMEN_REG);
1098 pmen &= ~DMA_PMEN_EPM;
1099 writel(pmen, iommu->reg + DMAR_PMEN_REG);
1100
1101 /* wait for the protected region status bit to clear */
1102 IOMMU_WAIT_OP(iommu, DMAR_PMEN_REG,
1103 readl, !(pmen & DMA_PMEN_PRS), pmen);
1104
1105 spin_unlock_irqrestore(&iommu->register_lock, flags);
1106 }
1107
1108 static int iommu_enable_translation(struct intel_iommu *iommu)
1109 {
1110 u32 sts;
1111 unsigned long flags;
1112
1113 spin_lock_irqsave(&iommu->register_lock, flags);
1114 iommu->gcmd |= DMA_GCMD_TE;
1115 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1116
1117 /* Make sure hardware complete it */
1118 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
1119 readl, (sts & DMA_GSTS_TES), sts);
1120
1121 spin_unlock_irqrestore(&iommu->register_lock, flags);
1122 return 0;
1123 }
1124
1125 static int iommu_disable_translation(struct intel_iommu *iommu)
1126 {
1127 u32 sts;
1128 unsigned long flag;
1129
1130 spin_lock_irqsave(&iommu->register_lock, flag);
1131 iommu->gcmd &= ~DMA_GCMD_TE;
1132 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1133
1134 /* Make sure hardware complete it */
1135 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
1136 readl, (!(sts & DMA_GSTS_TES)), sts);
1137
1138 spin_unlock_irqrestore(&iommu->register_lock, flag);
1139 return 0;
1140 }
1141
1142
1143 static int iommu_init_domains(struct intel_iommu *iommu)
1144 {
1145 unsigned long ndomains;
1146 unsigned long nlongs;
1147
1148 ndomains = cap_ndoms(iommu->cap);
1149 pr_debug("IOMMU %d: Number of Domains supportd <%ld>\n", iommu->seq_id,
1150 ndomains);
1151 nlongs = BITS_TO_LONGS(ndomains);
1152
1153 spin_lock_init(&iommu->lock);
1154
1155 /* TBD: there might be 64K domains,
1156 * consider other allocation for future chip
1157 */
1158 iommu->domain_ids = kcalloc(nlongs, sizeof(unsigned long), GFP_KERNEL);
1159 if (!iommu->domain_ids) {
1160 printk(KERN_ERR "Allocating domain id array failed\n");
1161 return -ENOMEM;
1162 }
1163 iommu->domains = kcalloc(ndomains, sizeof(struct dmar_domain *),
1164 GFP_KERNEL);
1165 if (!iommu->domains) {
1166 printk(KERN_ERR "Allocating domain array failed\n");
1167 return -ENOMEM;
1168 }
1169
1170 /*
1171 * if Caching mode is set, then invalid translations are tagged
1172 * with domainid 0. Hence we need to pre-allocate it.
1173 */
1174 if (cap_caching_mode(iommu->cap))
1175 set_bit(0, iommu->domain_ids);
1176 return 0;
1177 }
1178
1179
1180 static void domain_exit(struct dmar_domain *domain);
1181 static void vm_domain_exit(struct dmar_domain *domain);
1182
1183 void free_dmar_iommu(struct intel_iommu *iommu)
1184 {
1185 struct dmar_domain *domain;
1186 int i;
1187 unsigned long flags;
1188
1189 if ((iommu->domains) && (iommu->domain_ids)) {
1190 for_each_set_bit(i, iommu->domain_ids, cap_ndoms(iommu->cap)) {
1191 domain = iommu->domains[i];
1192 clear_bit(i, iommu->domain_ids);
1193
1194 spin_lock_irqsave(&domain->iommu_lock, flags);
1195 if (--domain->iommu_count == 0) {
1196 if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
1197 vm_domain_exit(domain);
1198 else
1199 domain_exit(domain);
1200 }
1201 spin_unlock_irqrestore(&domain->iommu_lock, flags);
1202 }
1203 }
1204
1205 if (iommu->gcmd & DMA_GCMD_TE)
1206 iommu_disable_translation(iommu);
1207
1208 if (iommu->irq) {
1209 set_irq_data(iommu->irq, NULL);
1210 /* This will mask the irq */
1211 free_irq(iommu->irq, iommu);
1212 destroy_irq(iommu->irq);
1213 }
1214
1215 kfree(iommu->domains);
1216 kfree(iommu->domain_ids);
1217
1218 g_iommus[iommu->seq_id] = NULL;
1219
1220 /* if all iommus are freed, free g_iommus */
1221 for (i = 0; i < g_num_of_iommus; i++) {
1222 if (g_iommus[i])
1223 break;
1224 }
1225
1226 if (i == g_num_of_iommus)
1227 kfree(g_iommus);
1228
1229 /* free context mapping */
1230 free_context_table(iommu);
1231 }
1232
1233 static struct dmar_domain *alloc_domain(void)
1234 {
1235 struct dmar_domain *domain;
1236
1237 domain = alloc_domain_mem();
1238 if (!domain)
1239 return NULL;
1240
1241 domain->nid = -1;
1242 memset(&domain->iommu_bmp, 0, sizeof(unsigned long));
1243 domain->flags = 0;
1244
1245 return domain;
1246 }
1247
1248 static int iommu_attach_domain(struct dmar_domain *domain,
1249 struct intel_iommu *iommu)
1250 {
1251 int num;
1252 unsigned long ndomains;
1253 unsigned long flags;
1254
1255 ndomains = cap_ndoms(iommu->cap);
1256
1257 spin_lock_irqsave(&iommu->lock, flags);
1258
1259 num = find_first_zero_bit(iommu->domain_ids, ndomains);
1260 if (num >= ndomains) {
1261 spin_unlock_irqrestore(&iommu->lock, flags);
1262 printk(KERN_ERR "IOMMU: no free domain ids\n");
1263 return -ENOMEM;
1264 }
1265
1266 domain->id = num;
1267 set_bit(num, iommu->domain_ids);
1268 set_bit(iommu->seq_id, &domain->iommu_bmp);
1269 iommu->domains[num] = domain;
1270 spin_unlock_irqrestore(&iommu->lock, flags);
1271
1272 return 0;
1273 }
1274
1275 static void iommu_detach_domain(struct dmar_domain *domain,
1276 struct intel_iommu *iommu)
1277 {
1278 unsigned long flags;
1279 int num, ndomains;
1280 int found = 0;
1281
1282 spin_lock_irqsave(&iommu->lock, flags);
1283 ndomains = cap_ndoms(iommu->cap);
1284 for_each_set_bit(num, iommu->domain_ids, ndomains) {
1285 if (iommu->domains[num] == domain) {
1286 found = 1;
1287 break;
1288 }
1289 }
1290
1291 if (found) {
1292 clear_bit(num, iommu->domain_ids);
1293 clear_bit(iommu->seq_id, &domain->iommu_bmp);
1294 iommu->domains[num] = NULL;
1295 }
1296 spin_unlock_irqrestore(&iommu->lock, flags);
1297 }
1298
1299 static struct iova_domain reserved_iova_list;
1300 static struct lock_class_key reserved_rbtree_key;
1301
1302 static void dmar_init_reserved_ranges(void)
1303 {
1304 struct pci_dev *pdev = NULL;
1305 struct iova *iova;
1306 int i;
1307
1308 init_iova_domain(&reserved_iova_list, DMA_32BIT_PFN);
1309
1310 lockdep_set_class(&reserved_iova_list.iova_rbtree_lock,
1311 &reserved_rbtree_key);
1312
1313 /* IOAPIC ranges shouldn't be accessed by DMA */
1314 iova = reserve_iova(&reserved_iova_list, IOVA_PFN(IOAPIC_RANGE_START),
1315 IOVA_PFN(IOAPIC_RANGE_END));
1316 if (!iova)
1317 printk(KERN_ERR "Reserve IOAPIC range failed\n");
1318
1319 /* Reserve all PCI MMIO to avoid peer-to-peer access */
1320 for_each_pci_dev(pdev) {
1321 struct resource *r;
1322
1323 for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1324 r = &pdev->resource[i];
1325 if (!r->flags || !(r->flags & IORESOURCE_MEM))
1326 continue;
1327 iova = reserve_iova(&reserved_iova_list,
1328 IOVA_PFN(r->start),
1329 IOVA_PFN(r->end));
1330 if (!iova)
1331 printk(KERN_ERR "Reserve iova failed\n");
1332 }
1333 }
1334
1335 }
1336
1337 static void domain_reserve_special_ranges(struct dmar_domain *domain)
1338 {
1339 copy_reserved_iova(&reserved_iova_list, &domain->iovad);
1340 }
1341
1342 static inline int guestwidth_to_adjustwidth(int gaw)
1343 {
1344 int agaw;
1345 int r = (gaw - 12) % 9;
1346
1347 if (r == 0)
1348 agaw = gaw;
1349 else
1350 agaw = gaw + 9 - r;
1351 if (agaw > 64)
1352 agaw = 64;
1353 return agaw;
1354 }
1355
1356 static int domain_init(struct dmar_domain *domain, int guest_width)
1357 {
1358 struct intel_iommu *iommu;
1359 int adjust_width, agaw;
1360 unsigned long sagaw;
1361
1362 init_iova_domain(&domain->iovad, DMA_32BIT_PFN);
1363 spin_lock_init(&domain->iommu_lock);
1364
1365 domain_reserve_special_ranges(domain);
1366
1367 /* calculate AGAW */
1368 iommu = domain_get_iommu(domain);
1369 if (guest_width > cap_mgaw(iommu->cap))
1370 guest_width = cap_mgaw(iommu->cap);
1371 domain->gaw = guest_width;
1372 adjust_width = guestwidth_to_adjustwidth(guest_width);
1373 agaw = width_to_agaw(adjust_width);
1374 sagaw = cap_sagaw(iommu->cap);
1375 if (!test_bit(agaw, &sagaw)) {
1376 /* hardware doesn't support it, choose a bigger one */
1377 pr_debug("IOMMU: hardware doesn't support agaw %d\n", agaw);
1378 agaw = find_next_bit(&sagaw, 5, agaw);
1379 if (agaw >= 5)
1380 return -ENODEV;
1381 }
1382 domain->agaw = agaw;
1383 INIT_LIST_HEAD(&domain->devices);
1384
1385 if (ecap_coherent(iommu->ecap))
1386 domain->iommu_coherency = 1;
1387 else
1388 domain->iommu_coherency = 0;
1389
1390 if (ecap_sc_support(iommu->ecap))
1391 domain->iommu_snooping = 1;
1392 else
1393 domain->iommu_snooping = 0;
1394
1395 domain->iommu_count = 1;
1396 domain->nid = iommu->node;
1397
1398 /* always allocate the top pgd */
1399 domain->pgd = (struct dma_pte *)alloc_pgtable_page(domain->nid);
1400 if (!domain->pgd)
1401 return -ENOMEM;
1402 __iommu_flush_cache(iommu, domain->pgd, PAGE_SIZE);
1403 return 0;
1404 }
1405
1406 static void domain_exit(struct dmar_domain *domain)
1407 {
1408 struct dmar_drhd_unit *drhd;
1409 struct intel_iommu *iommu;
1410
1411 /* Domain 0 is reserved, so dont process it */
1412 if (!domain)
1413 return;
1414
1415 domain_remove_dev_info(domain);
1416 /* destroy iovas */
1417 put_iova_domain(&domain->iovad);
1418
1419 /* clear ptes */
1420 dma_pte_clear_range(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
1421
1422 /* free page tables */
1423 dma_pte_free_pagetable(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
1424
1425 for_each_active_iommu(iommu, drhd)
1426 if (test_bit(iommu->seq_id, &domain->iommu_bmp))
1427 iommu_detach_domain(domain, iommu);
1428
1429 free_domain_mem(domain);
1430 }
1431
1432 static int domain_context_mapping_one(struct dmar_domain *domain, int segment,
1433 u8 bus, u8 devfn, int translation)
1434 {
1435 struct context_entry *context;
1436 unsigned long flags;
1437 struct intel_iommu *iommu;
1438 struct dma_pte *pgd;
1439 unsigned long num;
1440 unsigned long ndomains;
1441 int id;
1442 int agaw;
1443 struct device_domain_info *info = NULL;
1444
1445 pr_debug("Set context mapping for %02x:%02x.%d\n",
1446 bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
1447
1448 BUG_ON(!domain->pgd);
1449 BUG_ON(translation != CONTEXT_TT_PASS_THROUGH &&
1450 translation != CONTEXT_TT_MULTI_LEVEL);
1451
1452 iommu = device_to_iommu(segment, bus, devfn);
1453 if (!iommu)
1454 return -ENODEV;
1455
1456 context = device_to_context_entry(iommu, bus, devfn);
1457 if (!context)
1458 return -ENOMEM;
1459 spin_lock_irqsave(&iommu->lock, flags);
1460 if (context_present(context)) {
1461 spin_unlock_irqrestore(&iommu->lock, flags);
1462 return 0;
1463 }
1464
1465 id = domain->id;
1466 pgd = domain->pgd;
1467
1468 if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE ||
1469 domain->flags & DOMAIN_FLAG_STATIC_IDENTITY) {
1470 int found = 0;
1471
1472 /* find an available domain id for this device in iommu */
1473 ndomains = cap_ndoms(iommu->cap);
1474 for_each_set_bit(num, iommu->domain_ids, ndomains) {
1475 if (iommu->domains[num] == domain) {
1476 id = num;
1477 found = 1;
1478 break;
1479 }
1480 }
1481
1482 if (found == 0) {
1483 num = find_first_zero_bit(iommu->domain_ids, ndomains);
1484 if (num >= ndomains) {
1485 spin_unlock_irqrestore(&iommu->lock, flags);
1486 printk(KERN_ERR "IOMMU: no free domain ids\n");
1487 return -EFAULT;
1488 }
1489
1490 set_bit(num, iommu->domain_ids);
1491 iommu->domains[num] = domain;
1492 id = num;
1493 }
1494
1495 /* Skip top levels of page tables for
1496 * iommu which has less agaw than default.
1497 * Unnecessary for PT mode.
1498 */
1499 if (translation != CONTEXT_TT_PASS_THROUGH) {
1500 for (agaw = domain->agaw; agaw != iommu->agaw; agaw--) {
1501 pgd = phys_to_virt(dma_pte_addr(pgd));
1502 if (!dma_pte_present(pgd)) {
1503 spin_unlock_irqrestore(&iommu->lock, flags);
1504 return -ENOMEM;
1505 }
1506 }
1507 }
1508 }
1509
1510 context_set_domain_id(context, id);
1511
1512 if (translation != CONTEXT_TT_PASS_THROUGH) {
1513 info = iommu_support_dev_iotlb(domain, segment, bus, devfn);
1514 translation = info ? CONTEXT_TT_DEV_IOTLB :
1515 CONTEXT_TT_MULTI_LEVEL;
1516 }
1517 /*
1518 * In pass through mode, AW must be programmed to indicate the largest
1519 * AGAW value supported by hardware. And ASR is ignored by hardware.
1520 */
1521 if (unlikely(translation == CONTEXT_TT_PASS_THROUGH))
1522 context_set_address_width(context, iommu->msagaw);
1523 else {
1524 context_set_address_root(context, virt_to_phys(pgd));
1525 context_set_address_width(context, iommu->agaw);
1526 }
1527
1528 context_set_translation_type(context, translation);
1529 context_set_fault_enable(context);
1530 context_set_present(context);
1531 domain_flush_cache(domain, context, sizeof(*context));
1532
1533 /*
1534 * It's a non-present to present mapping. If hardware doesn't cache
1535 * non-present entry we only need to flush the write-buffer. If the
1536 * _does_ cache non-present entries, then it does so in the special
1537 * domain #0, which we have to flush:
1538 */
1539 if (cap_caching_mode(iommu->cap)) {
1540 iommu->flush.flush_context(iommu, 0,
1541 (((u16)bus) << 8) | devfn,
1542 DMA_CCMD_MASK_NOBIT,
1543 DMA_CCMD_DEVICE_INVL);
1544 iommu->flush.flush_iotlb(iommu, domain->id, 0, 0, DMA_TLB_DSI_FLUSH);
1545 } else {
1546 iommu_flush_write_buffer(iommu);
1547 }
1548 iommu_enable_dev_iotlb(info);
1549 spin_unlock_irqrestore(&iommu->lock, flags);
1550
1551 spin_lock_irqsave(&domain->iommu_lock, flags);
1552 if (!test_and_set_bit(iommu->seq_id, &domain->iommu_bmp)) {
1553 domain->iommu_count++;
1554 if (domain->iommu_count == 1)
1555 domain->nid = iommu->node;
1556 domain_update_iommu_cap(domain);
1557 }
1558 spin_unlock_irqrestore(&domain->iommu_lock, flags);
1559 return 0;
1560 }
1561
1562 static int
1563 domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev,
1564 int translation)
1565 {
1566 int ret;
1567 struct pci_dev *tmp, *parent;
1568
1569 ret = domain_context_mapping_one(domain, pci_domain_nr(pdev->bus),
1570 pdev->bus->number, pdev->devfn,
1571 translation);
1572 if (ret)
1573 return ret;
1574
1575 /* dependent device mapping */
1576 tmp = pci_find_upstream_pcie_bridge(pdev);
1577 if (!tmp)
1578 return 0;
1579 /* Secondary interface's bus number and devfn 0 */
1580 parent = pdev->bus->self;
1581 while (parent != tmp) {
1582 ret = domain_context_mapping_one(domain,
1583 pci_domain_nr(parent->bus),
1584 parent->bus->number,
1585 parent->devfn, translation);
1586 if (ret)
1587 return ret;
1588 parent = parent->bus->self;
1589 }
1590 if (pci_is_pcie(tmp)) /* this is a PCIe-to-PCI bridge */
1591 return domain_context_mapping_one(domain,
1592 pci_domain_nr(tmp->subordinate),
1593 tmp->subordinate->number, 0,
1594 translation);
1595 else /* this is a legacy PCI bridge */
1596 return domain_context_mapping_one(domain,
1597 pci_domain_nr(tmp->bus),
1598 tmp->bus->number,
1599 tmp->devfn,
1600 translation);
1601 }
1602
1603 static int domain_context_mapped(struct pci_dev *pdev)
1604 {
1605 int ret;
1606 struct pci_dev *tmp, *parent;
1607 struct intel_iommu *iommu;
1608
1609 iommu = device_to_iommu(pci_domain_nr(pdev->bus), pdev->bus->number,
1610 pdev->devfn);
1611 if (!iommu)
1612 return -ENODEV;
1613
1614 ret = device_context_mapped(iommu, pdev->bus->number, pdev->devfn);
1615 if (!ret)
1616 return ret;
1617 /* dependent device mapping */
1618 tmp = pci_find_upstream_pcie_bridge(pdev);
1619 if (!tmp)
1620 return ret;
1621 /* Secondary interface's bus number and devfn 0 */
1622 parent = pdev->bus->self;
1623 while (parent != tmp) {
1624 ret = device_context_mapped(iommu, parent->bus->number,
1625 parent->devfn);
1626 if (!ret)
1627 return ret;
1628 parent = parent->bus->self;
1629 }
1630 if (pci_is_pcie(tmp))
1631 return device_context_mapped(iommu, tmp->subordinate->number,
1632 0);
1633 else
1634 return device_context_mapped(iommu, tmp->bus->number,
1635 tmp->devfn);
1636 }
1637
1638 /* Returns a number of VTD pages, but aligned to MM page size */
1639 static inline unsigned long aligned_nrpages(unsigned long host_addr,
1640 size_t size)
1641 {
1642 host_addr &= ~PAGE_MASK;
1643 return PAGE_ALIGN(host_addr + size) >> VTD_PAGE_SHIFT;
1644 }
1645
1646 static int __domain_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
1647 struct scatterlist *sg, unsigned long phys_pfn,
1648 unsigned long nr_pages, int prot)
1649 {
1650 struct dma_pte *first_pte = NULL, *pte = NULL;
1651 phys_addr_t uninitialized_var(pteval);
1652 int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
1653 unsigned long sg_res;
1654
1655 BUG_ON(addr_width < BITS_PER_LONG && (iov_pfn + nr_pages - 1) >> addr_width);
1656
1657 if ((prot & (DMA_PTE_READ|DMA_PTE_WRITE)) == 0)
1658 return -EINVAL;
1659
1660 prot &= DMA_PTE_READ | DMA_PTE_WRITE | DMA_PTE_SNP;
1661
1662 if (sg)
1663 sg_res = 0;
1664 else {
1665 sg_res = nr_pages + 1;
1666 pteval = ((phys_addr_t)phys_pfn << VTD_PAGE_SHIFT) | prot;
1667 }
1668
1669 while (nr_pages--) {
1670 uint64_t tmp;
1671
1672 if (!sg_res) {
1673 sg_res = aligned_nrpages(sg->offset, sg->length);
1674 sg->dma_address = ((dma_addr_t)iov_pfn << VTD_PAGE_SHIFT) + sg->offset;
1675 sg->dma_length = sg->length;
1676 pteval = page_to_phys(sg_page(sg)) | prot;
1677 }
1678 if (!pte) {
1679 first_pte = pte = pfn_to_dma_pte(domain, iov_pfn);
1680 if (!pte)
1681 return -ENOMEM;
1682 }
1683 /* We don't need lock here, nobody else
1684 * touches the iova range
1685 */
1686 tmp = cmpxchg64_local(&pte->val, 0ULL, pteval);
1687 if (tmp) {
1688 static int dumps = 5;
1689 printk(KERN_CRIT "ERROR: DMA PTE for vPFN 0x%lx already set (to %llx not %llx)\n",
1690 iov_pfn, tmp, (unsigned long long)pteval);
1691 if (dumps) {
1692 dumps--;
1693 debug_dma_dump_mappings(NULL);
1694 }
1695 WARN_ON(1);
1696 }
1697 pte++;
1698 if (!nr_pages || first_pte_in_page(pte)) {
1699 domain_flush_cache(domain, first_pte,
1700 (void *)pte - (void *)first_pte);
1701 pte = NULL;
1702 }
1703 iov_pfn++;
1704 pteval += VTD_PAGE_SIZE;
1705 sg_res--;
1706 if (!sg_res)
1707 sg = sg_next(sg);
1708 }
1709 return 0;
1710 }
1711
1712 static inline int domain_sg_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
1713 struct scatterlist *sg, unsigned long nr_pages,
1714 int prot)
1715 {
1716 return __domain_mapping(domain, iov_pfn, sg, 0, nr_pages, prot);
1717 }
1718
1719 static inline int domain_pfn_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
1720 unsigned long phys_pfn, unsigned long nr_pages,
1721 int prot)
1722 {
1723 return __domain_mapping(domain, iov_pfn, NULL, phys_pfn, nr_pages, prot);
1724 }
1725
1726 static void iommu_detach_dev(struct intel_iommu *iommu, u8 bus, u8 devfn)
1727 {
1728 if (!iommu)
1729 return;
1730
1731 clear_context_table(iommu, bus, devfn);
1732 iommu->flush.flush_context(iommu, 0, 0, 0,
1733 DMA_CCMD_GLOBAL_INVL);
1734 iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH);
1735 }
1736
1737 static void domain_remove_dev_info(struct dmar_domain *domain)
1738 {
1739 struct device_domain_info *info;
1740 unsigned long flags;
1741 struct intel_iommu *iommu;
1742
1743 spin_lock_irqsave(&device_domain_lock, flags);
1744 while (!list_empty(&domain->devices)) {
1745 info = list_entry(domain->devices.next,
1746 struct device_domain_info, link);
1747 list_del(&info->link);
1748 list_del(&info->global);
1749 if (info->dev)
1750 info->dev->dev.archdata.iommu = NULL;
1751 spin_unlock_irqrestore(&device_domain_lock, flags);
1752
1753 iommu_disable_dev_iotlb(info);
1754 iommu = device_to_iommu(info->segment, info->bus, info->devfn);
1755 iommu_detach_dev(iommu, info->bus, info->devfn);
1756 free_devinfo_mem(info);
1757
1758 spin_lock_irqsave(&device_domain_lock, flags);
1759 }
1760 spin_unlock_irqrestore(&device_domain_lock, flags);
1761 }
1762
1763 /*
1764 * find_domain
1765 * Note: we use struct pci_dev->dev.archdata.iommu stores the info
1766 */
1767 static struct dmar_domain *
1768 find_domain(struct pci_dev *pdev)
1769 {
1770 struct device_domain_info *info;
1771
1772 /* No lock here, assumes no domain exit in normal case */
1773 info = pdev->dev.archdata.iommu;
1774 if (info)
1775 return info->domain;
1776 return NULL;
1777 }
1778
1779 /* domain is initialized */
1780 static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw)
1781 {
1782 struct dmar_domain *domain, *found = NULL;
1783 struct intel_iommu *iommu;
1784 struct dmar_drhd_unit *drhd;
1785 struct device_domain_info *info, *tmp;
1786 struct pci_dev *dev_tmp;
1787 unsigned long flags;
1788 int bus = 0, devfn = 0;
1789 int segment;
1790 int ret;
1791
1792 domain = find_domain(pdev);
1793 if (domain)
1794 return domain;
1795
1796 segment = pci_domain_nr(pdev->bus);
1797
1798 dev_tmp = pci_find_upstream_pcie_bridge(pdev);
1799 if (dev_tmp) {
1800 if (pci_is_pcie(dev_tmp)) {
1801 bus = dev_tmp->subordinate->number;
1802 devfn = 0;
1803 } else {
1804 bus = dev_tmp->bus->number;
1805 devfn = dev_tmp->devfn;
1806 }
1807 spin_lock_irqsave(&device_domain_lock, flags);
1808 list_for_each_entry(info, &device_domain_list, global) {
1809 if (info->segment == segment &&
1810 info->bus == bus && info->devfn == devfn) {
1811 found = info->domain;
1812 break;
1813 }
1814 }
1815 spin_unlock_irqrestore(&device_domain_lock, flags);
1816 /* pcie-pci bridge already has a domain, uses it */
1817 if (found) {
1818 domain = found;
1819 goto found_domain;
1820 }
1821 }
1822
1823 domain = alloc_domain();
1824 if (!domain)
1825 goto error;
1826
1827 /* Allocate new domain for the device */
1828 drhd = dmar_find_matched_drhd_unit(pdev);
1829 if (!drhd) {
1830 printk(KERN_ERR "IOMMU: can't find DMAR for device %s\n",
1831 pci_name(pdev));
1832 return NULL;
1833 }
1834 iommu = drhd->iommu;
1835
1836 ret = iommu_attach_domain(domain, iommu);
1837 if (ret) {
1838 domain_exit(domain);
1839 goto error;
1840 }
1841
1842 if (domain_init(domain, gaw)) {
1843 domain_exit(domain);
1844 goto error;
1845 }
1846
1847 /* register pcie-to-pci device */
1848 if (dev_tmp) {
1849 info = alloc_devinfo_mem();
1850 if (!info) {
1851 domain_exit(domain);
1852 goto error;
1853 }
1854 info->segment = segment;
1855 info->bus = bus;
1856 info->devfn = devfn;
1857 info->dev = NULL;
1858 info->domain = domain;
1859 /* This domain is shared by devices under p2p bridge */
1860 domain->flags |= DOMAIN_FLAG_P2P_MULTIPLE_DEVICES;
1861
1862 /* pcie-to-pci bridge already has a domain, uses it */
1863 found = NULL;
1864 spin_lock_irqsave(&device_domain_lock, flags);
1865 list_for_each_entry(tmp, &device_domain_list, global) {
1866 if (tmp->segment == segment &&
1867 tmp->bus == bus && tmp->devfn == devfn) {
1868 found = tmp->domain;
1869 break;
1870 }
1871 }
1872 if (found) {
1873 spin_unlock_irqrestore(&device_domain_lock, flags);
1874 free_devinfo_mem(info);
1875 domain_exit(domain);
1876 domain = found;
1877 } else {
1878 list_add(&info->link, &domain->devices);
1879 list_add(&info->global, &device_domain_list);
1880 spin_unlock_irqrestore(&device_domain_lock, flags);
1881 }
1882 }
1883
1884 found_domain:
1885 info = alloc_devinfo_mem();
1886 if (!info)
1887 goto error;
1888 info->segment = segment;
1889 info->bus = pdev->bus->number;
1890 info->devfn = pdev->devfn;
1891 info->dev = pdev;
1892 info->domain = domain;
1893 spin_lock_irqsave(&device_domain_lock, flags);
1894 /* somebody is fast */
1895 found = find_domain(pdev);
1896 if (found != NULL) {
1897 spin_unlock_irqrestore(&device_domain_lock, flags);
1898 if (found != domain) {
1899 domain_exit(domain);
1900 domain = found;
1901 }
1902 free_devinfo_mem(info);
1903 return domain;
1904 }
1905 list_add(&info->link, &domain->devices);
1906 list_add(&info->global, &device_domain_list);
1907 pdev->dev.archdata.iommu = info;
1908 spin_unlock_irqrestore(&device_domain_lock, flags);
1909 return domain;
1910 error:
1911 /* recheck it here, maybe others set it */
1912 return find_domain(pdev);
1913 }
1914
1915 static int iommu_identity_mapping;
1916 #define IDENTMAP_ALL 1
1917 #define IDENTMAP_GFX 2
1918 #define IDENTMAP_AZALIA 4
1919
1920 static int iommu_domain_identity_map(struct dmar_domain *domain,
1921 unsigned long long start,
1922 unsigned long long end)
1923 {
1924 unsigned long first_vpfn = start >> VTD_PAGE_SHIFT;
1925 unsigned long last_vpfn = end >> VTD_PAGE_SHIFT;
1926
1927 if (!reserve_iova(&domain->iovad, dma_to_mm_pfn(first_vpfn),
1928 dma_to_mm_pfn(last_vpfn))) {
1929 printk(KERN_ERR "IOMMU: reserve iova failed\n");
1930 return -ENOMEM;
1931 }
1932
1933 pr_debug("Mapping reserved region %llx-%llx for domain %d\n",
1934 start, end, domain->id);
1935 /*
1936 * RMRR range might have overlap with physical memory range,
1937 * clear it first
1938 */
1939 dma_pte_clear_range(domain, first_vpfn, last_vpfn);
1940
1941 return domain_pfn_mapping(domain, first_vpfn, first_vpfn,
1942 last_vpfn - first_vpfn + 1,
1943 DMA_PTE_READ|DMA_PTE_WRITE);
1944 }
1945
1946 static int iommu_prepare_identity_map(struct pci_dev *pdev,
1947 unsigned long long start,
1948 unsigned long long end)
1949 {
1950 struct dmar_domain *domain;
1951 int ret;
1952
1953 domain = get_domain_for_dev(pdev, DEFAULT_DOMAIN_ADDRESS_WIDTH);
1954 if (!domain)
1955 return -ENOMEM;
1956
1957 /* For _hardware_ passthrough, don't bother. But for software
1958 passthrough, we do it anyway -- it may indicate a memory
1959 range which is reserved in E820, so which didn't get set
1960 up to start with in si_domain */
1961 if (domain == si_domain && hw_pass_through) {
1962 printk("Ignoring identity map for HW passthrough device %s [0x%Lx - 0x%Lx]\n",
1963 pci_name(pdev), start, end);
1964 return 0;
1965 }
1966
1967 printk(KERN_INFO
1968 "IOMMU: Setting identity map for device %s [0x%Lx - 0x%Lx]\n",
1969 pci_name(pdev), start, end);
1970
1971 if (end < start) {
1972 WARN(1, "Your BIOS is broken; RMRR ends before it starts!\n"
1973 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
1974 dmi_get_system_info(DMI_BIOS_VENDOR),
1975 dmi_get_system_info(DMI_BIOS_VERSION),
1976 dmi_get_system_info(DMI_PRODUCT_VERSION));
1977 ret = -EIO;
1978 goto error;
1979 }
1980
1981 if (end >> agaw_to_width(domain->agaw)) {
1982 WARN(1, "Your BIOS is broken; RMRR exceeds permitted address width (%d bits)\n"
1983 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
1984 agaw_to_width(domain->agaw),
1985 dmi_get_system_info(DMI_BIOS_VENDOR),
1986 dmi_get_system_info(DMI_BIOS_VERSION),
1987 dmi_get_system_info(DMI_PRODUCT_VERSION));
1988 ret = -EIO;
1989 goto error;
1990 }
1991
1992 ret = iommu_domain_identity_map(domain, start, end);
1993 if (ret)
1994 goto error;
1995
1996 /* context entry init */
1997 ret = domain_context_mapping(domain, pdev, CONTEXT_TT_MULTI_LEVEL);
1998 if (ret)
1999 goto error;
2000
2001 return 0;
2002
2003 error:
2004 domain_exit(domain);
2005 return ret;
2006 }
2007
2008 static inline int iommu_prepare_rmrr_dev(struct dmar_rmrr_unit *rmrr,
2009 struct pci_dev *pdev)
2010 {
2011 if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO)
2012 return 0;
2013 return iommu_prepare_identity_map(pdev, rmrr->base_address,
2014 rmrr->end_address + 1);
2015 }
2016
2017 #ifdef CONFIG_DMAR_FLOPPY_WA
2018 static inline void iommu_prepare_isa(void)
2019 {
2020 struct pci_dev *pdev;
2021 int ret;
2022
2023 pdev = pci_get_class(PCI_CLASS_BRIDGE_ISA << 8, NULL);
2024 if (!pdev)
2025 return;
2026
2027 printk(KERN_INFO "IOMMU: Prepare 0-16MiB unity mapping for LPC\n");
2028 ret = iommu_prepare_identity_map(pdev, 0, 16*1024*1024);
2029
2030 if (ret)
2031 printk(KERN_ERR "IOMMU: Failed to create 0-16MiB identity map; "
2032 "floppy might not work\n");
2033
2034 }
2035 #else
2036 static inline void iommu_prepare_isa(void)
2037 {
2038 return;
2039 }
2040 #endif /* !CONFIG_DMAR_FLPY_WA */
2041
2042 static int md_domain_init(struct dmar_domain *domain, int guest_width);
2043
2044 static int __init si_domain_work_fn(unsigned long start_pfn,
2045 unsigned long end_pfn, void *datax)
2046 {
2047 int *ret = datax;
2048
2049 *ret = iommu_domain_identity_map(si_domain,
2050 (uint64_t)start_pfn << PAGE_SHIFT,
2051 (uint64_t)end_pfn << PAGE_SHIFT);
2052 return *ret;
2053
2054 }
2055
2056 static int __init si_domain_init(int hw)
2057 {
2058 struct dmar_drhd_unit *drhd;
2059 struct intel_iommu *iommu;
2060 int nid, ret = 0;
2061
2062 si_domain = alloc_domain();
2063 if (!si_domain)
2064 return -EFAULT;
2065
2066 pr_debug("Identity mapping domain is domain %d\n", si_domain->id);
2067
2068 for_each_active_iommu(iommu, drhd) {
2069 ret = iommu_attach_domain(si_domain, iommu);
2070 if (ret) {
2071 domain_exit(si_domain);
2072 return -EFAULT;
2073 }
2074 }
2075
2076 if (md_domain_init(si_domain, DEFAULT_DOMAIN_ADDRESS_WIDTH)) {
2077 domain_exit(si_domain);
2078 return -EFAULT;
2079 }
2080
2081 si_domain->flags = DOMAIN_FLAG_STATIC_IDENTITY;
2082
2083 if (hw)
2084 return 0;
2085
2086 for_each_online_node(nid) {
2087 work_with_active_regions(nid, si_domain_work_fn, &ret);
2088 if (ret)
2089 return ret;
2090 }
2091
2092 return 0;
2093 }
2094
2095 static void domain_remove_one_dev_info(struct dmar_domain *domain,
2096 struct pci_dev *pdev);
2097 static int identity_mapping(struct pci_dev *pdev)
2098 {
2099 struct device_domain_info *info;
2100
2101 if (likely(!iommu_identity_mapping))
2102 return 0;
2103
2104
2105 list_for_each_entry(info, &si_domain->devices, link)
2106 if (info->dev == pdev)
2107 return 1;
2108 return 0;
2109 }
2110
2111 static int domain_add_dev_info(struct dmar_domain *domain,
2112 struct pci_dev *pdev,
2113 int translation)
2114 {
2115 struct device_domain_info *info;
2116 unsigned long flags;
2117 int ret;
2118
2119 info = alloc_devinfo_mem();
2120 if (!info)
2121 return -ENOMEM;
2122
2123 ret = domain_context_mapping(domain, pdev, translation);
2124 if (ret) {
2125 free_devinfo_mem(info);
2126 return ret;
2127 }
2128
2129 info->segment = pci_domain_nr(pdev->bus);
2130 info->bus = pdev->bus->number;
2131 info->devfn = pdev->devfn;
2132 info->dev = pdev;
2133 info->domain = domain;
2134
2135 spin_lock_irqsave(&device_domain_lock, flags);
2136 list_add(&info->link, &domain->devices);
2137 list_add(&info->global, &device_domain_list);
2138 pdev->dev.archdata.iommu = info;
2139 spin_unlock_irqrestore(&device_domain_lock, flags);
2140
2141 return 0;
2142 }
2143
2144 static int iommu_should_identity_map(struct pci_dev *pdev, int startup)
2145 {
2146 if ((iommu_identity_mapping & IDENTMAP_AZALIA) && IS_AZALIA(pdev))
2147 return 1;
2148
2149 if ((iommu_identity_mapping & IDENTMAP_GFX) && IS_GFX_DEVICE(pdev))
2150 return 1;
2151
2152 if (!(iommu_identity_mapping & IDENTMAP_ALL))
2153 return 0;
2154
2155 /*
2156 * We want to start off with all devices in the 1:1 domain, and
2157 * take them out later if we find they can't access all of memory.
2158 *
2159 * However, we can't do this for PCI devices behind bridges,
2160 * because all PCI devices behind the same bridge will end up
2161 * with the same source-id on their transactions.
2162 *
2163 * Practically speaking, we can't change things around for these
2164 * devices at run-time, because we can't be sure there'll be no
2165 * DMA transactions in flight for any of their siblings.
2166 *
2167 * So PCI devices (unless they're on the root bus) as well as
2168 * their parent PCI-PCI or PCIe-PCI bridges must be left _out_ of
2169 * the 1:1 domain, just in _case_ one of their siblings turns out
2170 * not to be able to map all of memory.
2171 */
2172 if (!pci_is_pcie(pdev)) {
2173 if (!pci_is_root_bus(pdev->bus))
2174 return 0;
2175 if (pdev->class >> 8 == PCI_CLASS_BRIDGE_PCI)
2176 return 0;
2177 } else if (pdev->pcie_type == PCI_EXP_TYPE_PCI_BRIDGE)
2178 return 0;
2179
2180 /*
2181 * At boot time, we don't yet know if devices will be 64-bit capable.
2182 * Assume that they will -- if they turn out not to be, then we can
2183 * take them out of the 1:1 domain later.
2184 */
2185 if (!startup)
2186 return pdev->dma_mask > DMA_BIT_MASK(32);
2187
2188 return 1;
2189 }
2190
2191 static int __init iommu_prepare_static_identity_mapping(int hw)
2192 {
2193 struct pci_dev *pdev = NULL;
2194 int ret;
2195
2196 ret = si_domain_init(hw);
2197 if (ret)
2198 return -EFAULT;
2199
2200 for_each_pci_dev(pdev) {
2201 if (iommu_should_identity_map(pdev, 1)) {
2202 printk(KERN_INFO "IOMMU: %s identity mapping for device %s\n",
2203 hw ? "hardware" : "software", pci_name(pdev));
2204
2205 ret = domain_add_dev_info(si_domain, pdev,
2206 hw ? CONTEXT_TT_PASS_THROUGH :
2207 CONTEXT_TT_MULTI_LEVEL);
2208 if (ret)
2209 return ret;
2210 }
2211 }
2212
2213 return 0;
2214 }
2215
2216 int __init init_dmars(void)
2217 {
2218 struct dmar_drhd_unit *drhd;
2219 struct dmar_rmrr_unit *rmrr;
2220 struct pci_dev *pdev;
2221 struct intel_iommu *iommu;
2222 int i, ret;
2223
2224 /*
2225 * for each drhd
2226 * allocate root
2227 * initialize and program root entry to not present
2228 * endfor
2229 */
2230 for_each_drhd_unit(drhd) {
2231 g_num_of_iommus++;
2232 /*
2233 * lock not needed as this is only incremented in the single
2234 * threaded kernel __init code path all other access are read
2235 * only
2236 */
2237 }
2238
2239 g_iommus = kcalloc(g_num_of_iommus, sizeof(struct intel_iommu *),
2240 GFP_KERNEL);
2241 if (!g_iommus) {
2242 printk(KERN_ERR "Allocating global iommu array failed\n");
2243 ret = -ENOMEM;
2244 goto error;
2245 }
2246
2247 deferred_flush = kzalloc(g_num_of_iommus *
2248 sizeof(struct deferred_flush_tables), GFP_KERNEL);
2249 if (!deferred_flush) {
2250 ret = -ENOMEM;
2251 goto error;
2252 }
2253
2254 for_each_drhd_unit(drhd) {
2255 if (drhd->ignored)
2256 continue;
2257
2258 iommu = drhd->iommu;
2259 g_iommus[iommu->seq_id] = iommu;
2260
2261 ret = iommu_init_domains(iommu);
2262 if (ret)
2263 goto error;
2264
2265 /*
2266 * TBD:
2267 * we could share the same root & context tables
2268 * amoung all IOMMU's. Need to Split it later.
2269 */
2270 ret = iommu_alloc_root_entry(iommu);
2271 if (ret) {
2272 printk(KERN_ERR "IOMMU: allocate root entry failed\n");
2273 goto error;
2274 }
2275 if (!ecap_pass_through(iommu->ecap))
2276 hw_pass_through = 0;
2277 }
2278
2279 /*
2280 * Start from the sane iommu hardware state.
2281 */
2282 for_each_drhd_unit(drhd) {
2283 if (drhd->ignored)
2284 continue;
2285
2286 iommu = drhd->iommu;
2287
2288 /*
2289 * If the queued invalidation is already initialized by us
2290 * (for example, while enabling interrupt-remapping) then
2291 * we got the things already rolling from a sane state.
2292 */
2293 if (iommu->qi)
2294 continue;
2295
2296 /*
2297 * Clear any previous faults.
2298 */
2299 dmar_fault(-1, iommu);
2300 /*
2301 * Disable queued invalidation if supported and already enabled
2302 * before OS handover.
2303 */
2304 dmar_disable_qi(iommu);
2305 }
2306
2307 for_each_drhd_unit(drhd) {
2308 if (drhd->ignored)
2309 continue;
2310
2311 iommu = drhd->iommu;
2312
2313 if (dmar_enable_qi(iommu)) {
2314 /*
2315 * Queued Invalidate not enabled, use Register Based
2316 * Invalidate
2317 */
2318 iommu->flush.flush_context = __iommu_flush_context;
2319 iommu->flush.flush_iotlb = __iommu_flush_iotlb;
2320 printk(KERN_INFO "IOMMU %d 0x%Lx: using Register based "
2321 "invalidation\n",
2322 iommu->seq_id,
2323 (unsigned long long)drhd->reg_base_addr);
2324 } else {
2325 iommu->flush.flush_context = qi_flush_context;
2326 iommu->flush.flush_iotlb = qi_flush_iotlb;
2327 printk(KERN_INFO "IOMMU %d 0x%Lx: using Queued "
2328 "invalidation\n",
2329 iommu->seq_id,
2330 (unsigned long long)drhd->reg_base_addr);
2331 }
2332 }
2333
2334 if (iommu_pass_through)
2335 iommu_identity_mapping |= IDENTMAP_ALL;
2336
2337 #ifdef CONFIG_DMAR_BROKEN_GFX_WA
2338 iommu_identity_mapping |= IDENTMAP_GFX;
2339 #endif
2340
2341 check_tylersburg_isoch();
2342
2343 /*
2344 * If pass through is not set or not enabled, setup context entries for
2345 * identity mappings for rmrr, gfx, and isa and may fall back to static
2346 * identity mapping if iommu_identity_mapping is set.
2347 */
2348 if (iommu_identity_mapping) {
2349 ret = iommu_prepare_static_identity_mapping(hw_pass_through);
2350 if (ret) {
2351 printk(KERN_CRIT "Failed to setup IOMMU pass-through\n");
2352 goto error;
2353 }
2354 }
2355 /*
2356 * For each rmrr
2357 * for each dev attached to rmrr
2358 * do
2359 * locate drhd for dev, alloc domain for dev
2360 * allocate free domain
2361 * allocate page table entries for rmrr
2362 * if context not allocated for bus
2363 * allocate and init context
2364 * set present in root table for this bus
2365 * init context with domain, translation etc
2366 * endfor
2367 * endfor
2368 */
2369 printk(KERN_INFO "IOMMU: Setting RMRR:\n");
2370 for_each_rmrr_units(rmrr) {
2371 for (i = 0; i < rmrr->devices_cnt; i++) {
2372 pdev = rmrr->devices[i];
2373 /*
2374 * some BIOS lists non-exist devices in DMAR
2375 * table.
2376 */
2377 if (!pdev)
2378 continue;
2379 ret = iommu_prepare_rmrr_dev(rmrr, pdev);
2380 if (ret)
2381 printk(KERN_ERR
2382 "IOMMU: mapping reserved region failed\n");
2383 }
2384 }
2385
2386 iommu_prepare_isa();
2387
2388 /*
2389 * for each drhd
2390 * enable fault log
2391 * global invalidate context cache
2392 * global invalidate iotlb
2393 * enable translation
2394 */
2395 for_each_drhd_unit(drhd) {
2396 if (drhd->ignored)
2397 continue;
2398 iommu = drhd->iommu;
2399
2400 iommu_flush_write_buffer(iommu);
2401
2402 ret = dmar_set_interrupt(iommu);
2403 if (ret)
2404 goto error;
2405
2406 iommu_set_root_entry(iommu);
2407
2408 iommu->flush.flush_context(iommu, 0, 0, 0, DMA_CCMD_GLOBAL_INVL);
2409 iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH);
2410
2411 ret = iommu_enable_translation(iommu);
2412 if (ret)
2413 goto error;
2414
2415 iommu_disable_protect_mem_regions(iommu);
2416 }
2417
2418 return 0;
2419 error:
2420 for_each_drhd_unit(drhd) {
2421 if (drhd->ignored)
2422 continue;
2423 iommu = drhd->iommu;
2424 free_iommu(iommu);
2425 }
2426 kfree(g_iommus);
2427 return ret;
2428 }
2429
2430 /* This takes a number of _MM_ pages, not VTD pages */
2431 static struct iova *intel_alloc_iova(struct device *dev,
2432 struct dmar_domain *domain,
2433 unsigned long nrpages, uint64_t dma_mask)
2434 {
2435 struct pci_dev *pdev = to_pci_dev(dev);
2436 struct iova *iova = NULL;
2437
2438 /* Restrict dma_mask to the width that the iommu can handle */
2439 dma_mask = min_t(uint64_t, DOMAIN_MAX_ADDR(domain->gaw), dma_mask);
2440
2441 if (!dmar_forcedac && dma_mask > DMA_BIT_MASK(32)) {
2442 /*
2443 * First try to allocate an io virtual address in
2444 * DMA_BIT_MASK(32) and if that fails then try allocating
2445 * from higher range
2446 */
2447 iova = alloc_iova(&domain->iovad, nrpages,
2448 IOVA_PFN(DMA_BIT_MASK(32)), 1);
2449 if (iova)
2450 return iova;
2451 }
2452 iova = alloc_iova(&domain->iovad, nrpages, IOVA_PFN(dma_mask), 1);
2453 if (unlikely(!iova)) {
2454 printk(KERN_ERR "Allocating %ld-page iova for %s failed",
2455 nrpages, pci_name(pdev));
2456 return NULL;
2457 }
2458
2459 return iova;
2460 }
2461
2462 static struct dmar_domain *__get_valid_domain_for_dev(struct pci_dev *pdev)
2463 {
2464 struct dmar_domain *domain;
2465 int ret;
2466
2467 domain = get_domain_for_dev(pdev,
2468 DEFAULT_DOMAIN_ADDRESS_WIDTH);
2469 if (!domain) {
2470 printk(KERN_ERR
2471 "Allocating domain for %s failed", pci_name(pdev));
2472 return NULL;
2473 }
2474
2475 /* make sure context mapping is ok */
2476 if (unlikely(!domain_context_mapped(pdev))) {
2477 ret = domain_context_mapping(domain, pdev,
2478 CONTEXT_TT_MULTI_LEVEL);
2479 if (ret) {
2480 printk(KERN_ERR
2481 "Domain context map for %s failed",
2482 pci_name(pdev));
2483 return NULL;
2484 }
2485 }
2486
2487 return domain;
2488 }
2489
2490 static inline struct dmar_domain *get_valid_domain_for_dev(struct pci_dev *dev)
2491 {
2492 struct device_domain_info *info;
2493
2494 /* No lock here, assumes no domain exit in normal case */
2495 info = dev->dev.archdata.iommu;
2496 if (likely(info))
2497 return info->domain;
2498
2499 return __get_valid_domain_for_dev(dev);
2500 }
2501
2502 static int iommu_dummy(struct pci_dev *pdev)
2503 {
2504 return pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO;
2505 }
2506
2507 /* Check if the pdev needs to go through non-identity map and unmap process.*/
2508 static int iommu_no_mapping(struct device *dev)
2509 {
2510 struct pci_dev *pdev;
2511 int found;
2512
2513 if (unlikely(dev->bus != &pci_bus_type))
2514 return 1;
2515
2516 pdev = to_pci_dev(dev);
2517 if (iommu_dummy(pdev))
2518 return 1;
2519
2520 if (!iommu_identity_mapping)
2521 return 0;
2522
2523 found = identity_mapping(pdev);
2524 if (found) {
2525 if (iommu_should_identity_map(pdev, 0))
2526 return 1;
2527 else {
2528 /*
2529 * 32 bit DMA is removed from si_domain and fall back
2530 * to non-identity mapping.
2531 */
2532 domain_remove_one_dev_info(si_domain, pdev);
2533 printk(KERN_INFO "32bit %s uses non-identity mapping\n",
2534 pci_name(pdev));
2535 return 0;
2536 }
2537 } else {
2538 /*
2539 * In case of a detached 64 bit DMA device from vm, the device
2540 * is put into si_domain for identity mapping.
2541 */
2542 if (iommu_should_identity_map(pdev, 0)) {
2543 int ret;
2544 ret = domain_add_dev_info(si_domain, pdev,
2545 hw_pass_through ?
2546 CONTEXT_TT_PASS_THROUGH :
2547 CONTEXT_TT_MULTI_LEVEL);
2548 if (!ret) {
2549 printk(KERN_INFO "64bit %s uses identity mapping\n",
2550 pci_name(pdev));
2551 return 1;
2552 }
2553 }
2554 }
2555
2556 return 0;
2557 }
2558
2559 static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
2560 size_t size, int dir, u64 dma_mask)
2561 {
2562 struct pci_dev *pdev = to_pci_dev(hwdev);
2563 struct dmar_domain *domain;
2564 phys_addr_t start_paddr;
2565 struct iova *iova;
2566 int prot = 0;
2567 int ret;
2568 struct intel_iommu *iommu;
2569 unsigned long paddr_pfn = paddr >> PAGE_SHIFT;
2570
2571 BUG_ON(dir == DMA_NONE);
2572
2573 if (iommu_no_mapping(hwdev))
2574 return paddr;
2575
2576 domain = get_valid_domain_for_dev(pdev);
2577 if (!domain)
2578 return 0;
2579
2580 iommu = domain_get_iommu(domain);
2581 size = aligned_nrpages(paddr, size);
2582
2583 iova = intel_alloc_iova(hwdev, domain, dma_to_mm_pfn(size),
2584 pdev->dma_mask);
2585 if (!iova)
2586 goto error;
2587
2588 /*
2589 * Check if DMAR supports zero-length reads on write only
2590 * mappings..
2591 */
2592 if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL || \
2593 !cap_zlr(iommu->cap))
2594 prot |= DMA_PTE_READ;
2595 if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
2596 prot |= DMA_PTE_WRITE;
2597 /*
2598 * paddr - (paddr + size) might be partial page, we should map the whole
2599 * page. Note: if two part of one page are separately mapped, we
2600 * might have two guest_addr mapping to the same host paddr, but this
2601 * is not a big problem
2602 */
2603 ret = domain_pfn_mapping(domain, mm_to_dma_pfn(iova->pfn_lo),
2604 mm_to_dma_pfn(paddr_pfn), size, prot);
2605 if (ret)
2606 goto error;
2607
2608 /* it's a non-present to present mapping. Only flush if caching mode */
2609 if (cap_caching_mode(iommu->cap))
2610 iommu_flush_iotlb_psi(iommu, domain->id, mm_to_dma_pfn(iova->pfn_lo), size, 1);
2611 else
2612 iommu_flush_write_buffer(iommu);
2613
2614 start_paddr = (phys_addr_t)iova->pfn_lo << PAGE_SHIFT;
2615 start_paddr += paddr & ~PAGE_MASK;
2616 return start_paddr;
2617
2618 error:
2619 if (iova)
2620 __free_iova(&domain->iovad, iova);
2621 printk(KERN_ERR"Device %s request: %zx@%llx dir %d --- failed\n",
2622 pci_name(pdev), size, (unsigned long long)paddr, dir);
2623 return 0;
2624 }
2625
2626 static dma_addr_t intel_map_page(struct device *dev, struct page *page,
2627 unsigned long offset, size_t size,
2628 enum dma_data_direction dir,
2629 struct dma_attrs *attrs)
2630 {
2631 return __intel_map_single(dev, page_to_phys(page) + offset, size,
2632 dir, to_pci_dev(dev)->dma_mask);
2633 }
2634
2635 static void flush_unmaps(void)
2636 {
2637 int i, j;
2638
2639 timer_on = 0;
2640
2641 /* just flush them all */
2642 for (i = 0; i < g_num_of_iommus; i++) {
2643 struct intel_iommu *iommu = g_iommus[i];
2644 if (!iommu)
2645 continue;
2646
2647 if (!deferred_flush[i].next)
2648 continue;
2649
2650 /* In caching mode, global flushes turn emulation expensive */
2651 if (!cap_caching_mode(iommu->cap))
2652 iommu->flush.flush_iotlb(iommu, 0, 0, 0,
2653 DMA_TLB_GLOBAL_FLUSH);
2654 for (j = 0; j < deferred_flush[i].next; j++) {
2655 unsigned long mask;
2656 struct iova *iova = deferred_flush[i].iova[j];
2657 struct dmar_domain *domain = deferred_flush[i].domain[j];
2658
2659 /* On real hardware multiple invalidations are expensive */
2660 if (cap_caching_mode(iommu->cap))
2661 iommu_flush_iotlb_psi(iommu, domain->id,
2662 iova->pfn_lo, iova->pfn_hi - iova->pfn_lo + 1, 0);
2663 else {
2664 mask = ilog2(mm_to_dma_pfn(iova->pfn_hi - iova->pfn_lo + 1));
2665 iommu_flush_dev_iotlb(deferred_flush[i].domain[j],
2666 (uint64_t)iova->pfn_lo << PAGE_SHIFT, mask);
2667 }
2668 __free_iova(&deferred_flush[i].domain[j]->iovad, iova);
2669 }
2670 deferred_flush[i].next = 0;
2671 }
2672
2673 list_size = 0;
2674 }
2675
2676 static void flush_unmaps_timeout(unsigned long data)
2677 {
2678 unsigned long flags;
2679
2680 spin_lock_irqsave(&async_umap_flush_lock, flags);
2681 flush_unmaps();
2682 spin_unlock_irqrestore(&async_umap_flush_lock, flags);
2683 }
2684
2685 static void add_unmap(struct dmar_domain *dom, struct iova *iova)
2686 {
2687 unsigned long flags;
2688 int next, iommu_id;
2689 struct intel_iommu *iommu;
2690
2691 spin_lock_irqsave(&async_umap_flush_lock, flags);
2692 if (list_size == HIGH_WATER_MARK)
2693 flush_unmaps();
2694
2695 iommu = domain_get_iommu(dom);
2696 iommu_id = iommu->seq_id;
2697
2698 next = deferred_flush[iommu_id].next;
2699 deferred_flush[iommu_id].domain[next] = dom;
2700 deferred_flush[iommu_id].iova[next] = iova;
2701 deferred_flush[iommu_id].next++;
2702
2703 if (!timer_on) {
2704 mod_timer(&unmap_timer, jiffies + msecs_to_jiffies(10));
2705 timer_on = 1;
2706 }
2707 list_size++;
2708 spin_unlock_irqrestore(&async_umap_flush_lock, flags);
2709 }
2710
2711 static void intel_unmap_page(struct device *dev, dma_addr_t dev_addr,
2712 size_t size, enum dma_data_direction dir,
2713 struct dma_attrs *attrs)
2714 {
2715 struct pci_dev *pdev = to_pci_dev(dev);
2716 struct dmar_domain *domain;
2717 unsigned long start_pfn, last_pfn;
2718 struct iova *iova;
2719 struct intel_iommu *iommu;
2720
2721 if (iommu_no_mapping(dev))
2722 return;
2723
2724 domain = find_domain(pdev);
2725 BUG_ON(!domain);
2726
2727 iommu = domain_get_iommu(domain);
2728
2729 iova = find_iova(&domain->iovad, IOVA_PFN(dev_addr));
2730 if (WARN_ONCE(!iova, "Driver unmaps unmatched page at PFN %llx\n",
2731 (unsigned long long)dev_addr))
2732 return;
2733
2734 start_pfn = mm_to_dma_pfn(iova->pfn_lo);
2735 last_pfn = mm_to_dma_pfn(iova->pfn_hi + 1) - 1;
2736
2737 pr_debug("Device %s unmapping: pfn %lx-%lx\n",
2738 pci_name(pdev), start_pfn, last_pfn);
2739
2740 /* clear the whole page */
2741 dma_pte_clear_range(domain, start_pfn, last_pfn);
2742
2743 /* free page tables */
2744 dma_pte_free_pagetable(domain, start_pfn, last_pfn);
2745
2746 if (intel_iommu_strict) {
2747 iommu_flush_iotlb_psi(iommu, domain->id, start_pfn,
2748 last_pfn - start_pfn + 1, 0);
2749 /* free iova */
2750 __free_iova(&domain->iovad, iova);
2751 } else {
2752 add_unmap(domain, iova);
2753 /*
2754 * queue up the release of the unmap to save the 1/6th of the
2755 * cpu used up by the iotlb flush operation...
2756 */
2757 }
2758 }
2759
2760 static void *intel_alloc_coherent(struct device *hwdev, size_t size,
2761 dma_addr_t *dma_handle, gfp_t flags)
2762 {
2763 void *vaddr;
2764 int order;
2765
2766 size = PAGE_ALIGN(size);
2767 order = get_order(size);
2768
2769 if (!iommu_no_mapping(hwdev))
2770 flags &= ~(GFP_DMA | GFP_DMA32);
2771 else if (hwdev->coherent_dma_mask < dma_get_required_mask(hwdev)) {
2772 if (hwdev->coherent_dma_mask < DMA_BIT_MASK(32))
2773 flags |= GFP_DMA;
2774 else
2775 flags |= GFP_DMA32;
2776 }
2777
2778 vaddr = (void *)__get_free_pages(flags, order);
2779 if (!vaddr)
2780 return NULL;
2781 memset(vaddr, 0, size);
2782
2783 *dma_handle = __intel_map_single(hwdev, virt_to_bus(vaddr), size,
2784 DMA_BIDIRECTIONAL,
2785 hwdev->coherent_dma_mask);
2786 if (*dma_handle)
2787 return vaddr;
2788 free_pages((unsigned long)vaddr, order);
2789 return NULL;
2790 }
2791
2792 static void intel_free_coherent(struct device *hwdev, size_t size, void *vaddr,
2793 dma_addr_t dma_handle)
2794 {
2795 int order;
2796
2797 size = PAGE_ALIGN(size);
2798 order = get_order(size);
2799
2800 intel_unmap_page(hwdev, dma_handle, size, DMA_BIDIRECTIONAL, NULL);
2801 free_pages((unsigned long)vaddr, order);
2802 }
2803
2804 static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
2805 int nelems, enum dma_data_direction dir,
2806 struct dma_attrs *attrs)
2807 {
2808 struct pci_dev *pdev = to_pci_dev(hwdev);
2809 struct dmar_domain *domain;
2810 unsigned long start_pfn, last_pfn;
2811 struct iova *iova;
2812 struct intel_iommu *iommu;
2813
2814 if (iommu_no_mapping(hwdev))
2815 return;
2816
2817 domain = find_domain(pdev);
2818 BUG_ON(!domain);
2819
2820 iommu = domain_get_iommu(domain);
2821
2822 iova = find_iova(&domain->iovad, IOVA_PFN(sglist[0].dma_address));
2823 if (WARN_ONCE(!iova, "Driver unmaps unmatched sglist at PFN %llx\n",
2824 (unsigned long long)sglist[0].dma_address))
2825 return;
2826
2827 start_pfn = mm_to_dma_pfn(iova->pfn_lo);
2828 last_pfn = mm_to_dma_pfn(iova->pfn_hi + 1) - 1;
2829
2830 /* clear the whole page */
2831 dma_pte_clear_range(domain, start_pfn, last_pfn);
2832
2833 /* free page tables */
2834 dma_pte_free_pagetable(domain, start_pfn, last_pfn);
2835
2836 if (intel_iommu_strict) {
2837 iommu_flush_iotlb_psi(iommu, domain->id, start_pfn,
2838 last_pfn - start_pfn + 1, 0);
2839 /* free iova */
2840 __free_iova(&domain->iovad, iova);
2841 } else {
2842 add_unmap(domain, iova);
2843 /*
2844 * queue up the release of the unmap to save the 1/6th of the
2845 * cpu used up by the iotlb flush operation...
2846 */
2847 }
2848 }
2849
2850 static int intel_nontranslate_map_sg(struct device *hddev,
2851 struct scatterlist *sglist, int nelems, int dir)
2852 {
2853 int i;
2854 struct scatterlist *sg;
2855
2856 for_each_sg(sglist, sg, nelems, i) {
2857 BUG_ON(!sg_page(sg));
2858 sg->dma_address = page_to_phys(sg_page(sg)) + sg->offset;
2859 sg->dma_length = sg->length;
2860 }
2861 return nelems;
2862 }
2863
2864 static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
2865 enum dma_data_direction dir, struct dma_attrs *attrs)
2866 {
2867 int i;
2868 struct pci_dev *pdev = to_pci_dev(hwdev);
2869 struct dmar_domain *domain;
2870 size_t size = 0;
2871 int prot = 0;
2872 struct iova *iova = NULL;
2873 int ret;
2874 struct scatterlist *sg;
2875 unsigned long start_vpfn;
2876 struct intel_iommu *iommu;
2877
2878 BUG_ON(dir == DMA_NONE);
2879 if (iommu_no_mapping(hwdev))
2880 return intel_nontranslate_map_sg(hwdev, sglist, nelems, dir);
2881
2882 domain = get_valid_domain_for_dev(pdev);
2883 if (!domain)
2884 return 0;
2885
2886 iommu = domain_get_iommu(domain);
2887
2888 for_each_sg(sglist, sg, nelems, i)
2889 size += aligned_nrpages(sg->offset, sg->length);
2890
2891 iova = intel_alloc_iova(hwdev, domain, dma_to_mm_pfn(size),
2892 pdev->dma_mask);
2893 if (!iova) {
2894 sglist->dma_length = 0;
2895 return 0;
2896 }
2897
2898 /*
2899 * Check if DMAR supports zero-length reads on write only
2900 * mappings..
2901 */
2902 if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL || \
2903 !cap_zlr(iommu->cap))
2904 prot |= DMA_PTE_READ;
2905 if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
2906 prot |= DMA_PTE_WRITE;
2907
2908 start_vpfn = mm_to_dma_pfn(iova->pfn_lo);
2909
2910 ret = domain_sg_mapping(domain, start_vpfn, sglist, size, prot);
2911 if (unlikely(ret)) {
2912 /* clear the page */
2913 dma_pte_clear_range(domain, start_vpfn,
2914 start_vpfn + size - 1);
2915 /* free page tables */
2916 dma_pte_free_pagetable(domain, start_vpfn,
2917 start_vpfn + size - 1);
2918 /* free iova */
2919 __free_iova(&domain->iovad, iova);
2920 return 0;
2921 }
2922
2923 /* it's a non-present to present mapping. Only flush if caching mode */
2924 if (cap_caching_mode(iommu->cap))
2925 iommu_flush_iotlb_psi(iommu, domain->id, start_vpfn, size, 1);
2926 else
2927 iommu_flush_write_buffer(iommu);
2928
2929 return nelems;
2930 }
2931
2932 static int intel_mapping_error(struct device *dev, dma_addr_t dma_addr)
2933 {
2934 return !dma_addr;
2935 }
2936
2937 struct dma_map_ops intel_dma_ops = {
2938 .alloc_coherent = intel_alloc_coherent,
2939 .free_coherent = intel_free_coherent,
2940 .map_sg = intel_map_sg,
2941 .unmap_sg = intel_unmap_sg,
2942 .map_page = intel_map_page,
2943 .unmap_page = intel_unmap_page,
2944 .mapping_error = intel_mapping_error,
2945 };
2946
2947 static inline int iommu_domain_cache_init(void)
2948 {
2949 int ret = 0;
2950
2951 iommu_domain_cache = kmem_cache_create("iommu_domain",
2952 sizeof(struct dmar_domain),
2953 0,
2954 SLAB_HWCACHE_ALIGN,
2955
2956 NULL);
2957 if (!iommu_domain_cache) {
2958 printk(KERN_ERR "Couldn't create iommu_domain cache\n");
2959 ret = -ENOMEM;
2960 }
2961
2962 return ret;
2963 }
2964
2965 static inline int iommu_devinfo_cache_init(void)
2966 {
2967 int ret = 0;
2968
2969 iommu_devinfo_cache = kmem_cache_create("iommu_devinfo",
2970 sizeof(struct device_domain_info),
2971 0,
2972 SLAB_HWCACHE_ALIGN,
2973 NULL);
2974 if (!iommu_devinfo_cache) {
2975 printk(KERN_ERR "Couldn't create devinfo cache\n");
2976 ret = -ENOMEM;
2977 }
2978
2979 return ret;
2980 }
2981
2982 static inline int iommu_iova_cache_init(void)
2983 {
2984 int ret = 0;
2985
2986 iommu_iova_cache = kmem_cache_create("iommu_iova",
2987 sizeof(struct iova),
2988 0,
2989 SLAB_HWCACHE_ALIGN,
2990 NULL);
2991 if (!iommu_iova_cache) {
2992 printk(KERN_ERR "Couldn't create iova cache\n");
2993 ret = -ENOMEM;
2994 }
2995
2996 return ret;
2997 }
2998
2999 static int __init iommu_init_mempool(void)
3000 {
3001 int ret;
3002 ret = iommu_iova_cache_init();
3003 if (ret)
3004 return ret;
3005
3006 ret = iommu_domain_cache_init();
3007 if (ret)
3008 goto domain_error;
3009
3010 ret = iommu_devinfo_cache_init();
3011 if (!ret)
3012 return ret;
3013
3014 kmem_cache_destroy(iommu_domain_cache);
3015 domain_error:
3016 kmem_cache_destroy(iommu_iova_cache);
3017
3018 return -ENOMEM;
3019 }
3020
3021 static void __init iommu_exit_mempool(void)
3022 {
3023 kmem_cache_destroy(iommu_devinfo_cache);
3024 kmem_cache_destroy(iommu_domain_cache);
3025 kmem_cache_destroy(iommu_iova_cache);
3026
3027 }
3028
3029 static void quirk_ioat_snb_local_iommu(struct pci_dev *pdev)
3030 {
3031 struct dmar_drhd_unit *drhd;
3032 u32 vtbar;
3033 int rc;
3034
3035 /* We know that this device on this chipset has its own IOMMU.
3036 * If we find it under a different IOMMU, then the BIOS is lying
3037 * to us. Hope that the IOMMU for this device is actually
3038 * disabled, and it needs no translation...
3039 */
3040 rc = pci_bus_read_config_dword(pdev->bus, PCI_DEVFN(0, 0), 0xb0, &vtbar);
3041 if (rc) {
3042 /* "can't" happen */
3043 dev_info(&pdev->dev, "failed to run vt-d quirk\n");
3044 return;
3045 }
3046 vtbar &= 0xffff0000;
3047
3048 /* we know that the this iommu should be at offset 0xa000 from vtbar */
3049 drhd = dmar_find_matched_drhd_unit(pdev);
3050 if (WARN_TAINT_ONCE(!drhd || drhd->reg_base_addr - vtbar != 0xa000,
3051 TAINT_FIRMWARE_WORKAROUND,
3052 "BIOS assigned incorrect VT-d unit for Intel(R) QuickData Technology device\n"))
3053 pdev->dev.archdata.iommu = DUMMY_DEVICE_DOMAIN_INFO;
3054 }
3055 DECLARE_PCI_FIXUP_ENABLE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB, quirk_ioat_snb_local_iommu);
3056
3057 static void __init init_no_remapping_devices(void)
3058 {
3059 struct dmar_drhd_unit *drhd;
3060
3061 for_each_drhd_unit(drhd) {
3062 if (!drhd->include_all) {
3063 int i;
3064 for (i = 0; i < drhd->devices_cnt; i++)
3065 if (drhd->devices[i] != NULL)
3066 break;
3067 /* ignore DMAR unit if no pci devices exist */
3068 if (i == drhd->devices_cnt)
3069 drhd->ignored = 1;
3070 }
3071 }
3072
3073 if (dmar_map_gfx)
3074 return;
3075
3076 for_each_drhd_unit(drhd) {
3077 int i;
3078 if (drhd->ignored || drhd->include_all)
3079 continue;
3080
3081 for (i = 0; i < drhd->devices_cnt; i++)
3082 if (drhd->devices[i] &&
3083 !IS_GFX_DEVICE(drhd->devices[i]))
3084 break;
3085
3086 if (i < drhd->devices_cnt)
3087 continue;
3088
3089 /* bypass IOMMU if it is just for gfx devices */
3090 drhd->ignored = 1;
3091 for (i = 0; i < drhd->devices_cnt; i++) {
3092 if (!drhd->devices[i])
3093 continue;
3094 drhd->devices[i]->dev.archdata.iommu = DUMMY_DEVICE_DOMAIN_INFO;
3095 }
3096 }
3097 }
3098
3099 #ifdef CONFIG_SUSPEND
3100 static int init_iommu_hw(void)
3101 {
3102 struct dmar_drhd_unit *drhd;
3103 struct intel_iommu *iommu = NULL;
3104
3105 for_each_active_iommu(iommu, drhd)
3106 if (iommu->qi)
3107 dmar_reenable_qi(iommu);
3108
3109 for_each_active_iommu(iommu, drhd) {
3110 iommu_flush_write_buffer(iommu);
3111
3112 iommu_set_root_entry(iommu);
3113
3114 iommu->flush.flush_context(iommu, 0, 0, 0,
3115 DMA_CCMD_GLOBAL_INVL);
3116 iommu->flush.flush_iotlb(iommu, 0, 0, 0,
3117 DMA_TLB_GLOBAL_FLUSH);
3118 iommu_enable_translation(iommu);
3119 iommu_disable_protect_mem_regions(iommu);
3120 }
3121
3122 return 0;
3123 }
3124
3125 static void iommu_flush_all(void)
3126 {
3127 struct dmar_drhd_unit *drhd;
3128 struct intel_iommu *iommu;
3129
3130 for_each_active_iommu(iommu, drhd) {
3131 iommu->flush.flush_context(iommu, 0, 0, 0,
3132 DMA_CCMD_GLOBAL_INVL);
3133 iommu->flush.flush_iotlb(iommu, 0, 0, 0,
3134 DMA_TLB_GLOBAL_FLUSH);
3135 }
3136 }
3137
3138 static int iommu_suspend(struct sys_device *dev, pm_message_t state)
3139 {
3140 struct dmar_drhd_unit *drhd;
3141 struct intel_iommu *iommu = NULL;
3142 unsigned long flag;
3143
3144 for_each_active_iommu(iommu, drhd) {
3145 iommu->iommu_state = kzalloc(sizeof(u32) * MAX_SR_DMAR_REGS,
3146 GFP_ATOMIC);
3147 if (!iommu->iommu_state)
3148 goto nomem;
3149 }
3150
3151 iommu_flush_all();
3152
3153 for_each_active_iommu(iommu, drhd) {
3154 iommu_disable_translation(iommu);
3155
3156 spin_lock_irqsave(&iommu->register_lock, flag);
3157
3158 iommu->iommu_state[SR_DMAR_FECTL_REG] =
3159 readl(iommu->reg + DMAR_FECTL_REG);
3160 iommu->iommu_state[SR_DMAR_FEDATA_REG] =
3161 readl(iommu->reg + DMAR_FEDATA_REG);
3162 iommu->iommu_state[SR_DMAR_FEADDR_REG] =
3163 readl(iommu->reg + DMAR_FEADDR_REG);
3164 iommu->iommu_state[SR_DMAR_FEUADDR_REG] =
3165 readl(iommu->reg + DMAR_FEUADDR_REG);
3166
3167 spin_unlock_irqrestore(&iommu->register_lock, flag);
3168 }
3169 return 0;
3170
3171 nomem:
3172 for_each_active_iommu(iommu, drhd)
3173 kfree(iommu->iommu_state);
3174
3175 return -ENOMEM;
3176 }
3177
3178 static int iommu_resume(struct sys_device *dev)
3179 {
3180 struct dmar_drhd_unit *drhd;
3181 struct intel_iommu *iommu = NULL;
3182 unsigned long flag;
3183
3184 if (init_iommu_hw()) {
3185 WARN(1, "IOMMU setup failed, DMAR can not resume!\n");
3186 return -EIO;
3187 }
3188
3189 for_each_active_iommu(iommu, drhd) {
3190
3191 spin_lock_irqsave(&iommu->register_lock, flag);
3192
3193 writel(iommu->iommu_state[SR_DMAR_FECTL_REG],
3194 iommu->reg + DMAR_FECTL_REG);
3195 writel(iommu->iommu_state[SR_DMAR_FEDATA_REG],
3196 iommu->reg + DMAR_FEDATA_REG);
3197 writel(iommu->iommu_state[SR_DMAR_FEADDR_REG],
3198 iommu->reg + DMAR_FEADDR_REG);
3199 writel(iommu->iommu_state[SR_DMAR_FEUADDR_REG],
3200 iommu->reg + DMAR_FEUADDR_REG);
3201
3202 spin_unlock_irqrestore(&iommu->register_lock, flag);
3203 }
3204
3205 for_each_active_iommu(iommu, drhd)
3206 kfree(iommu->iommu_state);
3207
3208 return 0;
3209 }
3210
3211 static struct sysdev_class iommu_sysclass = {
3212 .name = "iommu",
3213 .resume = iommu_resume,
3214 .suspend = iommu_suspend,
3215 };
3216
3217 static struct sys_device device_iommu = {
3218 .cls = &iommu_sysclass,
3219 };
3220
3221 static int __init init_iommu_sysfs(void)
3222 {
3223 int error;
3224
3225 error = sysdev_class_register(&iommu_sysclass);
3226 if (error)
3227 return error;
3228
3229 error = sysdev_register(&device_iommu);
3230 if (error)
3231 sysdev_class_unregister(&iommu_sysclass);
3232
3233 return error;
3234 }
3235
3236 #else
3237 static int __init init_iommu_sysfs(void)
3238 {
3239 return 0;
3240 }
3241 #endif /* CONFIG_PM */
3242
3243 /*
3244 * Here we only respond to action of unbound device from driver.
3245 *
3246 * Added device is not attached to its DMAR domain here yet. That will happen
3247 * when mapping the device to iova.
3248 */
3249 static int device_notifier(struct notifier_block *nb,
3250 unsigned long action, void *data)
3251 {
3252 struct device *dev = data;
3253 struct pci_dev *pdev = to_pci_dev(dev);
3254 struct dmar_domain *domain;
3255
3256 if (iommu_no_mapping(dev))
3257 return 0;
3258
3259 domain = find_domain(pdev);
3260 if (!domain)
3261 return 0;
3262
3263 if (action == BUS_NOTIFY_UNBOUND_DRIVER && !iommu_pass_through)
3264 domain_remove_one_dev_info(domain, pdev);
3265
3266 return 0;
3267 }
3268
3269 static struct notifier_block device_nb = {
3270 .notifier_call = device_notifier,
3271 };
3272
3273 int __init intel_iommu_init(void)
3274 {
3275 int ret = 0;
3276 int force_on = 0;
3277
3278 /* VT-d is required for a TXT/tboot launch, so enforce that */
3279 force_on = tboot_force_iommu();
3280
3281 if (dmar_table_init()) {
3282 if (force_on)
3283 panic("tboot: Failed to initialize DMAR table\n");
3284 return -ENODEV;
3285 }
3286
3287 if (dmar_dev_scope_init()) {
3288 if (force_on)
3289 panic("tboot: Failed to initialize DMAR device scope\n");
3290 return -ENODEV;
3291 }
3292
3293 /*
3294 * Check the need for DMA-remapping initialization now.
3295 * Above initialization will also be used by Interrupt-remapping.
3296 */
3297 if (no_iommu || dmar_disabled)
3298 return -ENODEV;
3299
3300 iommu_init_mempool();
3301 dmar_init_reserved_ranges();
3302
3303 init_no_remapping_devices();
3304
3305 ret = init_dmars();
3306 if (ret) {
3307 if (force_on)
3308 panic("tboot: Failed to initialize DMARs\n");
3309 printk(KERN_ERR "IOMMU: dmar init failed\n");
3310 put_iova_domain(&reserved_iova_list);
3311 iommu_exit_mempool();
3312 return ret;
3313 }
3314 printk(KERN_INFO
3315 "PCI-DMA: Intel(R) Virtualization Technology for Directed I/O\n");
3316
3317 init_timer(&unmap_timer);
3318 #ifdef CONFIG_SWIOTLB
3319 swiotlb = 0;
3320 #endif
3321 dma_ops = &intel_dma_ops;
3322
3323 init_iommu_sysfs();
3324
3325 register_iommu(&intel_iommu_ops);
3326
3327 bus_register_notifier(&pci_bus_type, &device_nb);
3328
3329 return 0;
3330 }
3331
3332 static void iommu_detach_dependent_devices(struct intel_iommu *iommu,
3333 struct pci_dev *pdev)
3334 {
3335 struct pci_dev *tmp, *parent;
3336
3337 if (!iommu || !pdev)
3338 return;
3339
3340 /* dependent device detach */
3341 tmp = pci_find_upstream_pcie_bridge(pdev);
3342 /* Secondary interface's bus number and devfn 0 */
3343 if (tmp) {
3344 parent = pdev->bus->self;
3345 while (parent != tmp) {
3346 iommu_detach_dev(iommu, parent->bus->number,
3347 parent->devfn);
3348 parent = parent->bus->self;
3349 }
3350 if (pci_is_pcie(tmp)) /* this is a PCIe-to-PCI bridge */
3351 iommu_detach_dev(iommu,
3352 tmp->subordinate->number, 0);
3353 else /* this is a legacy PCI bridge */
3354 iommu_detach_dev(iommu, tmp->bus->number,
3355 tmp->devfn);
3356 }
3357 }
3358
3359 static void domain_remove_one_dev_info(struct dmar_domain *domain,
3360 struct pci_dev *pdev)
3361 {
3362 struct device_domain_info *info;
3363 struct intel_iommu *iommu;
3364 unsigned long flags;
3365 int found = 0;
3366 struct list_head *entry, *tmp;
3367
3368 iommu = device_to_iommu(pci_domain_nr(pdev->bus), pdev->bus->number,
3369 pdev->devfn);
3370 if (!iommu)
3371 return;
3372
3373 spin_lock_irqsave(&device_domain_lock, flags);
3374 list_for_each_safe(entry, tmp, &domain->devices) {
3375 info = list_entry(entry, struct device_domain_info, link);
3376 /* No need to compare PCI domain; it has to be the same */
3377 if (info->bus == pdev->bus->number &&
3378 info->devfn == pdev->devfn) {
3379 list_del(&info->link);
3380 list_del(&info->global);
3381 if (info->dev)
3382 info->dev->dev.archdata.iommu = NULL;
3383 spin_unlock_irqrestore(&device_domain_lock, flags);
3384
3385 iommu_disable_dev_iotlb(info);
3386 iommu_detach_dev(iommu, info->bus, info->devfn);
3387 iommu_detach_dependent_devices(iommu, pdev);
3388 free_devinfo_mem(info);
3389
3390 spin_lock_irqsave(&device_domain_lock, flags);
3391
3392 if (found)
3393 break;
3394 else
3395 continue;
3396 }
3397
3398 /* if there is no other devices under the same iommu
3399 * owned by this domain, clear this iommu in iommu_bmp
3400 * update iommu count and coherency
3401 */
3402 if (iommu == device_to_iommu(info->segment, info->bus,
3403 info->devfn))
3404 found = 1;
3405 }
3406
3407 if (found == 0) {
3408 unsigned long tmp_flags;
3409 spin_lock_irqsave(&domain->iommu_lock, tmp_flags);
3410 clear_bit(iommu->seq_id, &domain->iommu_bmp);
3411 domain->iommu_count--;
3412 domain_update_iommu_cap(domain);
3413 spin_unlock_irqrestore(&domain->iommu_lock, tmp_flags);
3414 }
3415
3416 spin_unlock_irqrestore(&device_domain_lock, flags);
3417 }
3418
3419 static void vm_domain_remove_all_dev_info(struct dmar_domain *domain)
3420 {
3421 struct device_domain_info *info;
3422 struct intel_iommu *iommu;
3423 unsigned long flags1, flags2;
3424
3425 spin_lock_irqsave(&device_domain_lock, flags1);
3426 while (!list_empty(&domain->devices)) {
3427 info = list_entry(domain->devices.next,
3428 struct device_domain_info, link);
3429 list_del(&info->link);
3430 list_del(&info->global);
3431 if (info->dev)
3432 info->dev->dev.archdata.iommu = NULL;
3433
3434 spin_unlock_irqrestore(&device_domain_lock, flags1);
3435
3436 iommu_disable_dev_iotlb(info);
3437 iommu = device_to_iommu(info->segment, info->bus, info->devfn);
3438 iommu_detach_dev(iommu, info->bus, info->devfn);
3439 iommu_detach_dependent_devices(iommu, info->dev);
3440
3441 /* clear this iommu in iommu_bmp, update iommu count
3442 * and capabilities
3443 */
3444 spin_lock_irqsave(&domain->iommu_lock, flags2);
3445 if (test_and_clear_bit(iommu->seq_id,
3446 &domain->iommu_bmp)) {
3447 domain->iommu_count--;
3448 domain_update_iommu_cap(domain);
3449 }
3450 spin_unlock_irqrestore(&domain->iommu_lock, flags2);
3451
3452 free_devinfo_mem(info);
3453 spin_lock_irqsave(&device_domain_lock, flags1);
3454 }
3455 spin_unlock_irqrestore(&device_domain_lock, flags1);
3456 }
3457
3458 /* domain id for virtual machine, it won't be set in context */
3459 static unsigned long vm_domid;
3460
3461 static struct dmar_domain *iommu_alloc_vm_domain(void)
3462 {
3463 struct dmar_domain *domain;
3464
3465 domain = alloc_domain_mem();
3466 if (!domain)
3467 return NULL;
3468
3469 domain->id = vm_domid++;
3470 domain->nid = -1;
3471 memset(&domain->iommu_bmp, 0, sizeof(unsigned long));
3472 domain->flags = DOMAIN_FLAG_VIRTUAL_MACHINE;
3473
3474 return domain;
3475 }
3476
3477 static int md_domain_init(struct dmar_domain *domain, int guest_width)
3478 {
3479 int adjust_width;
3480
3481 init_iova_domain(&domain->iovad, DMA_32BIT_PFN);
3482 spin_lock_init(&domain->iommu_lock);
3483
3484 domain_reserve_special_ranges(domain);
3485
3486 /* calculate AGAW */
3487 domain->gaw = guest_width;
3488 adjust_width = guestwidth_to_adjustwidth(guest_width);
3489 domain->agaw = width_to_agaw(adjust_width);
3490
3491 INIT_LIST_HEAD(&domain->devices);
3492
3493 domain->iommu_count = 0;
3494 domain->iommu_coherency = 0;
3495 domain->iommu_snooping = 0;
3496 domain->max_addr = 0;
3497 domain->nid = -1;
3498
3499 /* always allocate the top pgd */
3500 domain->pgd = (struct dma_pte *)alloc_pgtable_page(domain->nid);
3501 if (!domain->pgd)
3502 return -ENOMEM;
3503 domain_flush_cache(domain, domain->pgd, PAGE_SIZE);
3504 return 0;
3505 }
3506
3507 static void iommu_free_vm_domain(struct dmar_domain *domain)
3508 {
3509 unsigned long flags;
3510 struct dmar_drhd_unit *drhd;
3511 struct intel_iommu *iommu;
3512 unsigned long i;
3513 unsigned long ndomains;
3514
3515 for_each_drhd_unit(drhd) {
3516 if (drhd->ignored)
3517 continue;
3518 iommu = drhd->iommu;
3519
3520 ndomains = cap_ndoms(iommu->cap);
3521 for_each_set_bit(i, iommu->domain_ids, ndomains) {
3522 if (iommu->domains[i] == domain) {
3523 spin_lock_irqsave(&iommu->lock, flags);
3524 clear_bit(i, iommu->domain_ids);
3525 iommu->domains[i] = NULL;
3526 spin_unlock_irqrestore(&iommu->lock, flags);
3527 break;
3528 }
3529 }
3530 }
3531 }
3532
3533 static void vm_domain_exit(struct dmar_domain *domain)
3534 {
3535 /* Domain 0 is reserved, so dont process it */
3536 if (!domain)
3537 return;
3538
3539 vm_domain_remove_all_dev_info(domain);
3540 /* destroy iovas */
3541 put_iova_domain(&domain->iovad);
3542
3543 /* clear ptes */
3544 dma_pte_clear_range(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
3545
3546 /* free page tables */
3547 dma_pte_free_pagetable(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
3548
3549 iommu_free_vm_domain(domain);
3550 free_domain_mem(domain);
3551 }
3552
3553 static int intel_iommu_domain_init(struct iommu_domain *domain)
3554 {
3555 struct dmar_domain *dmar_domain;
3556
3557 dmar_domain = iommu_alloc_vm_domain();
3558 if (!dmar_domain) {
3559 printk(KERN_ERR
3560 "intel_iommu_domain_init: dmar_domain == NULL\n");
3561 return -ENOMEM;
3562 }
3563 if (md_domain_init(dmar_domain, DEFAULT_DOMAIN_ADDRESS_WIDTH)) {
3564 printk(KERN_ERR
3565 "intel_iommu_domain_init() failed\n");
3566 vm_domain_exit(dmar_domain);
3567 return -ENOMEM;
3568 }
3569 domain->priv = dmar_domain;
3570
3571 return 0;
3572 }
3573
3574 static void intel_iommu_domain_destroy(struct iommu_domain *domain)
3575 {
3576 struct dmar_domain *dmar_domain = domain->priv;
3577
3578 domain->priv = NULL;
3579 vm_domain_exit(dmar_domain);
3580 }
3581
3582 static int intel_iommu_attach_device(struct iommu_domain *domain,
3583 struct device *dev)
3584 {
3585 struct dmar_domain *dmar_domain = domain->priv;
3586 struct pci_dev *pdev = to_pci_dev(dev);
3587 struct intel_iommu *iommu;
3588 int addr_width;
3589
3590 /* normally pdev is not mapped */
3591 if (unlikely(domain_context_mapped(pdev))) {
3592 struct dmar_domain *old_domain;
3593
3594 old_domain = find_domain(pdev);
3595 if (old_domain) {
3596 if (dmar_domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE ||
3597 dmar_domain->flags & DOMAIN_FLAG_STATIC_IDENTITY)
3598 domain_remove_one_dev_info(old_domain, pdev);
3599 else
3600 domain_remove_dev_info(old_domain);
3601 }
3602 }
3603
3604 iommu = device_to_iommu(pci_domain_nr(pdev->bus), pdev->bus->number,
3605 pdev->devfn);
3606 if (!iommu)
3607 return -ENODEV;
3608
3609 /* check if this iommu agaw is sufficient for max mapped address */
3610 addr_width = agaw_to_width(iommu->agaw);
3611 if (addr_width > cap_mgaw(iommu->cap))
3612 addr_width = cap_mgaw(iommu->cap);
3613
3614 if (dmar_domain->max_addr > (1LL << addr_width)) {
3615 printk(KERN_ERR "%s: iommu width (%d) is not "
3616 "sufficient for the mapped address (%llx)\n",
3617 __func__, addr_width, dmar_domain->max_addr);
3618 return -EFAULT;
3619 }
3620 dmar_domain->gaw = addr_width;
3621
3622 /*
3623 * Knock out extra levels of page tables if necessary
3624 */
3625 while (iommu->agaw < dmar_domain->agaw) {
3626 struct dma_pte *pte;
3627
3628 pte = dmar_domain->pgd;
3629 if (dma_pte_present(pte)) {
3630 free_pgtable_page(dmar_domain->pgd);
3631 dmar_domain->pgd = (struct dma_pte *)
3632 phys_to_virt(dma_pte_addr(pte));
3633 }
3634 dmar_domain->agaw--;
3635 }
3636
3637 return domain_add_dev_info(dmar_domain, pdev, CONTEXT_TT_MULTI_LEVEL);
3638 }
3639
3640 static void intel_iommu_detach_device(struct iommu_domain *domain,
3641 struct device *dev)
3642 {
3643 struct dmar_domain *dmar_domain = domain->priv;
3644 struct pci_dev *pdev = to_pci_dev(dev);
3645
3646 domain_remove_one_dev_info(dmar_domain, pdev);
3647 }
3648
3649 static int intel_iommu_map(struct iommu_domain *domain,
3650 unsigned long iova, phys_addr_t hpa,
3651 int gfp_order, int iommu_prot)
3652 {
3653 struct dmar_domain *dmar_domain = domain->priv;
3654 u64 max_addr;
3655 int prot = 0;
3656 size_t size;
3657 int ret;
3658
3659 if (iommu_prot & IOMMU_READ)
3660 prot |= DMA_PTE_READ;
3661 if (iommu_prot & IOMMU_WRITE)
3662 prot |= DMA_PTE_WRITE;
3663 if ((iommu_prot & IOMMU_CACHE) && dmar_domain->iommu_snooping)
3664 prot |= DMA_PTE_SNP;
3665
3666 size = PAGE_SIZE << gfp_order;
3667 max_addr = iova + size;
3668 if (dmar_domain->max_addr < max_addr) {
3669 u64 end;
3670
3671 /* check if minimum agaw is sufficient for mapped address */
3672 end = __DOMAIN_MAX_ADDR(dmar_domain->gaw) + 1;
3673 if (end < max_addr) {
3674 printk(KERN_ERR "%s: iommu width (%d) is not "
3675 "sufficient for the mapped address (%llx)\n",
3676 __func__, dmar_domain->gaw, max_addr);
3677 return -EFAULT;
3678 }
3679 dmar_domain->max_addr = max_addr;
3680 }
3681 /* Round up size to next multiple of PAGE_SIZE, if it and
3682 the low bits of hpa would take us onto the next page */
3683 size = aligned_nrpages(hpa, size);
3684 ret = domain_pfn_mapping(dmar_domain, iova >> VTD_PAGE_SHIFT,
3685 hpa >> VTD_PAGE_SHIFT, size, prot);
3686 return ret;
3687 }
3688
3689 static int intel_iommu_unmap(struct iommu_domain *domain,
3690 unsigned long iova, int gfp_order)
3691 {
3692 struct dmar_domain *dmar_domain = domain->priv;
3693 size_t size = PAGE_SIZE << gfp_order;
3694
3695 dma_pte_clear_range(dmar_domain, iova >> VTD_PAGE_SHIFT,
3696 (iova + size - 1) >> VTD_PAGE_SHIFT);
3697
3698 if (dmar_domain->max_addr == iova + size)
3699 dmar_domain->max_addr = iova;
3700
3701 return gfp_order;
3702 }
3703
3704 static phys_addr_t intel_iommu_iova_to_phys(struct iommu_domain *domain,
3705 unsigned long iova)
3706 {
3707 struct dmar_domain *dmar_domain = domain->priv;
3708 struct dma_pte *pte;
3709 u64 phys = 0;
3710
3711 pte = pfn_to_dma_pte(dmar_domain, iova >> VTD_PAGE_SHIFT);
3712 if (pte)
3713 phys = dma_pte_addr(pte);
3714
3715 return phys;
3716 }
3717
3718 static int intel_iommu_domain_has_cap(struct iommu_domain *domain,
3719 unsigned long cap)
3720 {
3721 struct dmar_domain *dmar_domain = domain->priv;
3722
3723 if (cap == IOMMU_CAP_CACHE_COHERENCY)
3724 return dmar_domain->iommu_snooping;
3725 if (cap == IOMMU_CAP_INTR_REMAP)
3726 return intr_remapping_enabled;
3727
3728 return 0;
3729 }
3730
3731 static struct iommu_ops intel_iommu_ops = {
3732 .domain_init = intel_iommu_domain_init,
3733 .domain_destroy = intel_iommu_domain_destroy,
3734 .attach_dev = intel_iommu_attach_device,
3735 .detach_dev = intel_iommu_detach_device,
3736 .map = intel_iommu_map,
3737 .unmap = intel_iommu_unmap,
3738 .iova_to_phys = intel_iommu_iova_to_phys,
3739 .domain_has_cap = intel_iommu_domain_has_cap,
3740 };
3741
3742 static void __devinit quirk_iommu_rwbf(struct pci_dev *dev)
3743 {
3744 /*
3745 * Mobile 4 Series Chipset neglects to set RWBF capability,
3746 * but needs it:
3747 */
3748 printk(KERN_INFO "DMAR: Forcing write-buffer flush capability\n");
3749 rwbf_quirk = 1;
3750
3751 /* https://bugzilla.redhat.com/show_bug.cgi?id=538163 */
3752 if (dev->revision == 0x07) {
3753 printk(KERN_INFO "DMAR: Disabling IOMMU for graphics on this chipset\n");
3754 dmar_map_gfx = 0;
3755 }
3756 }
3757
3758 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x2a40, quirk_iommu_rwbf);
3759
3760 #define GGC 0x52
3761 #define GGC_MEMORY_SIZE_MASK (0xf << 8)
3762 #define GGC_MEMORY_SIZE_NONE (0x0 << 8)
3763 #define GGC_MEMORY_SIZE_1M (0x1 << 8)
3764 #define GGC_MEMORY_SIZE_2M (0x3 << 8)
3765 #define GGC_MEMORY_VT_ENABLED (0x8 << 8)
3766 #define GGC_MEMORY_SIZE_2M_VT (0x9 << 8)
3767 #define GGC_MEMORY_SIZE_3M_VT (0xa << 8)
3768 #define GGC_MEMORY_SIZE_4M_VT (0xb << 8)
3769
3770 static void __devinit quirk_calpella_no_shadow_gtt(struct pci_dev *dev)
3771 {
3772 unsigned short ggc;
3773
3774 if (pci_read_config_word(dev, GGC, &ggc))
3775 return;
3776
3777 if (!(ggc & GGC_MEMORY_VT_ENABLED)) {
3778 printk(KERN_INFO "DMAR: BIOS has allocated no shadow GTT; disabling IOMMU for graphics\n");
3779 dmar_map_gfx = 0;
3780 }
3781 }
3782 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x0040, quirk_calpella_no_shadow_gtt);
3783 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x0044, quirk_calpella_no_shadow_gtt);
3784 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x0062, quirk_calpella_no_shadow_gtt);
3785 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x006a, quirk_calpella_no_shadow_gtt);
3786
3787 /* On Tylersburg chipsets, some BIOSes have been known to enable the
3788 ISOCH DMAR unit for the Azalia sound device, but not give it any
3789 TLB entries, which causes it to deadlock. Check for that. We do
3790 this in a function called from init_dmars(), instead of in a PCI
3791 quirk, because we don't want to print the obnoxious "BIOS broken"
3792 message if VT-d is actually disabled.
3793 */
3794 static void __init check_tylersburg_isoch(void)
3795 {
3796 struct pci_dev *pdev;
3797 uint32_t vtisochctrl;
3798
3799 /* If there's no Azalia in the system anyway, forget it. */
3800 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 0x3a3e, NULL);
3801 if (!pdev)
3802 return;
3803 pci_dev_put(pdev);
3804
3805 /* System Management Registers. Might be hidden, in which case
3806 we can't do the sanity check. But that's OK, because the
3807 known-broken BIOSes _don't_ actually hide it, so far. */
3808 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 0x342e, NULL);
3809 if (!pdev)
3810 return;
3811
3812 if (pci_read_config_dword(pdev, 0x188, &vtisochctrl)) {
3813 pci_dev_put(pdev);
3814 return;
3815 }
3816
3817 pci_dev_put(pdev);
3818
3819 /* If Azalia DMA is routed to the non-isoch DMAR unit, fine. */
3820 if (vtisochctrl & 1)
3821 return;
3822
3823 /* Drop all bits other than the number of TLB entries */
3824 vtisochctrl &= 0x1c;
3825
3826 /* If we have the recommended number of TLB entries (16), fine. */
3827 if (vtisochctrl == 0x10)
3828 return;
3829
3830 /* Zero TLB entries? You get to ride the short bus to school. */
3831 if (!vtisochctrl) {
3832 WARN(1, "Your BIOS is broken; DMA routed to ISOCH DMAR unit but no TLB space.\n"
3833 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
3834 dmi_get_system_info(DMI_BIOS_VENDOR),
3835 dmi_get_system_info(DMI_BIOS_VERSION),
3836 dmi_get_system_info(DMI_PRODUCT_VERSION));
3837 iommu_identity_mapping |= IDENTMAP_AZALIA;
3838 return;
3839 }
3840
3841 printk(KERN_WARNING "DMAR: Recommended TLB entries for ISOCH unit is 16; your BIOS set %d\n",
3842 vtisochctrl);
3843 }
Linux with added FPC-III support