intel-iommu: Warn about unmatched unmap requests
[linux-2.6.git] / drivers / pci / intel-iommu.c
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 <asm/cacheflush.h>
41 #include <asm/iommu.h>
42 #include "pci.h"
43
44 #define ROOT_SIZE               VTD_PAGE_SIZE
45 #define CONTEXT_SIZE            VTD_PAGE_SIZE
46
47 #define IS_GFX_DEVICE(pdev) ((pdev->class >> 16) == PCI_BASE_CLASS_DISPLAY)
48 #define IS_ISA_DEVICE(pdev) ((pdev->class >> 8) == PCI_CLASS_BRIDGE_ISA)
49
50 #define IOAPIC_RANGE_START      (0xfee00000)
51 #define IOAPIC_RANGE_END        (0xfeefffff)
52 #define IOVA_START_ADDR         (0x1000)
53
54 #define DEFAULT_DOMAIN_ADDRESS_WIDTH 48
55
56 #define MAX_AGAW_WIDTH 64
57
58 #define DOMAIN_MAX_ADDR(gaw) ((((u64)1) << gaw) - 1)
59 #define DOMAIN_MAX_PFN(gaw)  ((((u64)1) << (gaw-VTD_PAGE_SHIFT)) - 1)
60
61 #define IOVA_PFN(addr)          ((addr) >> PAGE_SHIFT)
62 #define DMA_32BIT_PFN           IOVA_PFN(DMA_BIT_MASK(32))
63 #define DMA_64BIT_PFN           IOVA_PFN(DMA_BIT_MASK(64))
64
65
66 /* VT-d pages must always be _smaller_ than MM pages. Otherwise things
67    are never going to work. */
68 static inline unsigned long dma_to_mm_pfn(unsigned long dma_pfn)
69 {
70         return dma_pfn >> (PAGE_SHIFT - VTD_PAGE_SHIFT);
71 }
72
73 static inline unsigned long mm_to_dma_pfn(unsigned long mm_pfn)
74 {
75         return mm_pfn << (PAGE_SHIFT - VTD_PAGE_SHIFT);
76 }
77 static inline unsigned long page_to_dma_pfn(struct page *pg)
78 {
79         return mm_to_dma_pfn(page_to_pfn(pg));
80 }
81 static inline unsigned long virt_to_dma_pfn(void *p)
82 {
83         return page_to_dma_pfn(virt_to_page(p));
84 }
85
86 /* global iommu list, set NULL for ignored DMAR units */
87 static struct intel_iommu **g_iommus;
88
89 static int rwbf_quirk;
90
91 /*
92  * 0: Present
93  * 1-11: Reserved
94  * 12-63: Context Ptr (12 - (haw-1))
95  * 64-127: Reserved
96  */
97 struct root_entry {
98         u64     val;
99         u64     rsvd1;
100 };
101 #define ROOT_ENTRY_NR (VTD_PAGE_SIZE/sizeof(struct root_entry))
102 static inline bool root_present(struct root_entry *root)
103 {
104         return (root->val & 1);
105 }
106 static inline void set_root_present(struct root_entry *root)
107 {
108         root->val |= 1;
109 }
110 static inline void set_root_value(struct root_entry *root, unsigned long value)
111 {
112         root->val |= value & VTD_PAGE_MASK;
113 }
114
115 static inline struct context_entry *
116 get_context_addr_from_root(struct root_entry *root)
117 {
118         return (struct context_entry *)
119                 (root_present(root)?phys_to_virt(
120                 root->val & VTD_PAGE_MASK) :
121                 NULL);
122 }
123
124 /*
125  * low 64 bits:
126  * 0: present
127  * 1: fault processing disable
128  * 2-3: translation type
129  * 12-63: address space root
130  * high 64 bits:
131  * 0-2: address width
132  * 3-6: aval
133  * 8-23: domain id
134  */
135 struct context_entry {
136         u64 lo;
137         u64 hi;
138 };
139
140 static inline bool context_present(struct context_entry *context)
141 {
142         return (context->lo & 1);
143 }
144 static inline void context_set_present(struct context_entry *context)
145 {
146         context->lo |= 1;
147 }
148
149 static inline void context_set_fault_enable(struct context_entry *context)
150 {
151         context->lo &= (((u64)-1) << 2) | 1;
152 }
153
154 static inline void context_set_translation_type(struct context_entry *context,
155                                                 unsigned long value)
156 {
157         context->lo &= (((u64)-1) << 4) | 3;
158         context->lo |= (value & 3) << 2;
159 }
160
161 static inline void context_set_address_root(struct context_entry *context,
162                                             unsigned long value)
163 {
164         context->lo |= value & VTD_PAGE_MASK;
165 }
166
167 static inline void context_set_address_width(struct context_entry *context,
168                                              unsigned long value)
169 {
170         context->hi |= value & 7;
171 }
172
173 static inline void context_set_domain_id(struct context_entry *context,
174                                          unsigned long value)
175 {
176         context->hi |= (value & ((1 << 16) - 1)) << 8;
177 }
178
179 static inline void context_clear_entry(struct context_entry *context)
180 {
181         context->lo = 0;
182         context->hi = 0;
183 }
184
185 /*
186  * 0: readable
187  * 1: writable
188  * 2-6: reserved
189  * 7: super page
190  * 8-10: available
191  * 11: snoop behavior
192  * 12-63: Host physcial address
193  */
194 struct dma_pte {
195         u64 val;
196 };
197
198 static inline void dma_clear_pte(struct dma_pte *pte)
199 {
200         pte->val = 0;
201 }
202
203 static inline void dma_set_pte_readable(struct dma_pte *pte)
204 {
205         pte->val |= DMA_PTE_READ;
206 }
207
208 static inline void dma_set_pte_writable(struct dma_pte *pte)
209 {
210         pte->val |= DMA_PTE_WRITE;
211 }
212
213 static inline void dma_set_pte_snp(struct dma_pte *pte)
214 {
215         pte->val |= DMA_PTE_SNP;
216 }
217
218 static inline void dma_set_pte_prot(struct dma_pte *pte, unsigned long prot)
219 {
220         pte->val = (pte->val & ~3) | (prot & 3);
221 }
222
223 static inline u64 dma_pte_addr(struct dma_pte *pte)
224 {
225 #ifdef CONFIG_64BIT
226         return pte->val & VTD_PAGE_MASK;
227 #else
228         /* Must have a full atomic 64-bit read */
229         return  __cmpxchg64(pte, 0ULL, 0ULL) & VTD_PAGE_MASK;
230 #endif
231 }
232
233 static inline void dma_set_pte_pfn(struct dma_pte *pte, unsigned long pfn)
234 {
235         pte->val |= (uint64_t)pfn << VTD_PAGE_SHIFT;
236 }
237
238 static inline bool dma_pte_present(struct dma_pte *pte)
239 {
240         return (pte->val & 3) != 0;
241 }
242
243 /*
244  * This domain is a statically identity mapping domain.
245  *      1. This domain creats a static 1:1 mapping to all usable memory.
246  *      2. It maps to each iommu if successful.
247  *      3. Each iommu mapps to this domain if successful.
248  */
249 struct dmar_domain *si_domain;
250
251 /* devices under the same p2p bridge are owned in one domain */
252 #define DOMAIN_FLAG_P2P_MULTIPLE_DEVICES (1 << 0)
253
254 /* domain represents a virtual machine, more than one devices
255  * across iommus may be owned in one domain, e.g. kvm guest.
256  */
257 #define DOMAIN_FLAG_VIRTUAL_MACHINE     (1 << 1)
258
259 /* si_domain contains mulitple devices */
260 #define DOMAIN_FLAG_STATIC_IDENTITY     (1 << 2)
261
262 struct dmar_domain {
263         int     id;                     /* domain id */
264         unsigned long iommu_bmp;        /* bitmap of iommus this domain uses*/
265
266         struct list_head devices;       /* all devices' list */
267         struct iova_domain iovad;       /* iova's that belong to this domain */
268
269         struct dma_pte  *pgd;           /* virtual address */
270         int             gaw;            /* max guest address width */
271
272         /* adjusted guest address width, 0 is level 2 30-bit */
273         int             agaw;
274
275         int             flags;          /* flags to find out type of domain */
276
277         int             iommu_coherency;/* indicate coherency of iommu access */
278         int             iommu_snooping; /* indicate snooping control feature*/
279         int             iommu_count;    /* reference count of iommu */
280         spinlock_t      iommu_lock;     /* protect iommu set in domain */
281         u64             max_addr;       /* maximum mapped address */
282 };
283
284 /* PCI domain-device relationship */
285 struct device_domain_info {
286         struct list_head link;  /* link to domain siblings */
287         struct list_head global; /* link to global list */
288         int segment;            /* PCI domain */
289         u8 bus;                 /* PCI bus number */
290         u8 devfn;               /* PCI devfn number */
291         struct pci_dev *dev; /* it's NULL for PCIE-to-PCI bridge */
292         struct intel_iommu *iommu; /* IOMMU used by this device */
293         struct dmar_domain *domain; /* pointer to domain */
294 };
295
296 static void flush_unmaps_timeout(unsigned long data);
297
298 DEFINE_TIMER(unmap_timer,  flush_unmaps_timeout, 0, 0);
299
300 #define HIGH_WATER_MARK 250
301 struct deferred_flush_tables {
302         int next;
303         struct iova *iova[HIGH_WATER_MARK];
304         struct dmar_domain *domain[HIGH_WATER_MARK];
305 };
306
307 static struct deferred_flush_tables *deferred_flush;
308
309 /* bitmap for indexing intel_iommus */
310 static int g_num_of_iommus;
311
312 static DEFINE_SPINLOCK(async_umap_flush_lock);
313 static LIST_HEAD(unmaps_to_do);
314
315 static int timer_on;
316 static long list_size;
317
318 static void domain_remove_dev_info(struct dmar_domain *domain);
319
320 #ifdef CONFIG_DMAR_DEFAULT_ON
321 int dmar_disabled = 0;
322 #else
323 int dmar_disabled = 1;
324 #endif /*CONFIG_DMAR_DEFAULT_ON*/
325
326 static int __initdata dmar_map_gfx = 1;
327 static int dmar_forcedac;
328 static int intel_iommu_strict;
329
330 #define DUMMY_DEVICE_DOMAIN_INFO ((struct device_domain_info *)(-1))
331 static DEFINE_SPINLOCK(device_domain_lock);
332 static LIST_HEAD(device_domain_list);
333
334 static struct iommu_ops intel_iommu_ops;
335
336 static int __init intel_iommu_setup(char *str)
337 {
338         if (!str)
339                 return -EINVAL;
340         while (*str) {
341                 if (!strncmp(str, "on", 2)) {
342                         dmar_disabled = 0;
343                         printk(KERN_INFO "Intel-IOMMU: enabled\n");
344                 } else if (!strncmp(str, "off", 3)) {
345                         dmar_disabled = 1;
346                         printk(KERN_INFO "Intel-IOMMU: disabled\n");
347                 } else if (!strncmp(str, "igfx_off", 8)) {
348                         dmar_map_gfx = 0;
349                         printk(KERN_INFO
350                                 "Intel-IOMMU: disable GFX device mapping\n");
351                 } else if (!strncmp(str, "forcedac", 8)) {
352                         printk(KERN_INFO
353                                 "Intel-IOMMU: Forcing DAC for PCI devices\n");
354                         dmar_forcedac = 1;
355                 } else if (!strncmp(str, "strict", 6)) {
356                         printk(KERN_INFO
357                                 "Intel-IOMMU: disable batched IOTLB flush\n");
358                         intel_iommu_strict = 1;
359                 }
360
361                 str += strcspn(str, ",");
362                 while (*str == ',')
363                         str++;
364         }
365         return 0;
366 }
367 __setup("intel_iommu=", intel_iommu_setup);
368
369 static struct kmem_cache *iommu_domain_cache;
370 static struct kmem_cache *iommu_devinfo_cache;
371 static struct kmem_cache *iommu_iova_cache;
372
373 static inline void *iommu_kmem_cache_alloc(struct kmem_cache *cachep)
374 {
375         unsigned int flags;
376         void *vaddr;
377
378         /* trying to avoid low memory issues */
379         flags = current->flags & PF_MEMALLOC;
380         current->flags |= PF_MEMALLOC;
381         vaddr = kmem_cache_alloc(cachep, GFP_ATOMIC);
382         current->flags &= (~PF_MEMALLOC | flags);
383         return vaddr;
384 }
385
386
387 static inline void *alloc_pgtable_page(void)
388 {
389         unsigned int flags;
390         void *vaddr;
391
392         /* trying to avoid low memory issues */
393         flags = current->flags & PF_MEMALLOC;
394         current->flags |= PF_MEMALLOC;
395         vaddr = (void *)get_zeroed_page(GFP_ATOMIC);
396         current->flags &= (~PF_MEMALLOC | flags);
397         return vaddr;
398 }
399
400 static inline void free_pgtable_page(void *vaddr)
401 {
402         free_page((unsigned long)vaddr);
403 }
404
405 static inline void *alloc_domain_mem(void)
406 {
407         return iommu_kmem_cache_alloc(iommu_domain_cache);
408 }
409
410 static void free_domain_mem(void *vaddr)
411 {
412         kmem_cache_free(iommu_domain_cache, vaddr);
413 }
414
415 static inline void * alloc_devinfo_mem(void)
416 {
417         return iommu_kmem_cache_alloc(iommu_devinfo_cache);
418 }
419
420 static inline void free_devinfo_mem(void *vaddr)
421 {
422         kmem_cache_free(iommu_devinfo_cache, vaddr);
423 }
424
425 struct iova *alloc_iova_mem(void)
426 {
427         return iommu_kmem_cache_alloc(iommu_iova_cache);
428 }
429
430 void free_iova_mem(struct iova *iova)
431 {
432         kmem_cache_free(iommu_iova_cache, iova);
433 }
434
435
436 static inline int width_to_agaw(int width);
437
438 static int __iommu_calculate_agaw(struct intel_iommu *iommu, int max_gaw)
439 {
440         unsigned long sagaw;
441         int agaw = -1;
442
443         sagaw = cap_sagaw(iommu->cap);
444         for (agaw = width_to_agaw(max_gaw);
445              agaw >= 0; agaw--) {
446                 if (test_bit(agaw, &sagaw))
447                         break;
448         }
449
450         return agaw;
451 }
452
453 /*
454  * Calculate max SAGAW for each iommu.
455  */
456 int iommu_calculate_max_sagaw(struct intel_iommu *iommu)
457 {
458         return __iommu_calculate_agaw(iommu, MAX_AGAW_WIDTH);
459 }
460
461 /*
462  * calculate agaw for each iommu.
463  * "SAGAW" may be different across iommus, use a default agaw, and
464  * get a supported less agaw for iommus that don't support the default agaw.
465  */
466 int iommu_calculate_agaw(struct intel_iommu *iommu)
467 {
468         return __iommu_calculate_agaw(iommu, DEFAULT_DOMAIN_ADDRESS_WIDTH);
469 }
470
471 /* This functionin only returns single iommu in a domain */
472 static struct intel_iommu *domain_get_iommu(struct dmar_domain *domain)
473 {
474         int iommu_id;
475
476         /* si_domain and vm domain should not get here. */
477         BUG_ON(domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE);
478         BUG_ON(domain->flags & DOMAIN_FLAG_STATIC_IDENTITY);
479
480         iommu_id = find_first_bit(&domain->iommu_bmp, g_num_of_iommus);
481         if (iommu_id < 0 || iommu_id >= g_num_of_iommus)
482                 return NULL;
483
484         return g_iommus[iommu_id];
485 }
486
487 static void domain_update_iommu_coherency(struct dmar_domain *domain)
488 {
489         int i;
490
491         domain->iommu_coherency = 1;
492
493         i = find_first_bit(&domain->iommu_bmp, g_num_of_iommus);
494         for (; i < g_num_of_iommus; ) {
495                 if (!ecap_coherent(g_iommus[i]->ecap)) {
496                         domain->iommu_coherency = 0;
497                         break;
498                 }
499                 i = find_next_bit(&domain->iommu_bmp, g_num_of_iommus, i+1);
500         }
501 }
502
503 static void domain_update_iommu_snooping(struct dmar_domain *domain)
504 {
505         int i;
506
507         domain->iommu_snooping = 1;
508
509         i = find_first_bit(&domain->iommu_bmp, g_num_of_iommus);
510         for (; i < g_num_of_iommus; ) {
511                 if (!ecap_sc_support(g_iommus[i]->ecap)) {
512                         domain->iommu_snooping = 0;
513                         break;
514                 }
515                 i = find_next_bit(&domain->iommu_bmp, g_num_of_iommus, i+1);
516         }
517 }
518
519 /* Some capabilities may be different across iommus */
520 static void domain_update_iommu_cap(struct dmar_domain *domain)
521 {
522         domain_update_iommu_coherency(domain);
523         domain_update_iommu_snooping(domain);
524 }
525
526 static struct intel_iommu *device_to_iommu(int segment, u8 bus, u8 devfn)
527 {
528         struct dmar_drhd_unit *drhd = NULL;
529         int i;
530
531         for_each_drhd_unit(drhd) {
532                 if (drhd->ignored)
533                         continue;
534                 if (segment != drhd->segment)
535                         continue;
536
537                 for (i = 0; i < drhd->devices_cnt; i++) {
538                         if (drhd->devices[i] &&
539                             drhd->devices[i]->bus->number == bus &&
540                             drhd->devices[i]->devfn == devfn)
541                                 return drhd->iommu;
542                         if (drhd->devices[i] &&
543                             drhd->devices[i]->subordinate &&
544                             drhd->devices[i]->subordinate->number <= bus &&
545                             drhd->devices[i]->subordinate->subordinate >= bus)
546                                 return drhd->iommu;
547                 }
548
549                 if (drhd->include_all)
550                         return drhd->iommu;
551         }
552
553         return NULL;
554 }
555
556 static void domain_flush_cache(struct dmar_domain *domain,
557                                void *addr, int size)
558 {
559         if (!domain->iommu_coherency)
560                 clflush_cache_range(addr, size);
561 }
562
563 /* Gets context entry for a given bus and devfn */
564 static struct context_entry * device_to_context_entry(struct intel_iommu *iommu,
565                 u8 bus, u8 devfn)
566 {
567         struct root_entry *root;
568         struct context_entry *context;
569         unsigned long phy_addr;
570         unsigned long flags;
571
572         spin_lock_irqsave(&iommu->lock, flags);
573         root = &iommu->root_entry[bus];
574         context = get_context_addr_from_root(root);
575         if (!context) {
576                 context = (struct context_entry *)alloc_pgtable_page();
577                 if (!context) {
578                         spin_unlock_irqrestore(&iommu->lock, flags);
579                         return NULL;
580                 }
581                 __iommu_flush_cache(iommu, (void *)context, CONTEXT_SIZE);
582                 phy_addr = virt_to_phys((void *)context);
583                 set_root_value(root, phy_addr);
584                 set_root_present(root);
585                 __iommu_flush_cache(iommu, root, sizeof(*root));
586         }
587         spin_unlock_irqrestore(&iommu->lock, flags);
588         return &context[devfn];
589 }
590
591 static int device_context_mapped(struct intel_iommu *iommu, u8 bus, u8 devfn)
592 {
593         struct root_entry *root;
594         struct context_entry *context;
595         int ret;
596         unsigned long flags;
597
598         spin_lock_irqsave(&iommu->lock, flags);
599         root = &iommu->root_entry[bus];
600         context = get_context_addr_from_root(root);
601         if (!context) {
602                 ret = 0;
603                 goto out;
604         }
605         ret = context_present(&context[devfn]);
606 out:
607         spin_unlock_irqrestore(&iommu->lock, flags);
608         return ret;
609 }
610
611 static void clear_context_table(struct intel_iommu *iommu, u8 bus, u8 devfn)
612 {
613         struct root_entry *root;
614         struct context_entry *context;
615         unsigned long flags;
616
617         spin_lock_irqsave(&iommu->lock, flags);
618         root = &iommu->root_entry[bus];
619         context = get_context_addr_from_root(root);
620         if (context) {
621                 context_clear_entry(&context[devfn]);
622                 __iommu_flush_cache(iommu, &context[devfn], \
623                         sizeof(*context));
624         }
625         spin_unlock_irqrestore(&iommu->lock, flags);
626 }
627
628 static void free_context_table(struct intel_iommu *iommu)
629 {
630         struct root_entry *root;
631         int i;
632         unsigned long flags;
633         struct context_entry *context;
634
635         spin_lock_irqsave(&iommu->lock, flags);
636         if (!iommu->root_entry) {
637                 goto out;
638         }
639         for (i = 0; i < ROOT_ENTRY_NR; i++) {
640                 root = &iommu->root_entry[i];
641                 context = get_context_addr_from_root(root);
642                 if (context)
643                         free_pgtable_page(context);
644         }
645         free_pgtable_page(iommu->root_entry);
646         iommu->root_entry = NULL;
647 out:
648         spin_unlock_irqrestore(&iommu->lock, flags);
649 }
650
651 /* page table handling */
652 #define LEVEL_STRIDE            (9)
653 #define LEVEL_MASK              (((u64)1 << LEVEL_STRIDE) - 1)
654
655 static inline int agaw_to_level(int agaw)
656 {
657         return agaw + 2;
658 }
659
660 static inline int agaw_to_width(int agaw)
661 {
662         return 30 + agaw * LEVEL_STRIDE;
663
664 }
665
666 static inline int width_to_agaw(int width)
667 {
668         return (width - 30) / LEVEL_STRIDE;
669 }
670
671 static inline unsigned int level_to_offset_bits(int level)
672 {
673         return (level - 1) * LEVEL_STRIDE;
674 }
675
676 static inline int pfn_level_offset(unsigned long pfn, int level)
677 {
678         return (pfn >> level_to_offset_bits(level)) & LEVEL_MASK;
679 }
680
681 static inline unsigned long level_mask(int level)
682 {
683         return -1UL << level_to_offset_bits(level);
684 }
685
686 static inline unsigned long level_size(int level)
687 {
688         return 1UL << level_to_offset_bits(level);
689 }
690
691 static inline unsigned long align_to_level(unsigned long pfn, int level)
692 {
693         return (pfn + level_size(level) - 1) & level_mask(level);
694 }
695
696 static struct dma_pte *pfn_to_dma_pte(struct dmar_domain *domain,
697                                       unsigned long pfn)
698 {
699         int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
700         struct dma_pte *parent, *pte = NULL;
701         int level = agaw_to_level(domain->agaw);
702         int offset;
703
704         BUG_ON(!domain->pgd);
705         BUG_ON(addr_width < BITS_PER_LONG && pfn >> addr_width);
706         parent = domain->pgd;
707
708         while (level > 0) {
709                 void *tmp_page;
710
711                 offset = pfn_level_offset(pfn, level);
712                 pte = &parent[offset];
713                 if (level == 1)
714                         break;
715
716                 if (!dma_pte_present(pte)) {
717                         uint64_t pteval;
718
719                         tmp_page = alloc_pgtable_page();
720
721                         if (!tmp_page)
722                                 return NULL;
723
724                         domain_flush_cache(domain, tmp_page, VTD_PAGE_SIZE);
725                         pteval = (virt_to_dma_pfn(tmp_page) << VTD_PAGE_SHIFT) | DMA_PTE_READ | DMA_PTE_WRITE;
726                         if (cmpxchg64(&pte->val, 0ULL, pteval)) {
727                                 /* Someone else set it while we were thinking; use theirs. */
728                                 free_pgtable_page(tmp_page);
729                         } else {
730                                 dma_pte_addr(pte);
731                                 domain_flush_cache(domain, pte, sizeof(*pte));
732                         }
733                 }
734                 parent = phys_to_virt(dma_pte_addr(pte));
735                 level--;
736         }
737
738         return pte;
739 }
740
741 /* return address's pte at specific level */
742 static struct dma_pte *dma_pfn_level_pte(struct dmar_domain *domain,
743                                          unsigned long pfn,
744                                          int level)
745 {
746         struct dma_pte *parent, *pte = NULL;
747         int total = agaw_to_level(domain->agaw);
748         int offset;
749
750         parent = domain->pgd;
751         while (level <= total) {
752                 offset = pfn_level_offset(pfn, total);
753                 pte = &parent[offset];
754                 if (level == total)
755                         return pte;
756
757                 if (!dma_pte_present(pte))
758                         break;
759                 parent = phys_to_virt(dma_pte_addr(pte));
760                 total--;
761         }
762         return NULL;
763 }
764
765 /* clear last level pte, a tlb flush should be followed */
766 static void dma_pte_clear_range(struct dmar_domain *domain,
767                                 unsigned long start_pfn,
768                                 unsigned long last_pfn)
769 {
770         int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
771         struct dma_pte *first_pte, *pte;
772
773         BUG_ON(addr_width < BITS_PER_LONG && start_pfn >> addr_width);
774         BUG_ON(addr_width < BITS_PER_LONG && last_pfn >> addr_width);
775
776         /* we don't need lock here; nobody else touches the iova range */
777         while (start_pfn <= last_pfn) {
778                 first_pte = pte = dma_pfn_level_pte(domain, start_pfn, 1);
779                 if (!pte) {
780                         start_pfn = align_to_level(start_pfn + 1, 2);
781                         continue;
782                 }
783                 while (start_pfn <= last_pfn &&
784                        (unsigned long)pte >> VTD_PAGE_SHIFT ==
785                        (unsigned long)first_pte >> VTD_PAGE_SHIFT) {
786                         dma_clear_pte(pte);
787                         start_pfn++;
788                         pte++;
789                 }
790                 domain_flush_cache(domain, first_pte,
791                                    (void *)pte - (void *)first_pte);
792         }
793 }
794
795 /* free page table pages. last level pte should already be cleared */
796 static void dma_pte_free_pagetable(struct dmar_domain *domain,
797                                    unsigned long start_pfn,
798                                    unsigned long last_pfn)
799 {
800         int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
801         struct dma_pte *first_pte, *pte;
802         int total = agaw_to_level(domain->agaw);
803         int level;
804         unsigned long tmp;
805
806         BUG_ON(addr_width < BITS_PER_LONG && start_pfn >> addr_width);
807         BUG_ON(addr_width < BITS_PER_LONG && last_pfn >> addr_width);
808
809         /* We don't need lock here; nobody else touches the iova range */
810         level = 2;
811         while (level <= total) {
812                 tmp = align_to_level(start_pfn, level);
813
814                 /* If we can't even clear one PTE at this level, we're done */
815                 if (tmp + level_size(level) - 1 > last_pfn)
816                         return;
817
818                 while (tmp + level_size(level) - 1 <= last_pfn) {
819                         first_pte = pte = dma_pfn_level_pte(domain, tmp, level);
820                         if (!pte) {
821                                 tmp = align_to_level(tmp + 1, level + 1);
822                                 continue;
823                         }
824                         while (tmp + level_size(level) - 1 <= last_pfn &&
825                                (unsigned long)pte >> VTD_PAGE_SHIFT ==
826                                (unsigned long)first_pte >> VTD_PAGE_SHIFT) {
827                                 free_pgtable_page(phys_to_virt(dma_pte_addr(pte)));
828                                 dma_clear_pte(pte);
829                                 pte++;
830                                 tmp += level_size(level);
831                         }
832                         domain_flush_cache(domain, first_pte,
833                                            (void *)pte - (void *)first_pte);
834                         
835                 }
836                 level++;
837         }
838         /* free pgd */
839         if (start_pfn == 0 && last_pfn == DOMAIN_MAX_PFN(domain->gaw)) {
840                 free_pgtable_page(domain->pgd);
841                 domain->pgd = NULL;
842         }
843 }
844
845 /* iommu handling */
846 static int iommu_alloc_root_entry(struct intel_iommu *iommu)
847 {
848         struct root_entry *root;
849         unsigned long flags;
850
851         root = (struct root_entry *)alloc_pgtable_page();
852         if (!root)
853                 return -ENOMEM;
854
855         __iommu_flush_cache(iommu, root, ROOT_SIZE);
856
857         spin_lock_irqsave(&iommu->lock, flags);
858         iommu->root_entry = root;
859         spin_unlock_irqrestore(&iommu->lock, flags);
860
861         return 0;
862 }
863
864 static void iommu_set_root_entry(struct intel_iommu *iommu)
865 {
866         void *addr;
867         u32 sts;
868         unsigned long flag;
869
870         addr = iommu->root_entry;
871
872         spin_lock_irqsave(&iommu->register_lock, flag);
873         dmar_writeq(iommu->reg + DMAR_RTADDR_REG, virt_to_phys(addr));
874
875         writel(iommu->gcmd | DMA_GCMD_SRTP, iommu->reg + DMAR_GCMD_REG);
876
877         /* Make sure hardware complete it */
878         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
879                       readl, (sts & DMA_GSTS_RTPS), sts);
880
881         spin_unlock_irqrestore(&iommu->register_lock, flag);
882 }
883
884 static void iommu_flush_write_buffer(struct intel_iommu *iommu)
885 {
886         u32 val;
887         unsigned long flag;
888
889         if (!rwbf_quirk && !cap_rwbf(iommu->cap))
890                 return;
891
892         spin_lock_irqsave(&iommu->register_lock, flag);
893         writel(iommu->gcmd | DMA_GCMD_WBF, iommu->reg + DMAR_GCMD_REG);
894
895         /* Make sure hardware complete it */
896         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
897                       readl, (!(val & DMA_GSTS_WBFS)), val);
898
899         spin_unlock_irqrestore(&iommu->register_lock, flag);
900 }
901
902 /* return value determine if we need a write buffer flush */
903 static void __iommu_flush_context(struct intel_iommu *iommu,
904                                   u16 did, u16 source_id, u8 function_mask,
905                                   u64 type)
906 {
907         u64 val = 0;
908         unsigned long flag;
909
910         switch (type) {
911         case DMA_CCMD_GLOBAL_INVL:
912                 val = DMA_CCMD_GLOBAL_INVL;
913                 break;
914         case DMA_CCMD_DOMAIN_INVL:
915                 val = DMA_CCMD_DOMAIN_INVL|DMA_CCMD_DID(did);
916                 break;
917         case DMA_CCMD_DEVICE_INVL:
918                 val = DMA_CCMD_DEVICE_INVL|DMA_CCMD_DID(did)
919                         | DMA_CCMD_SID(source_id) | DMA_CCMD_FM(function_mask);
920                 break;
921         default:
922                 BUG();
923         }
924         val |= DMA_CCMD_ICC;
925
926         spin_lock_irqsave(&iommu->register_lock, flag);
927         dmar_writeq(iommu->reg + DMAR_CCMD_REG, val);
928
929         /* Make sure hardware complete it */
930         IOMMU_WAIT_OP(iommu, DMAR_CCMD_REG,
931                 dmar_readq, (!(val & DMA_CCMD_ICC)), val);
932
933         spin_unlock_irqrestore(&iommu->register_lock, flag);
934 }
935
936 /* return value determine if we need a write buffer flush */
937 static void __iommu_flush_iotlb(struct intel_iommu *iommu, u16 did,
938                                 u64 addr, unsigned int size_order, u64 type)
939 {
940         int tlb_offset = ecap_iotlb_offset(iommu->ecap);
941         u64 val = 0, val_iva = 0;
942         unsigned long flag;
943
944         switch (type) {
945         case DMA_TLB_GLOBAL_FLUSH:
946                 /* global flush doesn't need set IVA_REG */
947                 val = DMA_TLB_GLOBAL_FLUSH|DMA_TLB_IVT;
948                 break;
949         case DMA_TLB_DSI_FLUSH:
950                 val = DMA_TLB_DSI_FLUSH|DMA_TLB_IVT|DMA_TLB_DID(did);
951                 break;
952         case DMA_TLB_PSI_FLUSH:
953                 val = DMA_TLB_PSI_FLUSH|DMA_TLB_IVT|DMA_TLB_DID(did);
954                 /* Note: always flush non-leaf currently */
955                 val_iva = size_order | addr;
956                 break;
957         default:
958                 BUG();
959         }
960         /* Note: set drain read/write */
961 #if 0
962         /*
963          * This is probably to be super secure.. Looks like we can
964          * ignore it without any impact.
965          */
966         if (cap_read_drain(iommu->cap))
967                 val |= DMA_TLB_READ_DRAIN;
968 #endif
969         if (cap_write_drain(iommu->cap))
970                 val |= DMA_TLB_WRITE_DRAIN;
971
972         spin_lock_irqsave(&iommu->register_lock, flag);
973         /* Note: Only uses first TLB reg currently */
974         if (val_iva)
975                 dmar_writeq(iommu->reg + tlb_offset, val_iva);
976         dmar_writeq(iommu->reg + tlb_offset + 8, val);
977
978         /* Make sure hardware complete it */
979         IOMMU_WAIT_OP(iommu, tlb_offset + 8,
980                 dmar_readq, (!(val & DMA_TLB_IVT)), val);
981
982         spin_unlock_irqrestore(&iommu->register_lock, flag);
983
984         /* check IOTLB invalidation granularity */
985         if (DMA_TLB_IAIG(val) == 0)
986                 printk(KERN_ERR"IOMMU: flush IOTLB failed\n");
987         if (DMA_TLB_IAIG(val) != DMA_TLB_IIRG(type))
988                 pr_debug("IOMMU: tlb flush request %Lx, actual %Lx\n",
989                         (unsigned long long)DMA_TLB_IIRG(type),
990                         (unsigned long long)DMA_TLB_IAIG(val));
991 }
992
993 static struct device_domain_info *iommu_support_dev_iotlb(
994         struct dmar_domain *domain, int segment, u8 bus, u8 devfn)
995 {
996         int found = 0;
997         unsigned long flags;
998         struct device_domain_info *info;
999         struct intel_iommu *iommu = device_to_iommu(segment, bus, devfn);
1000
1001         if (!ecap_dev_iotlb_support(iommu->ecap))
1002                 return NULL;
1003
1004         if (!iommu->qi)
1005                 return NULL;
1006
1007         spin_lock_irqsave(&device_domain_lock, flags);
1008         list_for_each_entry(info, &domain->devices, link)
1009                 if (info->bus == bus && info->devfn == devfn) {
1010                         found = 1;
1011                         break;
1012                 }
1013         spin_unlock_irqrestore(&device_domain_lock, flags);
1014
1015         if (!found || !info->dev)
1016                 return NULL;
1017
1018         if (!pci_find_ext_capability(info->dev, PCI_EXT_CAP_ID_ATS))
1019                 return NULL;
1020
1021         if (!dmar_find_matched_atsr_unit(info->dev))
1022                 return NULL;
1023
1024         info->iommu = iommu;
1025
1026         return info;
1027 }
1028
1029 static void iommu_enable_dev_iotlb(struct device_domain_info *info)
1030 {
1031         if (!info)
1032                 return;
1033
1034         pci_enable_ats(info->dev, VTD_PAGE_SHIFT);
1035 }
1036
1037 static void iommu_disable_dev_iotlb(struct device_domain_info *info)
1038 {
1039         if (!info->dev || !pci_ats_enabled(info->dev))
1040                 return;
1041
1042         pci_disable_ats(info->dev);
1043 }
1044
1045 static void iommu_flush_dev_iotlb(struct dmar_domain *domain,
1046                                   u64 addr, unsigned mask)
1047 {
1048         u16 sid, qdep;
1049         unsigned long flags;
1050         struct device_domain_info *info;
1051
1052         spin_lock_irqsave(&device_domain_lock, flags);
1053         list_for_each_entry(info, &domain->devices, link) {
1054                 if (!info->dev || !pci_ats_enabled(info->dev))
1055                         continue;
1056
1057                 sid = info->bus << 8 | info->devfn;
1058                 qdep = pci_ats_queue_depth(info->dev);
1059                 qi_flush_dev_iotlb(info->iommu, sid, qdep, addr, mask);
1060         }
1061         spin_unlock_irqrestore(&device_domain_lock, flags);
1062 }
1063
1064 static void iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did,
1065                                   unsigned long pfn, unsigned int pages)
1066 {
1067         unsigned int mask = ilog2(__roundup_pow_of_two(pages));
1068         uint64_t addr = (uint64_t)pfn << VTD_PAGE_SHIFT;
1069
1070         BUG_ON(pages == 0);
1071
1072         /*
1073          * Fallback to domain selective flush if no PSI support or the size is
1074          * too big.
1075          * PSI requires page size to be 2 ^ x, and the base address is naturally
1076          * aligned to the size
1077          */
1078         if (!cap_pgsel_inv(iommu->cap) || mask > cap_max_amask_val(iommu->cap))
1079                 iommu->flush.flush_iotlb(iommu, did, 0, 0,
1080                                                 DMA_TLB_DSI_FLUSH);
1081         else
1082                 iommu->flush.flush_iotlb(iommu, did, addr, mask,
1083                                                 DMA_TLB_PSI_FLUSH);
1084
1085         /*
1086          * In caching mode, domain ID 0 is reserved for non-present to present
1087          * mapping flush. Device IOTLB doesn't need to be flushed in this case.
1088          */
1089         if (!cap_caching_mode(iommu->cap) || did)
1090                 iommu_flush_dev_iotlb(iommu->domains[did], addr, mask);
1091 }
1092
1093 static void iommu_disable_protect_mem_regions(struct intel_iommu *iommu)
1094 {
1095         u32 pmen;
1096         unsigned long flags;
1097
1098         spin_lock_irqsave(&iommu->register_lock, flags);
1099         pmen = readl(iommu->reg + DMAR_PMEN_REG);
1100         pmen &= ~DMA_PMEN_EPM;
1101         writel(pmen, iommu->reg + DMAR_PMEN_REG);
1102
1103         /* wait for the protected region status bit to clear */
1104         IOMMU_WAIT_OP(iommu, DMAR_PMEN_REG,
1105                 readl, !(pmen & DMA_PMEN_PRS), pmen);
1106
1107         spin_unlock_irqrestore(&iommu->register_lock, flags);
1108 }
1109
1110 static int iommu_enable_translation(struct intel_iommu *iommu)
1111 {
1112         u32 sts;
1113         unsigned long flags;
1114
1115         spin_lock_irqsave(&iommu->register_lock, flags);
1116         iommu->gcmd |= DMA_GCMD_TE;
1117         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1118
1119         /* Make sure hardware complete it */
1120         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
1121                       readl, (sts & DMA_GSTS_TES), sts);
1122
1123         spin_unlock_irqrestore(&iommu->register_lock, flags);
1124         return 0;
1125 }
1126
1127 static int iommu_disable_translation(struct intel_iommu *iommu)
1128 {
1129         u32 sts;
1130         unsigned long flag;
1131
1132         spin_lock_irqsave(&iommu->register_lock, flag);
1133         iommu->gcmd &= ~DMA_GCMD_TE;
1134         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1135
1136         /* Make sure hardware complete it */
1137         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
1138                       readl, (!(sts & DMA_GSTS_TES)), sts);
1139
1140         spin_unlock_irqrestore(&iommu->register_lock, flag);
1141         return 0;
1142 }
1143
1144
1145 static int iommu_init_domains(struct intel_iommu *iommu)
1146 {
1147         unsigned long ndomains;
1148         unsigned long nlongs;
1149
1150         ndomains = cap_ndoms(iommu->cap);
1151         pr_debug("Number of Domains supportd <%ld>\n", ndomains);
1152         nlongs = BITS_TO_LONGS(ndomains);
1153
1154         /* TBD: there might be 64K domains,
1155          * consider other allocation for future chip
1156          */
1157         iommu->domain_ids = kcalloc(nlongs, sizeof(unsigned long), GFP_KERNEL);
1158         if (!iommu->domain_ids) {
1159                 printk(KERN_ERR "Allocating domain id array failed\n");
1160                 return -ENOMEM;
1161         }
1162         iommu->domains = kcalloc(ndomains, sizeof(struct dmar_domain *),
1163                         GFP_KERNEL);
1164         if (!iommu->domains) {
1165                 printk(KERN_ERR "Allocating domain array failed\n");
1166                 kfree(iommu->domain_ids);
1167                 return -ENOMEM;
1168         }
1169
1170         spin_lock_init(&iommu->lock);
1171
1172         /*
1173          * if Caching mode is set, then invalid translations are tagged
1174          * with domainid 0. Hence we need to pre-allocate it.
1175          */
1176         if (cap_caching_mode(iommu->cap))
1177                 set_bit(0, iommu->domain_ids);
1178         return 0;
1179 }
1180
1181
1182 static void domain_exit(struct dmar_domain *domain);
1183 static void vm_domain_exit(struct dmar_domain *domain);
1184
1185 void free_dmar_iommu(struct intel_iommu *iommu)
1186 {
1187         struct dmar_domain *domain;
1188         int i;
1189         unsigned long flags;
1190
1191         i = find_first_bit(iommu->domain_ids, cap_ndoms(iommu->cap));
1192         for (; i < cap_ndoms(iommu->cap); ) {
1193                 domain = iommu->domains[i];
1194                 clear_bit(i, iommu->domain_ids);
1195
1196                 spin_lock_irqsave(&domain->iommu_lock, flags);
1197                 if (--domain->iommu_count == 0) {
1198                         if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
1199                                 vm_domain_exit(domain);
1200                         else
1201                                 domain_exit(domain);
1202                 }
1203                 spin_unlock_irqrestore(&domain->iommu_lock, flags);
1204
1205                 i = find_next_bit(iommu->domain_ids,
1206                         cap_ndoms(iommu->cap), i+1);
1207         }
1208
1209         if (iommu->gcmd & DMA_GCMD_TE)
1210                 iommu_disable_translation(iommu);
1211
1212         if (iommu->irq) {
1213                 set_irq_data(iommu->irq, NULL);
1214                 /* This will mask the irq */
1215                 free_irq(iommu->irq, iommu);
1216                 destroy_irq(iommu->irq);
1217         }
1218
1219         kfree(iommu->domains);
1220         kfree(iommu->domain_ids);
1221
1222         g_iommus[iommu->seq_id] = NULL;
1223
1224         /* if all iommus are freed, free g_iommus */
1225         for (i = 0; i < g_num_of_iommus; i++) {
1226                 if (g_iommus[i])
1227                         break;
1228         }
1229
1230         if (i == g_num_of_iommus)
1231                 kfree(g_iommus);
1232
1233         /* free context mapping */
1234         free_context_table(iommu);
1235 }
1236
1237 static struct dmar_domain *alloc_domain(void)
1238 {
1239         struct dmar_domain *domain;
1240
1241         domain = alloc_domain_mem();
1242         if (!domain)
1243                 return NULL;
1244
1245         memset(&domain->iommu_bmp, 0, sizeof(unsigned long));
1246         domain->flags = 0;
1247
1248         return domain;
1249 }
1250
1251 static int iommu_attach_domain(struct dmar_domain *domain,
1252                                struct intel_iommu *iommu)
1253 {
1254         int num;
1255         unsigned long ndomains;
1256         unsigned long flags;
1257
1258         ndomains = cap_ndoms(iommu->cap);
1259
1260         spin_lock_irqsave(&iommu->lock, flags);
1261
1262         num = find_first_zero_bit(iommu->domain_ids, ndomains);
1263         if (num >= ndomains) {
1264                 spin_unlock_irqrestore(&iommu->lock, flags);
1265                 printk(KERN_ERR "IOMMU: no free domain ids\n");
1266                 return -ENOMEM;
1267         }
1268
1269         domain->id = num;
1270         set_bit(num, iommu->domain_ids);
1271         set_bit(iommu->seq_id, &domain->iommu_bmp);
1272         iommu->domains[num] = domain;
1273         spin_unlock_irqrestore(&iommu->lock, flags);
1274
1275         return 0;
1276 }
1277
1278 static void iommu_detach_domain(struct dmar_domain *domain,
1279                                 struct intel_iommu *iommu)
1280 {
1281         unsigned long flags;
1282         int num, ndomains;
1283         int found = 0;
1284
1285         spin_lock_irqsave(&iommu->lock, flags);
1286         ndomains = cap_ndoms(iommu->cap);
1287         num = find_first_bit(iommu->domain_ids, ndomains);
1288         for (; num < ndomains; ) {
1289                 if (iommu->domains[num] == domain) {
1290                         found = 1;
1291                         break;
1292                 }
1293                 num = find_next_bit(iommu->domain_ids,
1294                                     cap_ndoms(iommu->cap), num+1);
1295         }
1296
1297         if (found) {
1298                 clear_bit(num, iommu->domain_ids);
1299                 clear_bit(iommu->seq_id, &domain->iommu_bmp);
1300                 iommu->domains[num] = NULL;
1301         }
1302         spin_unlock_irqrestore(&iommu->lock, flags);
1303 }
1304
1305 static struct iova_domain reserved_iova_list;
1306 static struct lock_class_key reserved_alloc_key;
1307 static struct lock_class_key reserved_rbtree_key;
1308
1309 static void dmar_init_reserved_ranges(void)
1310 {
1311         struct pci_dev *pdev = NULL;
1312         struct iova *iova;
1313         int i;
1314
1315         init_iova_domain(&reserved_iova_list, DMA_32BIT_PFN);
1316
1317         lockdep_set_class(&reserved_iova_list.iova_alloc_lock,
1318                 &reserved_alloc_key);
1319         lockdep_set_class(&reserved_iova_list.iova_rbtree_lock,
1320                 &reserved_rbtree_key);
1321
1322         /* IOAPIC ranges shouldn't be accessed by DMA */
1323         iova = reserve_iova(&reserved_iova_list, IOVA_PFN(IOAPIC_RANGE_START),
1324                 IOVA_PFN(IOAPIC_RANGE_END));
1325         if (!iova)
1326                 printk(KERN_ERR "Reserve IOAPIC range failed\n");
1327
1328         /* Reserve all PCI MMIO to avoid peer-to-peer access */
1329         for_each_pci_dev(pdev) {
1330                 struct resource *r;
1331
1332                 for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1333                         r = &pdev->resource[i];
1334                         if (!r->flags || !(r->flags & IORESOURCE_MEM))
1335                                 continue;
1336                         iova = reserve_iova(&reserved_iova_list,
1337                                             IOVA_PFN(r->start),
1338                                             IOVA_PFN(r->end));
1339                         if (!iova)
1340                                 printk(KERN_ERR "Reserve iova failed\n");
1341                 }
1342         }
1343
1344 }
1345
1346 static void domain_reserve_special_ranges(struct dmar_domain *domain)
1347 {
1348         copy_reserved_iova(&reserved_iova_list, &domain->iovad);
1349 }
1350
1351 static inline int guestwidth_to_adjustwidth(int gaw)
1352 {
1353         int agaw;
1354         int r = (gaw - 12) % 9;
1355
1356         if (r == 0)
1357                 agaw = gaw;
1358         else
1359                 agaw = gaw + 9 - r;
1360         if (agaw > 64)
1361                 agaw = 64;
1362         return agaw;
1363 }
1364
1365 static int domain_init(struct dmar_domain *domain, int guest_width)
1366 {
1367         struct intel_iommu *iommu;
1368         int adjust_width, agaw;
1369         unsigned long sagaw;
1370
1371         init_iova_domain(&domain->iovad, DMA_32BIT_PFN);
1372         spin_lock_init(&domain->iommu_lock);
1373
1374         domain_reserve_special_ranges(domain);
1375
1376         /* calculate AGAW */
1377         iommu = domain_get_iommu(domain);
1378         if (guest_width > cap_mgaw(iommu->cap))
1379                 guest_width = cap_mgaw(iommu->cap);
1380         domain->gaw = guest_width;
1381         adjust_width = guestwidth_to_adjustwidth(guest_width);
1382         agaw = width_to_agaw(adjust_width);
1383         sagaw = cap_sagaw(iommu->cap);
1384         if (!test_bit(agaw, &sagaw)) {
1385                 /* hardware doesn't support it, choose a bigger one */
1386                 pr_debug("IOMMU: hardware doesn't support agaw %d\n", agaw);
1387                 agaw = find_next_bit(&sagaw, 5, agaw);
1388                 if (agaw >= 5)
1389                         return -ENODEV;
1390         }
1391         domain->agaw = agaw;
1392         INIT_LIST_HEAD(&domain->devices);
1393
1394         if (ecap_coherent(iommu->ecap))
1395                 domain->iommu_coherency = 1;
1396         else
1397                 domain->iommu_coherency = 0;
1398
1399         if (ecap_sc_support(iommu->ecap))
1400                 domain->iommu_snooping = 1;
1401         else
1402                 domain->iommu_snooping = 0;
1403
1404         domain->iommu_count = 1;
1405
1406         /* always allocate the top pgd */
1407         domain->pgd = (struct dma_pte *)alloc_pgtable_page();
1408         if (!domain->pgd)
1409                 return -ENOMEM;
1410         __iommu_flush_cache(iommu, domain->pgd, PAGE_SIZE);
1411         return 0;
1412 }
1413
1414 static void domain_exit(struct dmar_domain *domain)
1415 {
1416         struct dmar_drhd_unit *drhd;
1417         struct intel_iommu *iommu;
1418
1419         /* Domain 0 is reserved, so dont process it */
1420         if (!domain)
1421                 return;
1422
1423         domain_remove_dev_info(domain);
1424         /* destroy iovas */
1425         put_iova_domain(&domain->iovad);
1426
1427         /* clear ptes */
1428         dma_pte_clear_range(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
1429
1430         /* free page tables */
1431         dma_pte_free_pagetable(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
1432
1433         for_each_active_iommu(iommu, drhd)
1434                 if (test_bit(iommu->seq_id, &domain->iommu_bmp))
1435                         iommu_detach_domain(domain, iommu);
1436
1437         free_domain_mem(domain);
1438 }
1439
1440 static int domain_context_mapping_one(struct dmar_domain *domain, int segment,
1441                                  u8 bus, u8 devfn, int translation)
1442 {
1443         struct context_entry *context;
1444         unsigned long flags;
1445         struct intel_iommu *iommu;
1446         struct dma_pte *pgd;
1447         unsigned long num;
1448         unsigned long ndomains;
1449         int id;
1450         int agaw;
1451         struct device_domain_info *info = NULL;
1452
1453         pr_debug("Set context mapping for %02x:%02x.%d\n",
1454                 bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
1455
1456         BUG_ON(!domain->pgd);
1457         BUG_ON(translation != CONTEXT_TT_PASS_THROUGH &&
1458                translation != CONTEXT_TT_MULTI_LEVEL);
1459
1460         iommu = device_to_iommu(segment, bus, devfn);
1461         if (!iommu)
1462                 return -ENODEV;
1463
1464         context = device_to_context_entry(iommu, bus, devfn);
1465         if (!context)
1466                 return -ENOMEM;
1467         spin_lock_irqsave(&iommu->lock, flags);
1468         if (context_present(context)) {
1469                 spin_unlock_irqrestore(&iommu->lock, flags);
1470                 return 0;
1471         }
1472
1473         id = domain->id;
1474         pgd = domain->pgd;
1475
1476         if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE ||
1477             domain->flags & DOMAIN_FLAG_STATIC_IDENTITY) {
1478                 int found = 0;
1479
1480                 /* find an available domain id for this device in iommu */
1481                 ndomains = cap_ndoms(iommu->cap);
1482                 num = find_first_bit(iommu->domain_ids, ndomains);
1483                 for (; num < ndomains; ) {
1484                         if (iommu->domains[num] == domain) {
1485                                 id = num;
1486                                 found = 1;
1487                                 break;
1488                         }
1489                         num = find_next_bit(iommu->domain_ids,
1490                                             cap_ndoms(iommu->cap), num+1);
1491                 }
1492
1493                 if (found == 0) {
1494                         num = find_first_zero_bit(iommu->domain_ids, ndomains);
1495                         if (num >= ndomains) {
1496                                 spin_unlock_irqrestore(&iommu->lock, flags);
1497                                 printk(KERN_ERR "IOMMU: no free domain ids\n");
1498                                 return -EFAULT;
1499                         }
1500
1501                         set_bit(num, iommu->domain_ids);
1502                         set_bit(iommu->seq_id, &domain->iommu_bmp);
1503                         iommu->domains[num] = domain;
1504                         id = num;
1505                 }
1506
1507                 /* Skip top levels of page tables for
1508                  * iommu which has less agaw than default.
1509                  */
1510                 for (agaw = domain->agaw; agaw != iommu->agaw; agaw--) {
1511                         pgd = phys_to_virt(dma_pte_addr(pgd));
1512                         if (!dma_pte_present(pgd)) {
1513                                 spin_unlock_irqrestore(&iommu->lock, flags);
1514                                 return -ENOMEM;
1515                         }
1516                 }
1517         }
1518
1519         context_set_domain_id(context, id);
1520
1521         if (translation != CONTEXT_TT_PASS_THROUGH) {
1522                 info = iommu_support_dev_iotlb(domain, segment, bus, devfn);
1523                 translation = info ? CONTEXT_TT_DEV_IOTLB :
1524                                      CONTEXT_TT_MULTI_LEVEL;
1525         }
1526         /*
1527          * In pass through mode, AW must be programmed to indicate the largest
1528          * AGAW value supported by hardware. And ASR is ignored by hardware.
1529          */
1530         if (unlikely(translation == CONTEXT_TT_PASS_THROUGH))
1531                 context_set_address_width(context, iommu->msagaw);
1532         else {
1533                 context_set_address_root(context, virt_to_phys(pgd));
1534                 context_set_address_width(context, iommu->agaw);
1535         }
1536
1537         context_set_translation_type(context, translation);
1538         context_set_fault_enable(context);
1539         context_set_present(context);
1540         domain_flush_cache(domain, context, sizeof(*context));
1541
1542         /*
1543          * It's a non-present to present mapping. If hardware doesn't cache
1544          * non-present entry we only need to flush the write-buffer. If the
1545          * _does_ cache non-present entries, then it does so in the special
1546          * domain #0, which we have to flush:
1547          */
1548         if (cap_caching_mode(iommu->cap)) {
1549                 iommu->flush.flush_context(iommu, 0,
1550                                            (((u16)bus) << 8) | devfn,
1551                                            DMA_CCMD_MASK_NOBIT,
1552                                            DMA_CCMD_DEVICE_INVL);
1553                 iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_DSI_FLUSH);
1554         } else {
1555                 iommu_flush_write_buffer(iommu);
1556         }
1557         iommu_enable_dev_iotlb(info);
1558         spin_unlock_irqrestore(&iommu->lock, flags);
1559
1560         spin_lock_irqsave(&domain->iommu_lock, flags);
1561         if (!test_and_set_bit(iommu->seq_id, &domain->iommu_bmp)) {
1562                 domain->iommu_count++;
1563                 domain_update_iommu_cap(domain);
1564         }
1565         spin_unlock_irqrestore(&domain->iommu_lock, flags);
1566         return 0;
1567 }
1568
1569 static int
1570 domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev,
1571                         int translation)
1572 {
1573         int ret;
1574         struct pci_dev *tmp, *parent;
1575
1576         ret = domain_context_mapping_one(domain, pci_domain_nr(pdev->bus),
1577                                          pdev->bus->number, pdev->devfn,
1578                                          translation);
1579         if (ret)
1580                 return ret;
1581
1582         /* dependent device mapping */
1583         tmp = pci_find_upstream_pcie_bridge(pdev);
1584         if (!tmp)
1585                 return 0;
1586         /* Secondary interface's bus number and devfn 0 */
1587         parent = pdev->bus->self;
1588         while (parent != tmp) {
1589                 ret = domain_context_mapping_one(domain,
1590                                                  pci_domain_nr(parent->bus),
1591                                                  parent->bus->number,
1592                                                  parent->devfn, translation);
1593                 if (ret)
1594                         return ret;
1595                 parent = parent->bus->self;
1596         }
1597         if (tmp->is_pcie) /* this is a PCIE-to-PCI bridge */
1598                 return domain_context_mapping_one(domain,
1599                                         pci_domain_nr(tmp->subordinate),
1600                                         tmp->subordinate->number, 0,
1601                                         translation);
1602         else /* this is a legacy PCI bridge */
1603                 return domain_context_mapping_one(domain,
1604                                                   pci_domain_nr(tmp->bus),
1605                                                   tmp->bus->number,
1606                                                   tmp->devfn,
1607                                                   translation);
1608 }
1609
1610 static int domain_context_mapped(struct pci_dev *pdev)
1611 {
1612         int ret;
1613         struct pci_dev *tmp, *parent;
1614         struct intel_iommu *iommu;
1615
1616         iommu = device_to_iommu(pci_domain_nr(pdev->bus), pdev->bus->number,
1617                                 pdev->devfn);
1618         if (!iommu)
1619                 return -ENODEV;
1620
1621         ret = device_context_mapped(iommu, pdev->bus->number, pdev->devfn);
1622         if (!ret)
1623                 return ret;
1624         /* dependent device mapping */
1625         tmp = pci_find_upstream_pcie_bridge(pdev);
1626         if (!tmp)
1627                 return ret;
1628         /* Secondary interface's bus number and devfn 0 */
1629         parent = pdev->bus->self;
1630         while (parent != tmp) {
1631                 ret = device_context_mapped(iommu, parent->bus->number,
1632                                             parent->devfn);
1633                 if (!ret)
1634                         return ret;
1635                 parent = parent->bus->self;
1636         }
1637         if (tmp->is_pcie)
1638                 return device_context_mapped(iommu, tmp->subordinate->number,
1639                                              0);
1640         else
1641                 return device_context_mapped(iommu, tmp->bus->number,
1642                                              tmp->devfn);
1643 }
1644
1645 static int __domain_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
1646                             struct scatterlist *sg, unsigned long phys_pfn,
1647                             unsigned long nr_pages, int prot)
1648 {
1649         struct dma_pte *first_pte = NULL, *pte = NULL;
1650         phys_addr_t uninitialized_var(pteval);
1651         int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
1652         unsigned long sg_res;
1653
1654         BUG_ON(addr_width < BITS_PER_LONG && (iov_pfn + nr_pages - 1) >> addr_width);
1655
1656         if ((prot & (DMA_PTE_READ|DMA_PTE_WRITE)) == 0)
1657                 return -EINVAL;
1658
1659         prot &= DMA_PTE_READ | DMA_PTE_WRITE | DMA_PTE_SNP;
1660
1661         if (sg)
1662                 sg_res = 0;
1663         else {
1664                 sg_res = nr_pages + 1;
1665                 pteval = ((phys_addr_t)phys_pfn << VTD_PAGE_SHIFT) | prot;
1666         }
1667
1668         while (nr_pages--) {
1669                 uint64_t tmp;
1670
1671                 if (!sg_res) {
1672                         sg_res = (sg->offset + sg->length + VTD_PAGE_SIZE - 1) >> VTD_PAGE_SHIFT;
1673                         sg->dma_address = ((dma_addr_t)iov_pfn << VTD_PAGE_SHIFT) + sg->offset;
1674                         sg->dma_length = sg->length;
1675                         pteval = page_to_phys(sg_page(sg)) | prot;
1676                 }
1677                 if (!pte) {
1678                         first_pte = pte = pfn_to_dma_pte(domain, iov_pfn);
1679                         if (!pte)
1680                                 return -ENOMEM;
1681                 }
1682                 /* We don't need lock here, nobody else
1683                  * touches the iova range
1684                  */
1685                 tmp = cmpxchg64(&pte->val, 0ULL, pteval);
1686                 if (tmp) {
1687                         static int dumps = 5;
1688                         printk(KERN_CRIT "ERROR: DMA PTE for vPFN 0x%lx already set (to %llx not %llx)\n",
1689                                iov_pfn, tmp, (unsigned long long)pteval);
1690                         if (dumps) {
1691                                 dumps--;
1692                                 debug_dma_dump_mappings(NULL);
1693                         }
1694                         WARN_ON(1);
1695                 }
1696                 pte++;
1697                 if (!nr_pages ||
1698                     (unsigned long)pte >> VTD_PAGE_SHIFT !=
1699                     (unsigned long)first_pte >> VTD_PAGE_SHIFT) {
1700                         domain_flush_cache(domain, first_pte,
1701                                            (void *)pte - (void *)first_pte);
1702                         pte = NULL;
1703                 }
1704                 iov_pfn++;
1705                 pteval += VTD_PAGE_SIZE;
1706                 sg_res--;
1707                 if (!sg_res)
1708                         sg = sg_next(sg);
1709         }
1710         return 0;
1711 }
1712
1713 static inline int domain_sg_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
1714                                     struct scatterlist *sg, unsigned long nr_pages,
1715                                     int prot)
1716 {
1717         return __domain_mapping(domain, iov_pfn, sg, 0, nr_pages, prot);
1718 }
1719
1720 static inline int domain_pfn_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
1721                                      unsigned long phys_pfn, unsigned long nr_pages,
1722                                      int prot)
1723 {
1724         return __domain_mapping(domain, iov_pfn, NULL, phys_pfn, nr_pages, prot);
1725 }
1726
1727 static void iommu_detach_dev(struct intel_iommu *iommu, u8 bus, u8 devfn)
1728 {
1729         if (!iommu)
1730                 return;
1731
1732         clear_context_table(iommu, bus, devfn);
1733         iommu->flush.flush_context(iommu, 0, 0, 0,
1734                                            DMA_CCMD_GLOBAL_INVL);
1735         iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH);
1736 }
1737
1738 static void domain_remove_dev_info(struct dmar_domain *domain)
1739 {
1740         struct device_domain_info *info;
1741         unsigned long flags;
1742         struct intel_iommu *iommu;
1743
1744         spin_lock_irqsave(&device_domain_lock, flags);
1745         while (!list_empty(&domain->devices)) {
1746                 info = list_entry(domain->devices.next,
1747                         struct device_domain_info, link);
1748                 list_del(&info->link);
1749                 list_del(&info->global);
1750                 if (info->dev)
1751                         info->dev->dev.archdata.iommu = NULL;
1752                 spin_unlock_irqrestore(&device_domain_lock, flags);
1753
1754                 iommu_disable_dev_iotlb(info);
1755                 iommu = device_to_iommu(info->segment, info->bus, info->devfn);
1756                 iommu_detach_dev(iommu, info->bus, info->devfn);
1757                 free_devinfo_mem(info);
1758
1759                 spin_lock_irqsave(&device_domain_lock, flags);
1760         }
1761         spin_unlock_irqrestore(&device_domain_lock, flags);
1762 }
1763
1764 /*
1765  * find_domain
1766  * Note: we use struct pci_dev->dev.archdata.iommu stores the info
1767  */
1768 static struct dmar_domain *
1769 find_domain(struct pci_dev *pdev)
1770 {
1771         struct device_domain_info *info;
1772
1773         /* No lock here, assumes no domain exit in normal case */
1774         info = pdev->dev.archdata.iommu;
1775         if (info)
1776                 return info->domain;
1777         return NULL;
1778 }
1779
1780 /* domain is initialized */
1781 static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw)
1782 {
1783         struct dmar_domain *domain, *found = NULL;
1784         struct intel_iommu *iommu;
1785         struct dmar_drhd_unit *drhd;
1786         struct device_domain_info *info, *tmp;
1787         struct pci_dev *dev_tmp;
1788         unsigned long flags;
1789         int bus = 0, devfn = 0;
1790         int segment;
1791         int ret;
1792
1793         domain = find_domain(pdev);
1794         if (domain)
1795                 return domain;
1796
1797         segment = pci_domain_nr(pdev->bus);
1798
1799         dev_tmp = pci_find_upstream_pcie_bridge(pdev);
1800         if (dev_tmp) {
1801                 if (dev_tmp->is_pcie) {
1802                         bus = dev_tmp->subordinate->number;
1803                         devfn = 0;
1804                 } else {
1805                         bus = dev_tmp->bus->number;
1806                         devfn = dev_tmp->devfn;
1807                 }
1808                 spin_lock_irqsave(&device_domain_lock, flags);
1809                 list_for_each_entry(info, &device_domain_list, global) {
1810                         if (info->segment == segment &&
1811                             info->bus == bus && info->devfn == devfn) {
1812                                 found = info->domain;
1813                                 break;
1814                         }
1815                 }
1816                 spin_unlock_irqrestore(&device_domain_lock, flags);
1817                 /* pcie-pci bridge already has a domain, uses it */
1818                 if (found) {
1819                         domain = found;
1820                         goto found_domain;
1821                 }
1822         }
1823
1824         domain = alloc_domain();
1825         if (!domain)
1826                 goto error;
1827
1828         /* Allocate new domain for the device */
1829         drhd = dmar_find_matched_drhd_unit(pdev);
1830         if (!drhd) {
1831                 printk(KERN_ERR "IOMMU: can't find DMAR for device %s\n",
1832                         pci_name(pdev));
1833                 return NULL;
1834         }
1835         iommu = drhd->iommu;
1836
1837         ret = iommu_attach_domain(domain, iommu);
1838         if (ret) {
1839                 domain_exit(domain);
1840                 goto error;
1841         }
1842
1843         if (domain_init(domain, gaw)) {
1844                 domain_exit(domain);
1845                 goto error;
1846         }
1847
1848         /* register pcie-to-pci device */
1849         if (dev_tmp) {
1850                 info = alloc_devinfo_mem();
1851                 if (!info) {
1852                         domain_exit(domain);
1853                         goto error;
1854                 }
1855                 info->segment = segment;
1856                 info->bus = bus;
1857                 info->devfn = devfn;
1858                 info->dev = NULL;
1859                 info->domain = domain;
1860                 /* This domain is shared by devices under p2p bridge */
1861                 domain->flags |= DOMAIN_FLAG_P2P_MULTIPLE_DEVICES;
1862
1863                 /* pcie-to-pci bridge already has a domain, uses it */
1864                 found = NULL;
1865                 spin_lock_irqsave(&device_domain_lock, flags);
1866                 list_for_each_entry(tmp, &device_domain_list, global) {
1867                         if (tmp->segment == segment &&
1868                             tmp->bus == bus && tmp->devfn == devfn) {
1869                                 found = tmp->domain;
1870                                 break;
1871                         }
1872                 }
1873                 if (found) {
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                 }
1881                 spin_unlock_irqrestore(&device_domain_lock, flags);
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
1917 static int iommu_domain_identity_map(struct dmar_domain *domain,
1918                                      unsigned long long start,
1919                                      unsigned long long end)
1920 {
1921         unsigned long first_vpfn = start >> VTD_PAGE_SHIFT;
1922         unsigned long last_vpfn = end >> VTD_PAGE_SHIFT;
1923
1924         if (!reserve_iova(&domain->iovad, dma_to_mm_pfn(first_vpfn),
1925                           dma_to_mm_pfn(last_vpfn))) {
1926                 printk(KERN_ERR "IOMMU: reserve iova failed\n");
1927                 return -ENOMEM;
1928         }
1929
1930         pr_debug("Mapping reserved region %llx-%llx for domain %d\n",
1931                  start, end, domain->id);
1932         /*
1933          * RMRR range might have overlap with physical memory range,
1934          * clear it first
1935          */
1936         dma_pte_clear_range(domain, first_vpfn, last_vpfn);
1937
1938         return domain_pfn_mapping(domain, first_vpfn, first_vpfn,
1939                                   last_vpfn - first_vpfn + 1,
1940                                   DMA_PTE_READ|DMA_PTE_WRITE);
1941 }
1942
1943 static int iommu_prepare_identity_map(struct pci_dev *pdev,
1944                                       unsigned long long start,
1945                                       unsigned long long end)
1946 {
1947         struct dmar_domain *domain;
1948         int ret;
1949
1950         printk(KERN_INFO
1951                "IOMMU: Setting identity map for device %s [0x%Lx - 0x%Lx]\n",
1952                pci_name(pdev), start, end);
1953
1954         domain = get_domain_for_dev(pdev, DEFAULT_DOMAIN_ADDRESS_WIDTH);
1955         if (!domain)
1956                 return -ENOMEM;
1957
1958         ret = iommu_domain_identity_map(domain, start, end);
1959         if (ret)
1960                 goto error;
1961
1962         /* context entry init */
1963         ret = domain_context_mapping(domain, pdev, CONTEXT_TT_MULTI_LEVEL);
1964         if (ret)
1965                 goto error;
1966
1967         return 0;
1968
1969  error:
1970         domain_exit(domain);
1971         return ret;
1972 }
1973
1974 static inline int iommu_prepare_rmrr_dev(struct dmar_rmrr_unit *rmrr,
1975         struct pci_dev *pdev)
1976 {
1977         if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO)
1978                 return 0;
1979         return iommu_prepare_identity_map(pdev, rmrr->base_address,
1980                 rmrr->end_address + 1);
1981 }
1982
1983 #ifdef CONFIG_DMAR_FLOPPY_WA
1984 static inline void iommu_prepare_isa(void)
1985 {
1986         struct pci_dev *pdev;
1987         int ret;
1988
1989         pdev = pci_get_class(PCI_CLASS_BRIDGE_ISA << 8, NULL);
1990         if (!pdev)
1991                 return;
1992
1993         printk(KERN_INFO "IOMMU: Prepare 0-16MiB unity mapping for LPC\n");
1994         ret = iommu_prepare_identity_map(pdev, 0, 16*1024*1024);
1995
1996         if (ret)
1997                 printk(KERN_ERR "IOMMU: Failed to create 0-16MiB identity map; "
1998                        "floppy might not work\n");
1999
2000 }
2001 #else
2002 static inline void iommu_prepare_isa(void)
2003 {
2004         return;
2005 }
2006 #endif /* !CONFIG_DMAR_FLPY_WA */
2007
2008 /* Initialize each context entry as pass through.*/
2009 static int __init init_context_pass_through(void)
2010 {
2011         struct pci_dev *pdev = NULL;
2012         struct dmar_domain *domain;
2013         int ret;
2014
2015         for_each_pci_dev(pdev) {
2016                 domain = get_domain_for_dev(pdev, DEFAULT_DOMAIN_ADDRESS_WIDTH);
2017                 ret = domain_context_mapping(domain, pdev,
2018                                              CONTEXT_TT_PASS_THROUGH);
2019                 if (ret)
2020                         return ret;
2021         }
2022         return 0;
2023 }
2024
2025 static int md_domain_init(struct dmar_domain *domain, int guest_width);
2026
2027 static int __init si_domain_work_fn(unsigned long start_pfn,
2028                                     unsigned long end_pfn, void *datax)
2029 {
2030         int *ret = datax;
2031
2032         *ret = iommu_domain_identity_map(si_domain,
2033                                          (uint64_t)start_pfn << PAGE_SHIFT,
2034                                          (uint64_t)end_pfn << PAGE_SHIFT);
2035         return *ret;
2036
2037 }
2038
2039 static int si_domain_init(void)
2040 {
2041         struct dmar_drhd_unit *drhd;
2042         struct intel_iommu *iommu;
2043         int nid, ret = 0;
2044
2045         si_domain = alloc_domain();
2046         if (!si_domain)
2047                 return -EFAULT;
2048
2049         pr_debug("Identity mapping domain is domain %d\n", si_domain->id);
2050
2051         for_each_active_iommu(iommu, drhd) {
2052                 ret = iommu_attach_domain(si_domain, iommu);
2053                 if (ret) {
2054                         domain_exit(si_domain);
2055                         return -EFAULT;
2056                 }
2057         }
2058
2059         if (md_domain_init(si_domain, DEFAULT_DOMAIN_ADDRESS_WIDTH)) {
2060                 domain_exit(si_domain);
2061                 return -EFAULT;
2062         }
2063
2064         si_domain->flags = DOMAIN_FLAG_STATIC_IDENTITY;
2065
2066         for_each_online_node(nid) {
2067                 work_with_active_regions(nid, si_domain_work_fn, &ret);
2068                 if (ret)
2069                         return ret;
2070         }
2071
2072         return 0;
2073 }
2074
2075 static void domain_remove_one_dev_info(struct dmar_domain *domain,
2076                                           struct pci_dev *pdev);
2077 static int identity_mapping(struct pci_dev *pdev)
2078 {
2079         struct device_domain_info *info;
2080
2081         if (likely(!iommu_identity_mapping))
2082                 return 0;
2083
2084
2085         list_for_each_entry(info, &si_domain->devices, link)
2086                 if (info->dev == pdev)
2087                         return 1;
2088         return 0;
2089 }
2090
2091 static int domain_add_dev_info(struct dmar_domain *domain,
2092                                   struct pci_dev *pdev)
2093 {
2094         struct device_domain_info *info;
2095         unsigned long flags;
2096
2097         info = alloc_devinfo_mem();
2098         if (!info)
2099                 return -ENOMEM;
2100
2101         info->segment = pci_domain_nr(pdev->bus);
2102         info->bus = pdev->bus->number;
2103         info->devfn = pdev->devfn;
2104         info->dev = pdev;
2105         info->domain = domain;
2106
2107         spin_lock_irqsave(&device_domain_lock, flags);
2108         list_add(&info->link, &domain->devices);
2109         list_add(&info->global, &device_domain_list);
2110         pdev->dev.archdata.iommu = info;
2111         spin_unlock_irqrestore(&device_domain_lock, flags);
2112
2113         return 0;
2114 }
2115
2116 static int iommu_prepare_static_identity_mapping(void)
2117 {
2118         struct pci_dev *pdev = NULL;
2119         int ret;
2120
2121         ret = si_domain_init();
2122         if (ret)
2123                 return -EFAULT;
2124
2125         for_each_pci_dev(pdev) {
2126                 printk(KERN_INFO "IOMMU: identity mapping for device %s\n",
2127                        pci_name(pdev));
2128
2129                 ret = domain_context_mapping(si_domain, pdev,
2130                                              CONTEXT_TT_MULTI_LEVEL);
2131                 if (ret)
2132                         return ret;
2133                 ret = domain_add_dev_info(si_domain, pdev);
2134                 if (ret)
2135                         return ret;
2136         }
2137
2138         return 0;
2139 }
2140
2141 int __init init_dmars(void)
2142 {
2143         struct dmar_drhd_unit *drhd;
2144         struct dmar_rmrr_unit *rmrr;
2145         struct pci_dev *pdev;
2146         struct intel_iommu *iommu;
2147         int i, ret;
2148         int pass_through = 1;
2149
2150         /*
2151          * In case pass through can not be enabled, iommu tries to use identity
2152          * mapping.
2153          */
2154         if (iommu_pass_through)
2155                 iommu_identity_mapping = 1;
2156
2157         /*
2158          * for each drhd
2159          *    allocate root
2160          *    initialize and program root entry to not present
2161          * endfor
2162          */
2163         for_each_drhd_unit(drhd) {
2164                 g_num_of_iommus++;
2165                 /*
2166                  * lock not needed as this is only incremented in the single
2167                  * threaded kernel __init code path all other access are read
2168                  * only
2169                  */
2170         }
2171
2172         g_iommus = kcalloc(g_num_of_iommus, sizeof(struct intel_iommu *),
2173                         GFP_KERNEL);
2174         if (!g_iommus) {
2175                 printk(KERN_ERR "Allocating global iommu array failed\n");
2176                 ret = -ENOMEM;
2177                 goto error;
2178         }
2179
2180         deferred_flush = kzalloc(g_num_of_iommus *
2181                 sizeof(struct deferred_flush_tables), GFP_KERNEL);
2182         if (!deferred_flush) {
2183                 kfree(g_iommus);
2184                 ret = -ENOMEM;
2185                 goto error;
2186         }
2187
2188         for_each_drhd_unit(drhd) {
2189                 if (drhd->ignored)
2190                         continue;
2191
2192                 iommu = drhd->iommu;
2193                 g_iommus[iommu->seq_id] = iommu;
2194
2195                 ret = iommu_init_domains(iommu);
2196                 if (ret)
2197                         goto error;
2198
2199                 /*
2200                  * TBD:
2201                  * we could share the same root & context tables
2202                  * amoung all IOMMU's. Need to Split it later.
2203                  */
2204                 ret = iommu_alloc_root_entry(iommu);
2205                 if (ret) {
2206                         printk(KERN_ERR "IOMMU: allocate root entry failed\n");
2207                         goto error;
2208                 }
2209                 if (!ecap_pass_through(iommu->ecap))
2210                         pass_through = 0;
2211         }
2212         if (iommu_pass_through)
2213                 if (!pass_through) {
2214                         printk(KERN_INFO
2215                                "Pass Through is not supported by hardware.\n");
2216                         iommu_pass_through = 0;
2217                 }
2218
2219         /*
2220          * Start from the sane iommu hardware state.
2221          */
2222         for_each_drhd_unit(drhd) {
2223                 if (drhd->ignored)
2224                         continue;
2225
2226                 iommu = drhd->iommu;
2227
2228                 /*
2229                  * If the queued invalidation is already initialized by us
2230                  * (for example, while enabling interrupt-remapping) then
2231                  * we got the things already rolling from a sane state.
2232                  */
2233                 if (iommu->qi)
2234                         continue;
2235
2236                 /*
2237                  * Clear any previous faults.
2238                  */
2239                 dmar_fault(-1, iommu);
2240                 /*
2241                  * Disable queued invalidation if supported and already enabled
2242                  * before OS handover.
2243                  */
2244                 dmar_disable_qi(iommu);
2245         }
2246
2247         for_each_drhd_unit(drhd) {
2248                 if (drhd->ignored)
2249                         continue;
2250
2251                 iommu = drhd->iommu;
2252
2253                 if (dmar_enable_qi(iommu)) {
2254                         /*
2255                          * Queued Invalidate not enabled, use Register Based
2256                          * Invalidate
2257                          */
2258                         iommu->flush.flush_context = __iommu_flush_context;
2259                         iommu->flush.flush_iotlb = __iommu_flush_iotlb;
2260                         printk(KERN_INFO "IOMMU 0x%Lx: using Register based "
2261                                "invalidation\n",
2262                                (unsigned long long)drhd->reg_base_addr);
2263                 } else {
2264                         iommu->flush.flush_context = qi_flush_context;
2265                         iommu->flush.flush_iotlb = qi_flush_iotlb;
2266                         printk(KERN_INFO "IOMMU 0x%Lx: using Queued "
2267                                "invalidation\n",
2268                                (unsigned long long)drhd->reg_base_addr);
2269                 }
2270         }
2271
2272         /*
2273          * If pass through is set and enabled, context entries of all pci
2274          * devices are intialized by pass through translation type.
2275          */
2276         if (iommu_pass_through) {
2277                 ret = init_context_pass_through();
2278                 if (ret) {
2279                         printk(KERN_ERR "IOMMU: Pass through init failed.\n");
2280                         iommu_pass_through = 0;
2281                 }
2282         }
2283
2284         /*
2285          * If pass through is not set or not enabled, setup context entries for
2286          * identity mappings for rmrr, gfx, and isa and may fall back to static
2287          * identity mapping if iommu_identity_mapping is set.
2288          */
2289         if (!iommu_pass_through) {
2290                 if (iommu_identity_mapping)
2291                         iommu_prepare_static_identity_mapping();
2292                 /*
2293                  * For each rmrr
2294                  *   for each dev attached to rmrr
2295                  *   do
2296                  *     locate drhd for dev, alloc domain for dev
2297                  *     allocate free domain
2298                  *     allocate page table entries for rmrr
2299                  *     if context not allocated for bus
2300                  *           allocate and init context
2301                  *           set present in root table for this bus
2302                  *     init context with domain, translation etc
2303                  *    endfor
2304                  * endfor
2305                  */
2306                 printk(KERN_INFO "IOMMU: Setting RMRR:\n");
2307                 for_each_rmrr_units(rmrr) {
2308                         for (i = 0; i < rmrr->devices_cnt; i++) {
2309                                 pdev = rmrr->devices[i];
2310                                 /*
2311                                  * some BIOS lists non-exist devices in DMAR
2312                                  * table.
2313                                  */
2314                                 if (!pdev)
2315                                         continue;
2316                                 ret = iommu_prepare_rmrr_dev(rmrr, pdev);
2317                                 if (ret)
2318                                         printk(KERN_ERR
2319                                  "IOMMU: mapping reserved region failed\n");
2320                         }
2321                 }
2322
2323                 iommu_prepare_isa();
2324         }
2325
2326         /*
2327          * for each drhd
2328          *   enable fault log
2329          *   global invalidate context cache
2330          *   global invalidate iotlb
2331          *   enable translation
2332          */
2333         for_each_drhd_unit(drhd) {
2334                 if (drhd->ignored)
2335                         continue;
2336                 iommu = drhd->iommu;
2337
2338                 iommu_flush_write_buffer(iommu);
2339
2340                 ret = dmar_set_interrupt(iommu);
2341                 if (ret)
2342                         goto error;
2343
2344                 iommu_set_root_entry(iommu);
2345
2346                 iommu->flush.flush_context(iommu, 0, 0, 0, DMA_CCMD_GLOBAL_INVL);
2347                 iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH);
2348                 iommu_disable_protect_mem_regions(iommu);
2349
2350                 ret = iommu_enable_translation(iommu);
2351                 if (ret)
2352                         goto error;
2353         }
2354
2355         return 0;
2356 error:
2357         for_each_drhd_unit(drhd) {
2358                 if (drhd->ignored)
2359                         continue;
2360                 iommu = drhd->iommu;
2361                 free_iommu(iommu);
2362         }
2363         kfree(g_iommus);
2364         return ret;
2365 }
2366
2367 static inline unsigned long aligned_nrpages(unsigned long host_addr,
2368                                             size_t size)
2369 {
2370         host_addr &= ~PAGE_MASK;
2371         host_addr += size + PAGE_SIZE - 1;
2372
2373         return host_addr >> VTD_PAGE_SHIFT;
2374 }
2375
2376 static struct iova *intel_alloc_iova(struct device *dev,
2377                                      struct dmar_domain *domain,
2378                                      unsigned long nrpages, uint64_t dma_mask)
2379 {
2380         struct pci_dev *pdev = to_pci_dev(dev);
2381         struct iova *iova = NULL;
2382
2383         /* Restrict dma_mask to the width that the iommu can handle */
2384         dma_mask = min_t(uint64_t, DOMAIN_MAX_ADDR(domain->gaw), dma_mask);
2385
2386         if (!dmar_forcedac && dma_mask > DMA_BIT_MASK(32)) {
2387                 /*
2388                  * First try to allocate an io virtual address in
2389                  * DMA_BIT_MASK(32) and if that fails then try allocating
2390                  * from higher range
2391                  */
2392                 iova = alloc_iova(&domain->iovad, nrpages,
2393                                   IOVA_PFN(DMA_BIT_MASK(32)), 1);
2394                 if (iova)
2395                         return iova;
2396         }
2397         iova = alloc_iova(&domain->iovad, nrpages, IOVA_PFN(dma_mask), 1);
2398         if (unlikely(!iova)) {
2399                 printk(KERN_ERR "Allocating %ld-page iova for %s failed",
2400                        nrpages, pci_name(pdev));
2401                 return NULL;
2402         }
2403
2404         return iova;
2405 }
2406
2407 static struct dmar_domain *
2408 get_valid_domain_for_dev(struct pci_dev *pdev)
2409 {
2410         struct dmar_domain *domain;
2411         int ret;
2412
2413         domain = get_domain_for_dev(pdev,
2414                         DEFAULT_DOMAIN_ADDRESS_WIDTH);
2415         if (!domain) {
2416                 printk(KERN_ERR
2417                         "Allocating domain for %s failed", pci_name(pdev));
2418                 return NULL;
2419         }
2420
2421         /* make sure context mapping is ok */
2422         if (unlikely(!domain_context_mapped(pdev))) {
2423                 ret = domain_context_mapping(domain, pdev,
2424                                              CONTEXT_TT_MULTI_LEVEL);
2425                 if (ret) {
2426                         printk(KERN_ERR
2427                                 "Domain context map for %s failed",
2428                                 pci_name(pdev));
2429                         return NULL;
2430                 }
2431         }
2432
2433         return domain;
2434 }
2435
2436 static int iommu_dummy(struct pci_dev *pdev)
2437 {
2438         return pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO;
2439 }
2440
2441 /* Check if the pdev needs to go through non-identity map and unmap process.*/
2442 static int iommu_no_mapping(struct pci_dev *pdev)
2443 {
2444         int found;
2445
2446         if (!iommu_identity_mapping)
2447                 return iommu_dummy(pdev);
2448
2449         found = identity_mapping(pdev);
2450         if (found) {
2451                 if (pdev->dma_mask > DMA_BIT_MASK(32))
2452                         return 1;
2453                 else {
2454                         /*
2455                          * 32 bit DMA is removed from si_domain and fall back
2456                          * to non-identity mapping.
2457                          */
2458                         domain_remove_one_dev_info(si_domain, pdev);
2459                         printk(KERN_INFO "32bit %s uses non-identity mapping\n",
2460                                pci_name(pdev));
2461                         return 0;
2462                 }
2463         } else {
2464                 /*
2465                  * In case of a detached 64 bit DMA device from vm, the device
2466                  * is put into si_domain for identity mapping.
2467                  */
2468                 if (pdev->dma_mask > DMA_BIT_MASK(32)) {
2469                         int ret;
2470                         ret = domain_add_dev_info(si_domain, pdev);
2471                         if (!ret) {
2472                                 printk(KERN_INFO "64bit %s uses identity mapping\n",
2473                                        pci_name(pdev));
2474                                 return 1;
2475                         }
2476                 }
2477         }
2478
2479         return iommu_dummy(pdev);
2480 }
2481
2482 static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
2483                                      size_t size, int dir, u64 dma_mask)
2484 {
2485         struct pci_dev *pdev = to_pci_dev(hwdev);
2486         struct dmar_domain *domain;
2487         phys_addr_t start_paddr;
2488         struct iova *iova;
2489         int prot = 0;
2490         int ret;
2491         struct intel_iommu *iommu;
2492
2493         BUG_ON(dir == DMA_NONE);
2494
2495         if (iommu_no_mapping(pdev))
2496                 return paddr;
2497
2498         domain = get_valid_domain_for_dev(pdev);
2499         if (!domain)
2500                 return 0;
2501
2502         iommu = domain_get_iommu(domain);
2503         size = aligned_nrpages(paddr, size);
2504
2505         iova = intel_alloc_iova(hwdev, domain, size, pdev->dma_mask);
2506         if (!iova)
2507                 goto error;
2508
2509         /*
2510          * Check if DMAR supports zero-length reads on write only
2511          * mappings..
2512          */
2513         if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL || \
2514                         !cap_zlr(iommu->cap))
2515                 prot |= DMA_PTE_READ;
2516         if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
2517                 prot |= DMA_PTE_WRITE;
2518         /*
2519          * paddr - (paddr + size) might be partial page, we should map the whole
2520          * page.  Note: if two part of one page are separately mapped, we
2521          * might have two guest_addr mapping to the same host paddr, but this
2522          * is not a big problem
2523          */
2524         ret = domain_pfn_mapping(domain, mm_to_dma_pfn(iova->pfn_lo),
2525                                  paddr >> VTD_PAGE_SHIFT, size, prot);
2526         if (ret)
2527                 goto error;
2528
2529         /* it's a non-present to present mapping. Only flush if caching mode */
2530         if (cap_caching_mode(iommu->cap))
2531                 iommu_flush_iotlb_psi(iommu, 0, mm_to_dma_pfn(iova->pfn_lo), size);
2532         else
2533                 iommu_flush_write_buffer(iommu);
2534
2535         start_paddr = (phys_addr_t)iova->pfn_lo << PAGE_SHIFT;
2536         start_paddr += paddr & ~PAGE_MASK;
2537         return start_paddr;
2538
2539 error:
2540         if (iova)
2541                 __free_iova(&domain->iovad, iova);
2542         printk(KERN_ERR"Device %s request: %zx@%llx dir %d --- failed\n",
2543                 pci_name(pdev), size, (unsigned long long)paddr, dir);
2544         return 0;
2545 }
2546
2547 static dma_addr_t intel_map_page(struct device *dev, struct page *page,
2548                                  unsigned long offset, size_t size,
2549                                  enum dma_data_direction dir,
2550                                  struct dma_attrs *attrs)
2551 {
2552         return __intel_map_single(dev, page_to_phys(page) + offset, size,
2553                                   dir, to_pci_dev(dev)->dma_mask);
2554 }
2555
2556 static void flush_unmaps(void)
2557 {
2558         int i, j;
2559
2560         timer_on = 0;
2561
2562         /* just flush them all */
2563         for (i = 0; i < g_num_of_iommus; i++) {
2564                 struct intel_iommu *iommu = g_iommus[i];
2565                 if (!iommu)
2566                         continue;
2567
2568                 if (!deferred_flush[i].next)
2569                         continue;
2570
2571                 iommu->flush.flush_iotlb(iommu, 0, 0, 0,
2572                                          DMA_TLB_GLOBAL_FLUSH);
2573                 for (j = 0; j < deferred_flush[i].next; j++) {
2574                         unsigned long mask;
2575                         struct iova *iova = deferred_flush[i].iova[j];
2576
2577                         mask = (iova->pfn_hi - iova->pfn_lo + 1) << PAGE_SHIFT;
2578                         mask = ilog2(mask >> VTD_PAGE_SHIFT);
2579                         iommu_flush_dev_iotlb(deferred_flush[i].domain[j],
2580                                         iova->pfn_lo << PAGE_SHIFT, mask);
2581                         __free_iova(&deferred_flush[i].domain[j]->iovad, iova);
2582                 }
2583                 deferred_flush[i].next = 0;
2584         }
2585
2586         list_size = 0;
2587 }
2588
2589 static void flush_unmaps_timeout(unsigned long data)
2590 {
2591         unsigned long flags;
2592
2593         spin_lock_irqsave(&async_umap_flush_lock, flags);
2594         flush_unmaps();
2595         spin_unlock_irqrestore(&async_umap_flush_lock, flags);
2596 }
2597
2598 static void add_unmap(struct dmar_domain *dom, struct iova *iova)
2599 {
2600         unsigned long flags;
2601         int next, iommu_id;
2602         struct intel_iommu *iommu;
2603
2604         spin_lock_irqsave(&async_umap_flush_lock, flags);
2605         if (list_size == HIGH_WATER_MARK)
2606                 flush_unmaps();
2607
2608         iommu = domain_get_iommu(dom);
2609         iommu_id = iommu->seq_id;
2610
2611         next = deferred_flush[iommu_id].next;
2612         deferred_flush[iommu_id].domain[next] = dom;
2613         deferred_flush[iommu_id].iova[next] = iova;
2614         deferred_flush[iommu_id].next++;
2615
2616         if (!timer_on) {
2617                 mod_timer(&unmap_timer, jiffies + msecs_to_jiffies(10));
2618                 timer_on = 1;
2619         }
2620         list_size++;
2621         spin_unlock_irqrestore(&async_umap_flush_lock, flags);
2622 }
2623
2624 static void intel_unmap_page(struct device *dev, dma_addr_t dev_addr,
2625                              size_t size, enum dma_data_direction dir,
2626                              struct dma_attrs *attrs)
2627 {
2628         struct pci_dev *pdev = to_pci_dev(dev);
2629         struct dmar_domain *domain;
2630         unsigned long start_pfn, last_pfn;
2631         struct iova *iova;
2632         struct intel_iommu *iommu;
2633
2634         if (iommu_no_mapping(pdev))
2635                 return;
2636
2637         domain = find_domain(pdev);
2638         BUG_ON(!domain);
2639
2640         iommu = domain_get_iommu(domain);
2641
2642         iova = find_iova(&domain->iovad, IOVA_PFN(dev_addr));
2643         if (WARN_ONCE(!iova, "Driver unmaps unmatched page at PFN %llx\n",
2644                       (unsigned long long)dev_addr))
2645                 return;
2646
2647         start_pfn = mm_to_dma_pfn(iova->pfn_lo);
2648         last_pfn = mm_to_dma_pfn(iova->pfn_hi + 1) - 1;
2649
2650         pr_debug("Device %s unmapping: pfn %lx-%lx\n",
2651                  pci_name(pdev), start_pfn, last_pfn);
2652
2653         /*  clear the whole page */
2654         dma_pte_clear_range(domain, start_pfn, last_pfn);
2655
2656         /* free page tables */
2657         dma_pte_free_pagetable(domain, start_pfn, last_pfn);
2658
2659         if (intel_iommu_strict) {
2660                 iommu_flush_iotlb_psi(iommu, domain->id, start_pfn,
2661                                       last_pfn - start_pfn + 1);
2662                 /* free iova */
2663                 __free_iova(&domain->iovad, iova);
2664         } else {
2665                 add_unmap(domain, iova);
2666                 /*
2667                  * queue up the release of the unmap to save the 1/6th of the
2668                  * cpu used up by the iotlb flush operation...
2669                  */
2670         }
2671 }
2672
2673 static void intel_unmap_single(struct device *dev, dma_addr_t dev_addr, size_t size,
2674                                int dir)
2675 {
2676         intel_unmap_page(dev, dev_addr, size, dir, NULL);
2677 }
2678
2679 static void *intel_alloc_coherent(struct device *hwdev, size_t size,
2680                                   dma_addr_t *dma_handle, gfp_t flags)
2681 {
2682         void *vaddr;
2683         int order;
2684
2685         size = PAGE_ALIGN(size);
2686         order = get_order(size);
2687         flags &= ~(GFP_DMA | GFP_DMA32);
2688
2689         vaddr = (void *)__get_free_pages(flags, order);
2690         if (!vaddr)
2691                 return NULL;
2692         memset(vaddr, 0, size);
2693
2694         *dma_handle = __intel_map_single(hwdev, virt_to_bus(vaddr), size,
2695                                          DMA_BIDIRECTIONAL,
2696                                          hwdev->coherent_dma_mask);
2697         if (*dma_handle)
2698                 return vaddr;
2699         free_pages((unsigned long)vaddr, order);
2700         return NULL;
2701 }
2702
2703 static void intel_free_coherent(struct device *hwdev, size_t size, void *vaddr,
2704                                 dma_addr_t dma_handle)
2705 {
2706         int order;
2707
2708         size = PAGE_ALIGN(size);
2709         order = get_order(size);
2710
2711         intel_unmap_single(hwdev, dma_handle, size, DMA_BIDIRECTIONAL);
2712         free_pages((unsigned long)vaddr, order);
2713 }
2714
2715 static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
2716                            int nelems, enum dma_data_direction dir,
2717                            struct dma_attrs *attrs)
2718 {
2719         struct pci_dev *pdev = to_pci_dev(hwdev);
2720         struct dmar_domain *domain;
2721         unsigned long start_pfn, last_pfn;
2722         struct iova *iova;
2723         struct intel_iommu *iommu;
2724
2725         if (iommu_no_mapping(pdev))
2726                 return;
2727
2728         domain = find_domain(pdev);
2729         BUG_ON(!domain);
2730
2731         iommu = domain_get_iommu(domain);
2732
2733         iova = find_iova(&domain->iovad, IOVA_PFN(sglist[0].dma_address));
2734         if (WARN_ONCE(!iova, "Driver unmaps unmatched sglist at PFN %llx\n",
2735                       (unsigned long long)sglist[0].dma_address))
2736                 return;
2737
2738         start_pfn = mm_to_dma_pfn(iova->pfn_lo);
2739         last_pfn = mm_to_dma_pfn(iova->pfn_hi + 1) - 1;
2740
2741         /*  clear the whole page */
2742         dma_pte_clear_range(domain, start_pfn, last_pfn);
2743
2744         /* free page tables */
2745         dma_pte_free_pagetable(domain, start_pfn, last_pfn);
2746
2747         iommu_flush_iotlb_psi(iommu, domain->id, start_pfn,
2748                               (last_pfn - start_pfn + 1));
2749
2750         /* free iova */
2751         __free_iova(&domain->iovad, iova);
2752 }
2753
2754 static int intel_nontranslate_map_sg(struct device *hddev,
2755         struct scatterlist *sglist, int nelems, int dir)
2756 {
2757         int i;
2758         struct scatterlist *sg;
2759
2760         for_each_sg(sglist, sg, nelems, i) {
2761                 BUG_ON(!sg_page(sg));
2762                 sg->dma_address = page_to_phys(sg_page(sg)) + sg->offset;
2763                 sg->dma_length = sg->length;
2764         }
2765         return nelems;
2766 }
2767
2768 static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
2769                         enum dma_data_direction dir, struct dma_attrs *attrs)
2770 {
2771         int i;
2772         struct pci_dev *pdev = to_pci_dev(hwdev);
2773         struct dmar_domain *domain;
2774         size_t size = 0;
2775         int prot = 0;
2776         size_t offset_pfn = 0;
2777         struct iova *iova = NULL;
2778         int ret;
2779         struct scatterlist *sg;
2780         unsigned long start_vpfn;
2781         struct intel_iommu *iommu;
2782
2783         BUG_ON(dir == DMA_NONE);
2784         if (iommu_no_mapping(pdev))
2785                 return intel_nontranslate_map_sg(hwdev, sglist, nelems, dir);
2786
2787         domain = get_valid_domain_for_dev(pdev);
2788         if (!domain)
2789                 return 0;
2790
2791         iommu = domain_get_iommu(domain);
2792
2793         for_each_sg(sglist, sg, nelems, i)
2794                 size += aligned_nrpages(sg->offset, sg->length);
2795
2796         iova = intel_alloc_iova(hwdev, domain, size, pdev->dma_mask);
2797         if (!iova) {
2798                 sglist->dma_length = 0;
2799                 return 0;
2800         }
2801
2802         /*
2803          * Check if DMAR supports zero-length reads on write only
2804          * mappings..
2805          */
2806         if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL || \
2807                         !cap_zlr(iommu->cap))
2808                 prot |= DMA_PTE_READ;
2809         if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
2810                 prot |= DMA_PTE_WRITE;
2811
2812         start_vpfn = mm_to_dma_pfn(iova->pfn_lo);
2813
2814         ret = domain_sg_mapping(domain, start_vpfn, sglist, mm_to_dma_pfn(size), prot);
2815         if (unlikely(ret)) {
2816                 /*  clear the page */
2817                 dma_pte_clear_range(domain, start_vpfn,
2818                                     start_vpfn + size - 1);
2819                 /* free page tables */
2820                 dma_pte_free_pagetable(domain, start_vpfn,
2821                                        start_vpfn + size - 1);
2822                 /* free iova */
2823                 __free_iova(&domain->iovad, iova);
2824                 return 0;
2825         }
2826
2827         /* it's a non-present to present mapping. Only flush if caching mode */
2828         if (cap_caching_mode(iommu->cap))
2829                 iommu_flush_iotlb_psi(iommu, 0, start_vpfn, offset_pfn);
2830         else
2831                 iommu_flush_write_buffer(iommu);
2832
2833         return nelems;
2834 }
2835
2836 static int intel_mapping_error(struct device *dev, dma_addr_t dma_addr)
2837 {
2838         return !dma_addr;
2839 }
2840
2841 struct dma_map_ops intel_dma_ops = {
2842         .alloc_coherent = intel_alloc_coherent,
2843         .free_coherent = intel_free_coherent,
2844         .map_sg = intel_map_sg,
2845         .unmap_sg = intel_unmap_sg,
2846         .map_page = intel_map_page,
2847         .unmap_page = intel_unmap_page,
2848         .mapping_error = intel_mapping_error,
2849 };
2850
2851 static inline int iommu_domain_cache_init(void)
2852 {
2853         int ret = 0;
2854
2855         iommu_domain_cache = kmem_cache_create("iommu_domain",
2856                                          sizeof(struct dmar_domain),
2857                                          0,
2858                                          SLAB_HWCACHE_ALIGN,
2859
2860                                          NULL);
2861         if (!iommu_domain_cache) {
2862                 printk(KERN_ERR "Couldn't create iommu_domain cache\n");
2863                 ret = -ENOMEM;
2864         }
2865
2866         return ret;
2867 }
2868
2869 static inline int iommu_devinfo_cache_init(void)
2870 {
2871         int ret = 0;
2872
2873         iommu_devinfo_cache = kmem_cache_create("iommu_devinfo",
2874                                          sizeof(struct device_domain_info),
2875                                          0,
2876                                          SLAB_HWCACHE_ALIGN,
2877                                          NULL);
2878         if (!iommu_devinfo_cache) {
2879                 printk(KERN_ERR "Couldn't create devinfo cache\n");
2880                 ret = -ENOMEM;
2881         }
2882
2883         return ret;
2884 }
2885
2886 static inline int iommu_iova_cache_init(void)
2887 {
2888         int ret = 0;
2889
2890         iommu_iova_cache = kmem_cache_create("iommu_iova",
2891                                          sizeof(struct iova),
2892                                          0,
2893                                          SLAB_HWCACHE_ALIGN,
2894                                          NULL);
2895         if (!iommu_iova_cache) {
2896                 printk(KERN_ERR "Couldn't create iova cache\n");
2897                 ret = -ENOMEM;
2898         }
2899
2900         return ret;
2901 }
2902
2903 static int __init iommu_init_mempool(void)
2904 {
2905         int ret;
2906         ret = iommu_iova_cache_init();
2907         if (ret)
2908                 return ret;
2909
2910         ret = iommu_domain_cache_init();
2911         if (ret)
2912                 goto domain_error;
2913
2914         ret = iommu_devinfo_cache_init();
2915         if (!ret)
2916                 return ret;
2917
2918         kmem_cache_destroy(iommu_domain_cache);
2919 domain_error:
2920         kmem_cache_destroy(iommu_iova_cache);
2921
2922         return -ENOMEM;
2923 }
2924
2925 static void __init iommu_exit_mempool(void)
2926 {
2927         kmem_cache_destroy(iommu_devinfo_cache);
2928         kmem_cache_destroy(iommu_domain_cache);
2929         kmem_cache_destroy(iommu_iova_cache);
2930
2931 }
2932
2933 static void __init init_no_remapping_devices(void)
2934 {
2935         struct dmar_drhd_unit *drhd;
2936
2937         for_each_drhd_unit(drhd) {
2938                 if (!drhd->include_all) {
2939                         int i;
2940                         for (i = 0; i < drhd->devices_cnt; i++)
2941                                 if (drhd->devices[i] != NULL)
2942                                         break;
2943                         /* ignore DMAR unit if no pci devices exist */
2944                         if (i == drhd->devices_cnt)
2945                                 drhd->ignored = 1;
2946                 }
2947         }
2948
2949         if (dmar_map_gfx)
2950                 return;
2951
2952         for_each_drhd_unit(drhd) {
2953                 int i;
2954                 if (drhd->ignored || drhd->include_all)
2955                         continue;
2956
2957                 for (i = 0; i < drhd->devices_cnt; i++)
2958                         if (drhd->devices[i] &&
2959                                 !IS_GFX_DEVICE(drhd->devices[i]))
2960                                 break;
2961
2962                 if (i < drhd->devices_cnt)
2963                         continue;
2964
2965                 /* bypass IOMMU if it is just for gfx devices */
2966                 drhd->ignored = 1;
2967                 for (i = 0; i < drhd->devices_cnt; i++) {
2968                         if (!drhd->devices[i])
2969                                 continue;
2970                         drhd->devices[i]->dev.archdata.iommu = DUMMY_DEVICE_DOMAIN_INFO;
2971                 }
2972         }
2973 }
2974
2975 #ifdef CONFIG_SUSPEND
2976 static int init_iommu_hw(void)
2977 {
2978         struct dmar_drhd_unit *drhd;
2979         struct intel_iommu *iommu = NULL;
2980
2981         for_each_active_iommu(iommu, drhd)
2982                 if (iommu->qi)
2983                         dmar_reenable_qi(iommu);
2984
2985         for_each_active_iommu(iommu, drhd) {
2986                 iommu_flush_write_buffer(iommu);
2987
2988                 iommu_set_root_entry(iommu);
2989
2990                 iommu->flush.flush_context(iommu, 0, 0, 0,
2991                                            DMA_CCMD_GLOBAL_INVL);
2992                 iommu->flush.flush_iotlb(iommu, 0, 0, 0,
2993                                          DMA_TLB_GLOBAL_FLUSH);
2994                 iommu_disable_protect_mem_regions(iommu);
2995                 iommu_enable_translation(iommu);
2996         }
2997
2998         return 0;
2999 }
3000
3001 static void iommu_flush_all(void)
3002 {
3003         struct dmar_drhd_unit *drhd;
3004         struct intel_iommu *iommu;
3005
3006         for_each_active_iommu(iommu, drhd) {
3007                 iommu->flush.flush_context(iommu, 0, 0, 0,
3008                                            DMA_CCMD_GLOBAL_INVL);
3009                 iommu->flush.flush_iotlb(iommu, 0, 0, 0,
3010                                          DMA_TLB_GLOBAL_FLUSH);
3011         }
3012 }
3013
3014 static int iommu_suspend(struct sys_device *dev, pm_message_t state)
3015 {
3016         struct dmar_drhd_unit *drhd;
3017         struct intel_iommu *iommu = NULL;
3018         unsigned long flag;
3019
3020         for_each_active_iommu(iommu, drhd) {
3021                 iommu->iommu_state = kzalloc(sizeof(u32) * MAX_SR_DMAR_REGS,
3022                                                  GFP_ATOMIC);
3023                 if (!iommu->iommu_state)
3024                         goto nomem;
3025         }
3026
3027         iommu_flush_all();
3028
3029         for_each_active_iommu(iommu, drhd) {
3030                 iommu_disable_translation(iommu);
3031
3032                 spin_lock_irqsave(&iommu->register_lock, flag);
3033
3034                 iommu->iommu_state[SR_DMAR_FECTL_REG] =
3035                         readl(iommu->reg + DMAR_FECTL_REG);
3036                 iommu->iommu_state[SR_DMAR_FEDATA_REG] =
3037                         readl(iommu->reg + DMAR_FEDATA_REG);
3038                 iommu->iommu_state[SR_DMAR_FEADDR_REG] =
3039                         readl(iommu->reg + DMAR_FEADDR_REG);
3040                 iommu->iommu_state[SR_DMAR_FEUADDR_REG] =
3041                         readl(iommu->reg + DMAR_FEUADDR_REG);
3042
3043                 spin_unlock_irqrestore(&iommu->register_lock, flag);
3044         }
3045         return 0;
3046
3047 nomem:
3048         for_each_active_iommu(iommu, drhd)
3049                 kfree(iommu->iommu_state);
3050
3051         return -ENOMEM;
3052 }
3053
3054 static int iommu_resume(struct sys_device *dev)
3055 {
3056         struct dmar_drhd_unit *drhd;
3057         struct intel_iommu *iommu = NULL;
3058         unsigned long flag;
3059
3060         if (init_iommu_hw()) {
3061                 WARN(1, "IOMMU setup failed, DMAR can not resume!\n");
3062                 return -EIO;
3063         }
3064
3065         for_each_active_iommu(iommu, drhd) {
3066
3067                 spin_lock_irqsave(&iommu->register_lock, flag);
3068
3069                 writel(iommu->iommu_state[SR_DMAR_FECTL_REG],
3070                         iommu->reg + DMAR_FECTL_REG);
3071                 writel(iommu->iommu_state[SR_DMAR_FEDATA_REG],
3072                         iommu->reg + DMAR_FEDATA_REG);
3073                 writel(iommu->iommu_state[SR_DMAR_FEADDR_REG],
3074                         iommu->reg + DMAR_FEADDR_REG);
3075                 writel(iommu->iommu_state[SR_DMAR_FEUADDR_REG],
3076                         iommu->reg + DMAR_FEUADDR_REG);
3077
3078                 spin_unlock_irqrestore(&iommu->register_lock, flag);
3079         }
3080
3081         for_each_active_iommu(iommu, drhd)
3082                 kfree(iommu->iommu_state);
3083
3084         return 0;
3085 }
3086
3087 static struct sysdev_class iommu_sysclass = {
3088         .name           = "iommu",
3089         .resume         = iommu_resume,
3090         .suspend        = iommu_suspend,
3091 };
3092
3093 static struct sys_device device_iommu = {
3094         .cls    = &iommu_sysclass,
3095 };
3096
3097 static int __init init_iommu_sysfs(void)
3098 {
3099         int error;
3100
3101         error = sysdev_class_register(&iommu_sysclass);
3102         if (error)
3103                 return error;
3104
3105         error = sysdev_register(&device_iommu);
3106         if (error)
3107                 sysdev_class_unregister(&iommu_sysclass);
3108
3109         return error;
3110 }
3111
3112 #else
3113 static int __init init_iommu_sysfs(void)
3114 {
3115         return 0;
3116 }
3117 #endif  /* CONFIG_PM */
3118
3119 int __init intel_iommu_init(void)
3120 {
3121         int ret = 0;
3122
3123         if (dmar_table_init())
3124                 return  -ENODEV;
3125
3126         if (dmar_dev_scope_init())
3127                 return  -ENODEV;
3128
3129         /*
3130          * Check the need for DMA-remapping initialization now.
3131          * Above initialization will also be used by Interrupt-remapping.
3132          */
3133         if (no_iommu || (swiotlb && !iommu_pass_through) || dmar_disabled)
3134                 return -ENODEV;
3135
3136         iommu_init_mempool();
3137         dmar_init_reserved_ranges();
3138
3139         init_no_remapping_devices();
3140
3141         ret = init_dmars();
3142         if (ret) {
3143                 printk(KERN_ERR "IOMMU: dmar init failed\n");
3144                 put_iova_domain(&reserved_iova_list);
3145                 iommu_exit_mempool();
3146                 return ret;
3147         }
3148         printk(KERN_INFO
3149         "PCI-DMA: Intel(R) Virtualization Technology for Directed I/O\n");
3150
3151         init_timer(&unmap_timer);
3152         force_iommu = 1;
3153
3154         if (!iommu_pass_through) {
3155                 printk(KERN_INFO
3156                        "Multi-level page-table translation for DMAR.\n");
3157                 dma_ops = &intel_dma_ops;
3158         } else
3159                 printk(KERN_INFO
3160                        "DMAR: Pass through translation for DMAR.\n");
3161
3162         init_iommu_sysfs();
3163
3164         register_iommu(&intel_iommu_ops);
3165
3166         return 0;
3167 }
3168
3169 static void iommu_detach_dependent_devices(struct intel_iommu *iommu,
3170                                            struct pci_dev *pdev)
3171 {
3172         struct pci_dev *tmp, *parent;
3173
3174         if (!iommu || !pdev)
3175                 return;
3176
3177         /* dependent device detach */
3178         tmp = pci_find_upstream_pcie_bridge(pdev);
3179         /* Secondary interface's bus number and devfn 0 */
3180         if (tmp) {
3181                 parent = pdev->bus->self;
3182                 while (parent != tmp) {
3183                         iommu_detach_dev(iommu, parent->bus->number,
3184                                          parent->devfn);
3185                         parent = parent->bus->self;
3186                 }
3187                 if (tmp->is_pcie) /* this is a PCIE-to-PCI bridge */
3188                         iommu_detach_dev(iommu,
3189                                 tmp->subordinate->number, 0);
3190                 else /* this is a legacy PCI bridge */
3191                         iommu_detach_dev(iommu, tmp->bus->number,
3192                                          tmp->devfn);
3193         }
3194 }
3195
3196 static void domain_remove_one_dev_info(struct dmar_domain *domain,
3197                                           struct pci_dev *pdev)
3198 {
3199         struct device_domain_info *info;
3200         struct intel_iommu *iommu;
3201         unsigned long flags;
3202         int found = 0;
3203         struct list_head *entry, *tmp;
3204
3205         iommu = device_to_iommu(pci_domain_nr(pdev->bus), pdev->bus->number,
3206                                 pdev->devfn);
3207         if (!iommu)
3208                 return;
3209
3210         spin_lock_irqsave(&device_domain_lock, flags);
3211         list_for_each_safe(entry, tmp, &domain->devices) {
3212                 info = list_entry(entry, struct device_domain_info, link);
3213                 /* No need to compare PCI domain; it has to be the same */
3214                 if (info->bus == pdev->bus->number &&
3215                     info->devfn == pdev->devfn) {
3216                         list_del(&info->link);
3217                         list_del(&info->global);
3218                         if (info->dev)
3219                                 info->dev->dev.archdata.iommu = NULL;
3220                         spin_unlock_irqrestore(&device_domain_lock, flags);
3221
3222                         iommu_disable_dev_iotlb(info);
3223                         iommu_detach_dev(iommu, info->bus, info->devfn);
3224                         iommu_detach_dependent_devices(iommu, pdev);
3225                         free_devinfo_mem(info);
3226
3227                         spin_lock_irqsave(&device_domain_lock, flags);
3228
3229                         if (found)
3230                                 break;
3231                         else
3232                                 continue;
3233                 }
3234
3235                 /* if there is no other devices under the same iommu
3236                  * owned by this domain, clear this iommu in iommu_bmp
3237                  * update iommu count and coherency
3238                  */
3239                 if (iommu == device_to_iommu(info->segment, info->bus,
3240                                             info->devfn))
3241                         found = 1;
3242         }
3243
3244         if (found == 0) {
3245                 unsigned long tmp_flags;
3246                 spin_lock_irqsave(&domain->iommu_lock, tmp_flags);
3247                 clear_bit(iommu->seq_id, &domain->iommu_bmp);
3248                 domain->iommu_count--;
3249                 domain_update_iommu_cap(domain);
3250                 spin_unlock_irqrestore(&domain->iommu_lock, tmp_flags);
3251         }
3252
3253         spin_unlock_irqrestore(&device_domain_lock, flags);
3254 }
3255
3256 static void vm_domain_remove_all_dev_info(struct dmar_domain *domain)
3257 {
3258         struct device_domain_info *info;
3259         struct intel_iommu *iommu;
3260         unsigned long flags1, flags2;
3261
3262         spin_lock_irqsave(&device_domain_lock, flags1);
3263         while (!list_empty(&domain->devices)) {
3264                 info = list_entry(domain->devices.next,
3265                         struct device_domain_info, link);
3266                 list_del(&info->link);
3267                 list_del(&info->global);
3268                 if (info->dev)
3269                         info->dev->dev.archdata.iommu = NULL;
3270
3271                 spin_unlock_irqrestore(&device_domain_lock, flags1);
3272
3273                 iommu_disable_dev_iotlb(info);
3274                 iommu = device_to_iommu(info->segment, info->bus, info->devfn);
3275                 iommu_detach_dev(iommu, info->bus, info->devfn);
3276                 iommu_detach_dependent_devices(iommu, info->dev);
3277
3278                 /* clear this iommu in iommu_bmp, update iommu count
3279                  * and capabilities
3280                  */
3281                 spin_lock_irqsave(&domain->iommu_lock, flags2);
3282                 if (test_and_clear_bit(iommu->seq_id,
3283                                        &domain->iommu_bmp)) {
3284                         domain->iommu_count--;
3285                         domain_update_iommu_cap(domain);
3286                 }
3287                 spin_unlock_irqrestore(&domain->iommu_lock, flags2);
3288
3289                 free_devinfo_mem(info);
3290                 spin_lock_irqsave(&device_domain_lock, flags1);
3291         }
3292         spin_unlock_irqrestore(&device_domain_lock, flags1);
3293 }
3294
3295 /* domain id for virtual machine, it won't be set in context */
3296 static unsigned long vm_domid;
3297
3298 static int vm_domain_min_agaw(struct dmar_domain *domain)
3299 {
3300         int i;
3301         int min_agaw = domain->agaw;
3302
3303         i = find_first_bit(&domain->iommu_bmp, g_num_of_iommus);
3304         for (; i < g_num_of_iommus; ) {
3305                 if (min_agaw > g_iommus[i]->agaw)
3306                         min_agaw = g_iommus[i]->agaw;
3307
3308                 i = find_next_bit(&domain->iommu_bmp, g_num_of_iommus, i+1);
3309         }
3310
3311         return min_agaw;
3312 }
3313
3314 static struct dmar_domain *iommu_alloc_vm_domain(void)
3315 {
3316         struct dmar_domain *domain;
3317
3318         domain = alloc_domain_mem();
3319         if (!domain)
3320                 return NULL;
3321
3322         domain->id = vm_domid++;
3323         memset(&domain->iommu_bmp, 0, sizeof(unsigned long));
3324         domain->flags = DOMAIN_FLAG_VIRTUAL_MACHINE;
3325
3326         return domain;
3327 }
3328
3329 static int md_domain_init(struct dmar_domain *domain, int guest_width)
3330 {
3331         int adjust_width;
3332
3333         init_iova_domain(&domain->iovad, DMA_32BIT_PFN);
3334         spin_lock_init(&domain->iommu_lock);
3335
3336         domain_reserve_special_ranges(domain);
3337
3338         /* calculate AGAW */
3339         domain->gaw = guest_width;
3340         adjust_width = guestwidth_to_adjustwidth(guest_width);
3341         domain->agaw = width_to_agaw(adjust_width);
3342
3343         INIT_LIST_HEAD(&domain->devices);
3344
3345         domain->iommu_count = 0;
3346         domain->iommu_coherency = 0;
3347         domain->max_addr = 0;
3348
3349         /* always allocate the top pgd */
3350         domain->pgd = (struct dma_pte *)alloc_pgtable_page();
3351         if (!domain->pgd)
3352                 return -ENOMEM;
3353         domain_flush_cache(domain, domain->pgd, PAGE_SIZE);
3354         return 0;
3355 }
3356
3357 static void iommu_free_vm_domain(struct dmar_domain *domain)
3358 {
3359         unsigned long flags;
3360         struct dmar_drhd_unit *drhd;
3361         struct intel_iommu *iommu;
3362         unsigned long i;
3363         unsigned long ndomains;
3364
3365         for_each_drhd_unit(drhd) {
3366                 if (drhd->ignored)
3367                         continue;
3368                 iommu = drhd->iommu;
3369
3370                 ndomains = cap_ndoms(iommu->cap);
3371                 i = find_first_bit(iommu->domain_ids, ndomains);
3372                 for (; i < ndomains; ) {
3373                         if (iommu->domains[i] == domain) {
3374                                 spin_lock_irqsave(&iommu->lock, flags);
3375                                 clear_bit(i, iommu->domain_ids);
3376                                 iommu->domains[i] = NULL;
3377                                 spin_unlock_irqrestore(&iommu->lock, flags);
3378                                 break;
3379                         }
3380                         i = find_next_bit(iommu->domain_ids, ndomains, i+1);
3381                 }
3382         }
3383 }
3384
3385 static void vm_domain_exit(struct dmar_domain *domain)
3386 {
3387         /* Domain 0 is reserved, so dont process it */
3388         if (!domain)
3389                 return;
3390
3391         vm_domain_remove_all_dev_info(domain);
3392         /* destroy iovas */
3393         put_iova_domain(&domain->iovad);
3394
3395         /* clear ptes */
3396         dma_pte_clear_range(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
3397
3398         /* free page tables */
3399         dma_pte_free_pagetable(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
3400
3401         iommu_free_vm_domain(domain);
3402         free_domain_mem(domain);
3403 }
3404
3405 static int intel_iommu_domain_init(struct iommu_domain *domain)
3406 {
3407         struct dmar_domain *dmar_domain;
3408
3409         dmar_domain = iommu_alloc_vm_domain();
3410         if (!dmar_domain) {
3411                 printk(KERN_ERR
3412                         "intel_iommu_domain_init: dmar_domain == NULL\n");
3413                 return -ENOMEM;
3414         }
3415         if (md_domain_init(dmar_domain, DEFAULT_DOMAIN_ADDRESS_WIDTH)) {
3416                 printk(KERN_ERR
3417                         "intel_iommu_domain_init() failed\n");
3418                 vm_domain_exit(dmar_domain);
3419                 return -ENOMEM;
3420         }
3421         domain->priv = dmar_domain;
3422
3423         return 0;
3424 }
3425
3426 static void intel_iommu_domain_destroy(struct iommu_domain *domain)
3427 {
3428         struct dmar_domain *dmar_domain = domain->priv;
3429
3430         domain->priv = NULL;
3431         vm_domain_exit(dmar_domain);
3432 }
3433
3434 static int intel_iommu_attach_device(struct iommu_domain *domain,
3435                                      struct device *dev)
3436 {
3437         struct dmar_domain *dmar_domain = domain->priv;
3438         struct pci_dev *pdev = to_pci_dev(dev);
3439         struct intel_iommu *iommu;
3440         int addr_width;
3441         u64 end;
3442         int ret;
3443
3444         /* normally pdev is not mapped */
3445         if (unlikely(domain_context_mapped(pdev))) {
3446                 struct dmar_domain *old_domain;
3447
3448                 old_domain = find_domain(pdev);
3449                 if (old_domain) {
3450                         if (dmar_domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE ||
3451                             dmar_domain->flags & DOMAIN_FLAG_STATIC_IDENTITY)
3452                                 domain_remove_one_dev_info(old_domain, pdev);
3453                         else
3454                                 domain_remove_dev_info(old_domain);
3455                 }
3456         }
3457
3458         iommu = device_to_iommu(pci_domain_nr(pdev->bus), pdev->bus->number,
3459                                 pdev->devfn);
3460         if (!iommu)
3461                 return -ENODEV;
3462
3463         /* check if this iommu agaw is sufficient for max mapped address */
3464         addr_width = agaw_to_width(iommu->agaw);
3465         end = DOMAIN_MAX_ADDR(addr_width);
3466         end = end & VTD_PAGE_MASK;
3467         if (end < dmar_domain->max_addr) {
3468                 printk(KERN_ERR "%s: iommu agaw (%d) is not "
3469                        "sufficient for the mapped address (%llx)\n",
3470                        __func__, iommu->agaw, dmar_domain->max_addr);
3471                 return -EFAULT;
3472         }
3473
3474         ret = domain_add_dev_info(dmar_domain, pdev);
3475         if (ret)
3476                 return ret;
3477
3478         ret = domain_context_mapping(dmar_domain, pdev, CONTEXT_TT_MULTI_LEVEL);
3479         return ret;
3480 }
3481
3482 static void intel_iommu_detach_device(struct iommu_domain *domain,
3483                                       struct device *dev)
3484 {
3485         struct dmar_domain *dmar_domain = domain->priv;
3486         struct pci_dev *pdev = to_pci_dev(dev);
3487
3488         domain_remove_one_dev_info(dmar_domain, pdev);
3489 }
3490
3491 static int intel_iommu_map_range(struct iommu_domain *domain,
3492                                  unsigned long iova, phys_addr_t hpa,
3493                                  size_t size, int iommu_prot)
3494 {
3495         struct dmar_domain *dmar_domain = domain->priv;
3496         u64 max_addr;
3497         int addr_width;
3498         int prot = 0;
3499         int ret;
3500
3501         if (iommu_prot & IOMMU_READ)
3502                 prot |= DMA_PTE_READ;
3503         if (iommu_prot & IOMMU_WRITE)
3504                 prot |= DMA_PTE_WRITE;
3505         if ((iommu_prot & IOMMU_CACHE) && dmar_domain->iommu_snooping)
3506                 prot |= DMA_PTE_SNP;
3507
3508         max_addr = iova + size;
3509         if (dmar_domain->max_addr < max_addr) {
3510                 int min_agaw;
3511                 u64 end;
3512
3513                 /* check if minimum agaw is sufficient for mapped address */
3514                 min_agaw = vm_domain_min_agaw(dmar_domain);
3515                 addr_width = agaw_to_width(min_agaw);
3516                 end = DOMAIN_MAX_ADDR(addr_width);
3517                 end = end & VTD_PAGE_MASK;
3518                 if (end < max_addr) {
3519                         printk(KERN_ERR "%s: iommu agaw (%d) is not "
3520                                "sufficient for the mapped address (%llx)\n",
3521                                __func__, min_agaw, max_addr);
3522                         return -EFAULT;
3523                 }
3524                 dmar_domain->max_addr = max_addr;
3525         }
3526         /* Round up size to next multiple of PAGE_SIZE, if it and
3527            the low bits of hpa would take us onto the next page */
3528         size = aligned_nrpages(hpa, size);
3529         ret = domain_pfn_mapping(dmar_domain, iova >> VTD_PAGE_SHIFT,
3530                                  hpa >> VTD_PAGE_SHIFT, size, prot);
3531         return ret;
3532 }
3533
3534 static void intel_iommu_unmap_range(struct iommu_domain *domain,
3535                                     unsigned long iova, size_t size)
3536 {
3537         struct dmar_domain *dmar_domain = domain->priv;
3538
3539         dma_pte_clear_range(dmar_domain, iova >> VTD_PAGE_SHIFT,
3540                             (iova + size - 1) >> VTD_PAGE_SHIFT);
3541
3542         if (dmar_domain->max_addr == iova + size)
3543                 dmar_domain->max_addr = iova;
3544 }
3545
3546 static phys_addr_t intel_iommu_iova_to_phys(struct iommu_domain *domain,
3547                                             unsigned long iova)
3548 {
3549         struct dmar_domain *dmar_domain = domain->priv;
3550         struct dma_pte *pte;
3551         u64 phys = 0;
3552
3553         pte = pfn_to_dma_pte(dmar_domain, iova >> VTD_PAGE_SHIFT);
3554         if (pte)
3555                 phys = dma_pte_addr(pte);
3556
3557         return phys;
3558 }
3559
3560 static int intel_iommu_domain_has_cap(struct iommu_domain *domain,
3561                                       unsigned long cap)
3562 {
3563         struct dmar_domain *dmar_domain = domain->priv;
3564
3565         if (cap == IOMMU_CAP_CACHE_COHERENCY)
3566                 return dmar_domain->iommu_snooping;
3567
3568         return 0;
3569 }
3570
3571 static struct iommu_ops intel_iommu_ops = {
3572         .domain_init    = intel_iommu_domain_init,
3573         .domain_destroy = intel_iommu_domain_destroy,
3574         .attach_dev     = intel_iommu_attach_device,
3575         .detach_dev     = intel_iommu_detach_device,
3576         .map            = intel_iommu_map_range,
3577         .unmap          = intel_iommu_unmap_range,
3578         .iova_to_phys   = intel_iommu_iova_to_phys,
3579         .domain_has_cap = intel_iommu_domain_has_cap,
3580 };
3581
3582 static void __devinit quirk_iommu_rwbf(struct pci_dev *dev)
3583 {
3584         /*
3585          * Mobile 4 Series Chipset neglects to set RWBF capability,
3586          * but needs it:
3587          */
3588         printk(KERN_INFO "DMAR: Forcing write-buffer flush capability\n");
3589         rwbf_quirk = 1;
3590 }
3591
3592 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x2a40, quirk_iommu_rwbf);