sdk: include arch/arm/ Linux includes along with arch/arm64/ ones
[oweals/openwrt.git] / target / linux / layerscape / patches-4.9 / 810-iommu-support-layerscape.patch
1 From 0a6c701f92e1aa368c44632fa0985e92703354ed Mon Sep 17 00:00:00 2001
2 From: Yangbo Lu <yangbo.lu@nxp.com>
3 Date: Wed, 17 Jan 2018 15:35:48 +0800
4 Subject: [PATCH 22/30] iommu: support layerscape
5
6 This is an integrated patch for layerscape smmu support.
7
8 Signed-off-by: Eric Auger <eric.auger@redhat.com>
9 Signed-off-by: Robin Murphy <robin.murphy@arm.com>
10 Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
11 Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
12 Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
13 ---
14  drivers/iommu/amd_iommu.c    |  56 ++++++----
15  drivers/iommu/arm-smmu-v3.c  | 111 ++++++++++++++------
16  drivers/iommu/arm-smmu.c     | 100 +++++++++++++++---
17  drivers/iommu/dma-iommu.c    | 242 ++++++++++++++++++++++++++++++++++++-------
18  drivers/iommu/intel-iommu.c  |  92 ++++++++++++----
19  drivers/iommu/iommu.c        | 219 ++++++++++++++++++++++++++++++++++++---
20  drivers/iommu/mtk_iommu.c    |   2 +
21  drivers/iommu/mtk_iommu_v1.c |   2 +
22  include/linux/dma-iommu.h    |  11 ++
23  include/linux/iommu.h        |  55 +++++++---
24  10 files changed, 739 insertions(+), 151 deletions(-)
25
26 --- a/drivers/iommu/amd_iommu.c
27 +++ b/drivers/iommu/amd_iommu.c
28 @@ -373,6 +373,8 @@ static struct iommu_group *acpihid_devic
29  
30         if (!entry->group)
31                 entry->group = generic_device_group(dev);
32 +       else
33 +               iommu_group_ref_get(entry->group);
34  
35         return entry->group;
36  }
37 @@ -3160,9 +3162,10 @@ static bool amd_iommu_capable(enum iommu
38         return false;
39  }
40  
41 -static void amd_iommu_get_dm_regions(struct device *dev,
42 -                                    struct list_head *head)
43 +static void amd_iommu_get_resv_regions(struct device *dev,
44 +                                      struct list_head *head)
45  {
46 +       struct iommu_resv_region *region;
47         struct unity_map_entry *entry;
48         int devid;
49  
50 @@ -3171,41 +3174,56 @@ static void amd_iommu_get_dm_regions(str
51                 return;
52  
53         list_for_each_entry(entry, &amd_iommu_unity_map, list) {
54 -               struct iommu_dm_region *region;
55 +               size_t length;
56 +               int prot = 0;
57  
58                 if (devid < entry->devid_start || devid > entry->devid_end)
59                         continue;
60  
61 -               region = kzalloc(sizeof(*region), GFP_KERNEL);
62 +               length = entry->address_end - entry->address_start;
63 +               if (entry->prot & IOMMU_PROT_IR)
64 +                       prot |= IOMMU_READ;
65 +               if (entry->prot & IOMMU_PROT_IW)
66 +                       prot |= IOMMU_WRITE;
67 +
68 +               region = iommu_alloc_resv_region(entry->address_start,
69 +                                                length, prot,
70 +                                                IOMMU_RESV_DIRECT);
71                 if (!region) {
72                         pr_err("Out of memory allocating dm-regions for %s\n",
73                                 dev_name(dev));
74                         return;
75                 }
76 -
77 -               region->start = entry->address_start;
78 -               region->length = entry->address_end - entry->address_start;
79 -               if (entry->prot & IOMMU_PROT_IR)
80 -                       region->prot |= IOMMU_READ;
81 -               if (entry->prot & IOMMU_PROT_IW)
82 -                       region->prot |= IOMMU_WRITE;
83 -
84                 list_add_tail(&region->list, head);
85         }
86 +
87 +       region = iommu_alloc_resv_region(MSI_RANGE_START,
88 +                                        MSI_RANGE_END - MSI_RANGE_START + 1,
89 +                                        0, IOMMU_RESV_MSI);
90 +       if (!region)
91 +               return;
92 +       list_add_tail(&region->list, head);
93 +
94 +       region = iommu_alloc_resv_region(HT_RANGE_START,
95 +                                        HT_RANGE_END - HT_RANGE_START + 1,
96 +                                        0, IOMMU_RESV_RESERVED);
97 +       if (!region)
98 +               return;
99 +       list_add_tail(&region->list, head);
100  }
101  
102 -static void amd_iommu_put_dm_regions(struct device *dev,
103 +static void amd_iommu_put_resv_regions(struct device *dev,
104                                      struct list_head *head)
105  {
106 -       struct iommu_dm_region *entry, *next;
107 +       struct iommu_resv_region *entry, *next;
108  
109         list_for_each_entry_safe(entry, next, head, list)
110                 kfree(entry);
111  }
112  
113 -static void amd_iommu_apply_dm_region(struct device *dev,
114 +static void amd_iommu_apply_resv_region(struct device *dev,
115                                       struct iommu_domain *domain,
116 -                                     struct iommu_dm_region *region)
117 +                                     struct iommu_resv_region *region)
118  {
119         struct dma_ops_domain *dma_dom = to_dma_ops_domain(to_pdomain(domain));
120         unsigned long start, end;
121 @@ -3229,9 +3247,9 @@ static const struct iommu_ops amd_iommu_
122         .add_device = amd_iommu_add_device,
123         .remove_device = amd_iommu_remove_device,
124         .device_group = amd_iommu_device_group,
125 -       .get_dm_regions = amd_iommu_get_dm_regions,
126 -       .put_dm_regions = amd_iommu_put_dm_regions,
127 -       .apply_dm_region = amd_iommu_apply_dm_region,
128 +       .get_resv_regions = amd_iommu_get_resv_regions,
129 +       .put_resv_regions = amd_iommu_put_resv_regions,
130 +       .apply_resv_region = amd_iommu_apply_resv_region,
131         .pgsize_bitmap  = AMD_IOMMU_PGSIZES,
132  };
133  
134 --- a/drivers/iommu/arm-smmu-v3.c
135 +++ b/drivers/iommu/arm-smmu-v3.c
136 @@ -410,6 +410,9 @@
137  /* High-level queue structures */
138  #define ARM_SMMU_POLL_TIMEOUT_US       100
139  
140 +#define MSI_IOVA_BASE                  0x8000000
141 +#define MSI_IOVA_LENGTH                        0x100000
142 +
143  static bool disable_bypass;
144  module_param_named(disable_bypass, disable_bypass, bool, S_IRUGO);
145  MODULE_PARM_DESC(disable_bypass,
146 @@ -552,9 +555,14 @@ struct arm_smmu_s2_cfg {
147  };
148  
149  struct arm_smmu_strtab_ent {
150 -       bool                            valid;
151 -
152 -       bool                            bypass; /* Overrides s1/s2 config */
153 +       /*
154 +        * An STE is "assigned" if the master emitting the corresponding SID
155 +        * is attached to a domain. The behaviour of an unassigned STE is
156 +        * determined by the disable_bypass parameter, whereas an assigned
157 +        * STE behaves according to s1_cfg/s2_cfg, which themselves are
158 +        * configured according to the domain type.
159 +        */
160 +       bool                            assigned;
161         struct arm_smmu_s1_cfg          *s1_cfg;
162         struct arm_smmu_s2_cfg          *s2_cfg;
163  };
164 @@ -627,6 +635,7 @@ enum arm_smmu_domain_stage {
165         ARM_SMMU_DOMAIN_S1 = 0,
166         ARM_SMMU_DOMAIN_S2,
167         ARM_SMMU_DOMAIN_NESTED,
168 +       ARM_SMMU_DOMAIN_BYPASS,
169  };
170  
171  struct arm_smmu_domain {
172 @@ -1000,9 +1009,9 @@ static void arm_smmu_write_strtab_ent(st
173          * This is hideously complicated, but we only really care about
174          * three cases at the moment:
175          *
176 -        * 1. Invalid (all zero) -> bypass  (init)
177 -        * 2. Bypass -> translation (attach)
178 -        * 3. Translation -> bypass (detach)
179 +        * 1. Invalid (all zero) -> bypass/fault (init)
180 +        * 2. Bypass/fault -> translation/bypass (attach)
181 +        * 3. Translation/bypass -> bypass/fault (detach)
182          *
183          * Given that we can't update the STE atomically and the SMMU
184          * doesn't read the thing in a defined order, that leaves us
185 @@ -1041,11 +1050,15 @@ static void arm_smmu_write_strtab_ent(st
186         }
187  
188         /* Nuke the existing STE_0 value, as we're going to rewrite it */
189 -       val = ste->valid ? STRTAB_STE_0_V : 0;
190 +       val = STRTAB_STE_0_V;
191 +
192 +       /* Bypass/fault */
193 +       if (!ste->assigned || !(ste->s1_cfg || ste->s2_cfg)) {
194 +               if (!ste->assigned && disable_bypass)
195 +                       val |= STRTAB_STE_0_CFG_ABORT;
196 +               else
197 +                       val |= STRTAB_STE_0_CFG_BYPASS;
198  
199 -       if (ste->bypass) {
200 -               val |= disable_bypass ? STRTAB_STE_0_CFG_ABORT
201 -                                     : STRTAB_STE_0_CFG_BYPASS;
202                 dst[0] = cpu_to_le64(val);
203                 dst[1] = cpu_to_le64(STRTAB_STE_1_SHCFG_INCOMING
204                          << STRTAB_STE_1_SHCFG_SHIFT);
205 @@ -1108,10 +1121,7 @@ static void arm_smmu_write_strtab_ent(st
206  static void arm_smmu_init_bypass_stes(u64 *strtab, unsigned int nent)
207  {
208         unsigned int i;
209 -       struct arm_smmu_strtab_ent ste = {
210 -               .valid  = true,
211 -               .bypass = true,
212 -       };
213 +       struct arm_smmu_strtab_ent ste = { .assigned = false };
214  
215         for (i = 0; i < nent; ++i) {
216                 arm_smmu_write_strtab_ent(NULL, -1, strtab, &ste);
217 @@ -1364,8 +1374,6 @@ static bool arm_smmu_capable(enum iommu_
218         switch (cap) {
219         case IOMMU_CAP_CACHE_COHERENCY:
220                 return true;
221 -       case IOMMU_CAP_INTR_REMAP:
222 -               return true; /* MSIs are just memory writes */
223         case IOMMU_CAP_NOEXEC:
224                 return true;
225         default:
226 @@ -1377,7 +1385,9 @@ static struct iommu_domain *arm_smmu_dom
227  {
228         struct arm_smmu_domain *smmu_domain;
229  
230 -       if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
231 +       if (type != IOMMU_DOMAIN_UNMANAGED &&
232 +           type != IOMMU_DOMAIN_DMA &&
233 +           type != IOMMU_DOMAIN_IDENTITY)
234                 return NULL;
235  
236         /*
237 @@ -1508,6 +1518,11 @@ static int arm_smmu_domain_finalise(stru
238         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
239         struct arm_smmu_device *smmu = smmu_domain->smmu;
240  
241 +       if (domain->type == IOMMU_DOMAIN_IDENTITY) {
242 +               smmu_domain->stage = ARM_SMMU_DOMAIN_BYPASS;
243 +               return 0;
244 +       }
245 +
246         /* Restrict the stage to what we can actually support */
247         if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S1))
248                 smmu_domain->stage = ARM_SMMU_DOMAIN_S2;
249 @@ -1580,7 +1595,7 @@ static __le64 *arm_smmu_get_step_for_sid
250         return step;
251  }
252  
253 -static int arm_smmu_install_ste_for_dev(struct iommu_fwspec *fwspec)
254 +static void arm_smmu_install_ste_for_dev(struct iommu_fwspec *fwspec)
255  {
256         int i, j;
257         struct arm_smmu_master_data *master = fwspec->iommu_priv;
258 @@ -1599,17 +1614,14 @@ static int arm_smmu_install_ste_for_dev(
259  
260                 arm_smmu_write_strtab_ent(smmu, sid, step, &master->ste);
261         }
262 -
263 -       return 0;
264  }
265  
266  static void arm_smmu_detach_dev(struct device *dev)
267  {
268         struct arm_smmu_master_data *master = dev->iommu_fwspec->iommu_priv;
269  
270 -       master->ste.bypass = true;
271 -       if (arm_smmu_install_ste_for_dev(dev->iommu_fwspec) < 0)
272 -               dev_warn(dev, "failed to install bypass STE\n");
273 +       master->ste.assigned = false;
274 +       arm_smmu_install_ste_for_dev(dev->iommu_fwspec);
275  }
276  
277  static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
278 @@ -1628,7 +1640,7 @@ static int arm_smmu_attach_dev(struct io
279         ste = &master->ste;
280  
281         /* Already attached to a different domain? */
282 -       if (!ste->bypass)
283 +       if (ste->assigned)
284                 arm_smmu_detach_dev(dev);
285  
286         mutex_lock(&smmu_domain->init_mutex);
287 @@ -1649,10 +1661,12 @@ static int arm_smmu_attach_dev(struct io
288                 goto out_unlock;
289         }
290  
291 -       ste->bypass = false;
292 -       ste->valid = true;
293 +       ste->assigned = true;
294  
295 -       if (smmu_domain->stage == ARM_SMMU_DOMAIN_S1) {
296 +       if (smmu_domain->stage == ARM_SMMU_DOMAIN_BYPASS) {
297 +               ste->s1_cfg = NULL;
298 +               ste->s2_cfg = NULL;
299 +       } else if (smmu_domain->stage == ARM_SMMU_DOMAIN_S1) {
300                 ste->s1_cfg = &smmu_domain->s1_cfg;
301                 ste->s2_cfg = NULL;
302                 arm_smmu_write_ctx_desc(smmu, ste->s1_cfg);
303 @@ -1661,10 +1675,7 @@ static int arm_smmu_attach_dev(struct io
304                 ste->s2_cfg = &smmu_domain->s2_cfg;
305         }
306  
307 -       ret = arm_smmu_install_ste_for_dev(dev->iommu_fwspec);
308 -       if (ret < 0)
309 -               ste->valid = false;
310 -
311 +       arm_smmu_install_ste_for_dev(dev->iommu_fwspec);
312  out_unlock:
313         mutex_unlock(&smmu_domain->init_mutex);
314         return ret;
315 @@ -1712,6 +1723,9 @@ arm_smmu_iova_to_phys(struct iommu_domai
316         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
317         struct io_pgtable_ops *ops = smmu_domain->pgtbl_ops;
318  
319 +       if (domain->type == IOMMU_DOMAIN_IDENTITY)
320 +               return iova;
321 +
322         if (!ops)
323                 return 0;
324  
325 @@ -1810,7 +1824,7 @@ static void arm_smmu_remove_device(struc
326                 return;
327  
328         master = fwspec->iommu_priv;
329 -       if (master && master->ste.valid)
330 +       if (master && master->ste.assigned)
331                 arm_smmu_detach_dev(dev);
332         iommu_group_remove_device(dev);
333         kfree(master);
334 @@ -1839,6 +1853,9 @@ static int arm_smmu_domain_get_attr(stru
335  {
336         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
337  
338 +       if (domain->type != IOMMU_DOMAIN_UNMANAGED)
339 +               return -EINVAL;
340 +
341         switch (attr) {
342         case DOMAIN_ATTR_NESTING:
343                 *(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
344 @@ -1854,6 +1871,9 @@ static int arm_smmu_domain_set_attr(stru
345         int ret = 0;
346         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
347  
348 +       if (domain->type != IOMMU_DOMAIN_UNMANAGED)
349 +               return -EINVAL;
350 +
351         mutex_lock(&smmu_domain->init_mutex);
352  
353         switch (attr) {
354 @@ -1883,6 +1903,31 @@ static int arm_smmu_of_xlate(struct devi
355         return iommu_fwspec_add_ids(dev, args->args, 1);
356  }
357  
358 +static void arm_smmu_get_resv_regions(struct device *dev,
359 +                                     struct list_head *head)
360 +{
361 +       struct iommu_resv_region *region;
362 +       int prot = IOMMU_WRITE | IOMMU_NOEXEC | IOMMU_MMIO;
363 +
364 +       region = iommu_alloc_resv_region(MSI_IOVA_BASE, MSI_IOVA_LENGTH,
365 +                                        prot, IOMMU_RESV_SW_MSI);
366 +       if (!region)
367 +               return;
368 +
369 +       list_add_tail(&region->list, head);
370 +
371 +       iommu_dma_get_resv_regions(dev, head);
372 +}
373 +
374 +static void arm_smmu_put_resv_regions(struct device *dev,
375 +                                     struct list_head *head)
376 +{
377 +       struct iommu_resv_region *entry, *next;
378 +
379 +       list_for_each_entry_safe(entry, next, head, list)
380 +               kfree(entry);
381 +}
382 +
383  static struct iommu_ops arm_smmu_ops = {
384         .capable                = arm_smmu_capable,
385         .domain_alloc           = arm_smmu_domain_alloc,
386 @@ -1898,6 +1943,8 @@ static struct iommu_ops arm_smmu_ops = {
387         .domain_get_attr        = arm_smmu_domain_get_attr,
388         .domain_set_attr        = arm_smmu_domain_set_attr,
389         .of_xlate               = arm_smmu_of_xlate,
390 +       .get_resv_regions       = arm_smmu_get_resv_regions,
391 +       .put_resv_regions       = arm_smmu_put_resv_regions,
392         .pgsize_bitmap          = -1UL, /* Restricted during device attach */
393  };
394  
395 --- a/drivers/iommu/arm-smmu.c
396 +++ b/drivers/iommu/arm-smmu.c
397 @@ -49,6 +49,7 @@
398  #include <linux/spinlock.h>
399  
400  #include <linux/amba/bus.h>
401 +#include "../staging/fsl-mc/include/mc-bus.h"
402  
403  #include "io-pgtable.h"
404  
405 @@ -247,6 +248,7 @@ enum arm_smmu_s2cr_privcfg {
406  #define ARM_MMU500_ACTLR_CPRE          (1 << 1)
407  
408  #define ARM_MMU500_ACR_CACHE_LOCK      (1 << 26)
409 +#define ARM_MMU500_ACR_SMTNMB_TLBEN    (1 << 8)
410  
411  #define CB_PAR_F                       (1 << 0)
412  
413 @@ -278,6 +280,9 @@ enum arm_smmu_s2cr_privcfg {
414  
415  #define FSYNR0_WNR                     (1 << 4)
416  
417 +#define MSI_IOVA_BASE                  0x8000000
418 +#define MSI_IOVA_LENGTH                        0x100000
419 +
420  static int force_stage;
421  module_param(force_stage, int, S_IRUGO);
422  MODULE_PARM_DESC(force_stage,
423 @@ -401,6 +406,7 @@ enum arm_smmu_domain_stage {
424         ARM_SMMU_DOMAIN_S1 = 0,
425         ARM_SMMU_DOMAIN_S2,
426         ARM_SMMU_DOMAIN_NESTED,
427 +       ARM_SMMU_DOMAIN_BYPASS,
428  };
429  
430  struct arm_smmu_domain {
431 @@ -821,6 +827,12 @@ static int arm_smmu_init_domain_context(
432         if (smmu_domain->smmu)
433                 goto out_unlock;
434  
435 +       if (domain->type == IOMMU_DOMAIN_IDENTITY) {
436 +               smmu_domain->stage = ARM_SMMU_DOMAIN_BYPASS;
437 +               smmu_domain->smmu = smmu;
438 +               goto out_unlock;
439 +       }
440 +
441         /*
442          * Mapping the requested stage onto what we support is surprisingly
443          * complicated, mainly because the spec allows S1+S2 SMMUs without
444 @@ -981,7 +993,7 @@ static void arm_smmu_destroy_domain_cont
445         void __iomem *cb_base;
446         int irq;
447  
448 -       if (!smmu)
449 +       if (!smmu || domain->type == IOMMU_DOMAIN_IDENTITY)
450                 return;
451  
452         /*
453 @@ -1004,7 +1016,9 @@ static struct iommu_domain *arm_smmu_dom
454  {
455         struct arm_smmu_domain *smmu_domain;
456  
457 -       if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
458 +       if (type != IOMMU_DOMAIN_UNMANAGED &&
459 +           type != IOMMU_DOMAIN_DMA &&
460 +           type != IOMMU_DOMAIN_IDENTITY)
461                 return NULL;
462         /*
463          * Allocate the domain and initialise some of its data structures.
464 @@ -1202,10 +1216,15 @@ static int arm_smmu_domain_add_master(st
465  {
466         struct arm_smmu_device *smmu = smmu_domain->smmu;
467         struct arm_smmu_s2cr *s2cr = smmu->s2crs;
468 -       enum arm_smmu_s2cr_type type = S2CR_TYPE_TRANS;
469         u8 cbndx = smmu_domain->cfg.cbndx;
470 +       enum arm_smmu_s2cr_type type;
471         int i, idx;
472  
473 +       if (smmu_domain->stage == ARM_SMMU_DOMAIN_BYPASS)
474 +               type = S2CR_TYPE_BYPASS;
475 +       else
476 +               type = S2CR_TYPE_TRANS;
477 +
478         for_each_cfg_sme(fwspec, i, idx) {
479                 if (type == s2cr[idx].type && cbndx == s2cr[idx].cbndx)
480                         continue;
481 @@ -1343,6 +1362,9 @@ static phys_addr_t arm_smmu_iova_to_phys
482         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
483         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
484  
485 +       if (domain->type == IOMMU_DOMAIN_IDENTITY)
486 +               return iova;
487 +
488         if (!ops)
489                 return 0;
490  
491 @@ -1368,8 +1390,6 @@ static bool arm_smmu_capable(enum iommu_
492                  * requests.
493                  */
494                 return true;
495 -       case IOMMU_CAP_INTR_REMAP:
496 -               return true; /* MSIs are just memory writes */
497         case IOMMU_CAP_NOEXEC:
498                 return true;
499         default:
500 @@ -1478,10 +1498,12 @@ static struct iommu_group *arm_smmu_devi
501         }
502  
503         if (group)
504 -               return group;
505 +               return iommu_group_ref_get(group);
506  
507         if (dev_is_pci(dev))
508                 group = pci_device_group(dev);
509 +       else if (dev_is_fsl_mc(dev))
510 +               group = fsl_mc_device_group(dev);
511         else
512                 group = generic_device_group(dev);
513  
514 @@ -1493,6 +1515,9 @@ static int arm_smmu_domain_get_attr(stru
515  {
516         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
517  
518 +       if (domain->type != IOMMU_DOMAIN_UNMANAGED)
519 +               return -EINVAL;
520 +
521         switch (attr) {
522         case DOMAIN_ATTR_NESTING:
523                 *(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
524 @@ -1508,6 +1533,9 @@ static int arm_smmu_domain_set_attr(stru
525         int ret = 0;
526         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
527  
528 +       if (domain->type != IOMMU_DOMAIN_UNMANAGED)
529 +               return -EINVAL;
530 +
531         mutex_lock(&smmu_domain->init_mutex);
532  
533         switch (attr) {
534 @@ -1534,17 +1562,44 @@ out_unlock:
535  
536  static int arm_smmu_of_xlate(struct device *dev, struct of_phandle_args *args)
537  {
538 -       u32 fwid = 0;
539 +       u32 mask, fwid = 0;
540  
541         if (args->args_count > 0)
542                 fwid |= (u16)args->args[0];
543  
544         if (args->args_count > 1)
545                 fwid |= (u16)args->args[1] << SMR_MASK_SHIFT;
546 +       else if (!of_property_read_u32(args->np, "stream-match-mask", &mask))
547 +               fwid |= (u16)mask << SMR_MASK_SHIFT;
548  
549         return iommu_fwspec_add_ids(dev, &fwid, 1);
550  }
551  
552 +static void arm_smmu_get_resv_regions(struct device *dev,
553 +                                     struct list_head *head)
554 +{
555 +       struct iommu_resv_region *region;
556 +       int prot = IOMMU_WRITE | IOMMU_NOEXEC | IOMMU_MMIO;
557 +
558 +       region = iommu_alloc_resv_region(MSI_IOVA_BASE, MSI_IOVA_LENGTH,
559 +                                        prot, IOMMU_RESV_SW_MSI);
560 +       if (!region)
561 +               return;
562 +
563 +       list_add_tail(&region->list, head);
564 +
565 +       iommu_dma_get_resv_regions(dev, head);
566 +}
567 +
568 +static void arm_smmu_put_resv_regions(struct device *dev,
569 +                                     struct list_head *head)
570 +{
571 +       struct iommu_resv_region *entry, *next;
572 +
573 +       list_for_each_entry_safe(entry, next, head, list)
574 +               kfree(entry);
575 +}
576 +
577  static struct iommu_ops arm_smmu_ops = {
578         .capable                = arm_smmu_capable,
579         .domain_alloc           = arm_smmu_domain_alloc,
580 @@ -1560,6 +1615,8 @@ static struct iommu_ops arm_smmu_ops = {
581         .domain_get_attr        = arm_smmu_domain_get_attr,
582         .domain_set_attr        = arm_smmu_domain_set_attr,
583         .of_xlate               = arm_smmu_of_xlate,
584 +       .get_resv_regions       = arm_smmu_get_resv_regions,
585 +       .put_resv_regions       = arm_smmu_put_resv_regions,
586         .pgsize_bitmap          = -1UL, /* Restricted during device attach */
587  };
588  
589 @@ -1581,16 +1638,22 @@ static void arm_smmu_device_reset(struct
590         for (i = 0; i < smmu->num_mapping_groups; ++i)
591                 arm_smmu_write_sme(smmu, i);
592  
593 -       /*
594 -        * Before clearing ARM_MMU500_ACTLR_CPRE, need to
595 -        * clear CACHE_LOCK bit of ACR first. And, CACHE_LOCK
596 -        * bit is only present in MMU-500r2 onwards.
597 -        */
598 -       reg = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID7);
599 -       major = (reg >> ID7_MAJOR_SHIFT) & ID7_MAJOR_MASK;
600 -       if ((smmu->model == ARM_MMU500) && (major >= 2)) {
601 +       if (smmu->model == ARM_MMU500) {
602 +               /*
603 +                * Before clearing ARM_MMU500_ACTLR_CPRE, need to
604 +                * clear CACHE_LOCK bit of ACR first. And, CACHE_LOCK
605 +                * bit is only present in MMU-500r2 onwards.
606 +                */
607 +               reg = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID7);
608 +               major = (reg >> ID7_MAJOR_SHIFT) & ID7_MAJOR_MASK;
609                 reg = readl_relaxed(gr0_base + ARM_SMMU_GR0_sACR);
610 -               reg &= ~ARM_MMU500_ACR_CACHE_LOCK;
611 +               if (major >= 2)
612 +                       reg &= ~ARM_MMU500_ACR_CACHE_LOCK;
613 +               /*
614 +                * Allow unmatched Stream IDs to allocate bypass
615 +                * TLB entries for reduced latency.
616 +                */
617 +               reg |= ARM_MMU500_ACR_SMTNMB_TLBEN;
618                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_sACR);
619         }
620  
621 @@ -2024,6 +2087,11 @@ static int arm_smmu_device_dt_probe(stru
622                 bus_set_iommu(&pci_bus_type, &arm_smmu_ops);
623         }
624  #endif
625 +#ifdef CONFIG_FSL_MC_BUS
626 +       if (!iommu_present(&fsl_mc_bus_type))
627 +               bus_set_iommu(&fsl_mc_bus_type, &arm_smmu_ops);
628 +#endif
629 +
630         return 0;
631  }
632  
633 --- a/drivers/iommu/dma-iommu.c
634 +++ b/drivers/iommu/dma-iommu.c
635 @@ -37,15 +37,50 @@ struct iommu_dma_msi_page {
636         phys_addr_t             phys;
637  };
638  
639 +enum iommu_dma_cookie_type {
640 +       IOMMU_DMA_IOVA_COOKIE,
641 +       IOMMU_DMA_MSI_COOKIE,
642 +};
643 +
644  struct iommu_dma_cookie {
645 -       struct iova_domain      iovad;
646 -       struct list_head        msi_page_list;
647 -       spinlock_t              msi_lock;
648 +       enum iommu_dma_cookie_type      type;
649 +       union {
650 +               /* Full allocator for IOMMU_DMA_IOVA_COOKIE */
651 +               struct iova_domain      iovad;
652 +               /* Trivial linear page allocator for IOMMU_DMA_MSI_COOKIE */
653 +               dma_addr_t              msi_iova;
654 +       };
655 +       struct list_head                msi_page_list;
656 +       spinlock_t                      msi_lock;
657  };
658  
659 +static inline size_t cookie_msi_granule(struct iommu_dma_cookie *cookie)
660 +{
661 +       if (cookie->type == IOMMU_DMA_IOVA_COOKIE)
662 +               return cookie->iovad.granule;
663 +       return PAGE_SIZE;
664 +}
665 +
666  static inline struct iova_domain *cookie_iovad(struct iommu_domain *domain)
667  {
668 -       return &((struct iommu_dma_cookie *)domain->iova_cookie)->iovad;
669 +       struct iommu_dma_cookie *cookie = domain->iova_cookie;
670 +
671 +       if (cookie->type == IOMMU_DMA_IOVA_COOKIE)
672 +               return &cookie->iovad;
673 +       return NULL;
674 +}
675 +
676 +static struct iommu_dma_cookie *cookie_alloc(enum iommu_dma_cookie_type type)
677 +{
678 +       struct iommu_dma_cookie *cookie;
679 +
680 +       cookie = kzalloc(sizeof(*cookie), GFP_KERNEL);
681 +       if (cookie) {
682 +               spin_lock_init(&cookie->msi_lock);
683 +               INIT_LIST_HEAD(&cookie->msi_page_list);
684 +               cookie->type = type;
685 +       }
686 +       return cookie;
687  }
688  
689  int iommu_dma_init(void)
690 @@ -62,25 +97,53 @@ int iommu_dma_init(void)
691   */
692  int iommu_get_dma_cookie(struct iommu_domain *domain)
693  {
694 +       if (domain->iova_cookie)
695 +               return -EEXIST;
696 +
697 +       domain->iova_cookie = cookie_alloc(IOMMU_DMA_IOVA_COOKIE);
698 +       if (!domain->iova_cookie)
699 +               return -ENOMEM;
700 +
701 +       return 0;
702 +}
703 +EXPORT_SYMBOL(iommu_get_dma_cookie);
704 +
705 +/**
706 + * iommu_get_msi_cookie - Acquire just MSI remapping resources
707 + * @domain: IOMMU domain to prepare
708 + * @base: Start address of IOVA region for MSI mappings
709 + *
710 + * Users who manage their own IOVA allocation and do not want DMA API support,
711 + * but would still like to take advantage of automatic MSI remapping, can use
712 + * this to initialise their own domain appropriately. Users should reserve a
713 + * contiguous IOVA region, starting at @base, large enough to accommodate the
714 + * number of PAGE_SIZE mappings necessary to cover every MSI doorbell address
715 + * used by the devices attached to @domain.
716 + */
717 +int iommu_get_msi_cookie(struct iommu_domain *domain, dma_addr_t base)
718 +{
719         struct iommu_dma_cookie *cookie;
720  
721 +       if (domain->type != IOMMU_DOMAIN_UNMANAGED)
722 +               return -EINVAL;
723 +
724         if (domain->iova_cookie)
725                 return -EEXIST;
726  
727 -       cookie = kzalloc(sizeof(*cookie), GFP_KERNEL);
728 +       cookie = cookie_alloc(IOMMU_DMA_MSI_COOKIE);
729         if (!cookie)
730                 return -ENOMEM;
731  
732 -       spin_lock_init(&cookie->msi_lock);
733 -       INIT_LIST_HEAD(&cookie->msi_page_list);
734 +       cookie->msi_iova = base;
735         domain->iova_cookie = cookie;
736         return 0;
737  }
738 -EXPORT_SYMBOL(iommu_get_dma_cookie);
739 +EXPORT_SYMBOL(iommu_get_msi_cookie);
740  
741  /**
742   * iommu_put_dma_cookie - Release a domain's DMA mapping resources
743 - * @domain: IOMMU domain previously prepared by iommu_get_dma_cookie()
744 + * @domain: IOMMU domain previously prepared by iommu_get_dma_cookie() or
745 + *          iommu_get_msi_cookie()
746   *
747   * IOMMU drivers should normally call this from their domain_free callback.
748   */
749 @@ -92,7 +155,7 @@ void iommu_put_dma_cookie(struct iommu_d
750         if (!cookie)
751                 return;
752  
753 -       if (cookie->iovad.granule)
754 +       if (cookie->type == IOMMU_DMA_IOVA_COOKIE && cookie->iovad.granule)
755                 put_iova_domain(&cookie->iovad);
756  
757         list_for_each_entry_safe(msi, tmp, &cookie->msi_page_list, list) {
758 @@ -104,21 +167,99 @@ void iommu_put_dma_cookie(struct iommu_d
759  }
760  EXPORT_SYMBOL(iommu_put_dma_cookie);
761  
762 -static void iova_reserve_pci_windows(struct pci_dev *dev,
763 -               struct iova_domain *iovad)
764 +/**
765 + * iommu_dma_get_resv_regions - Reserved region driver helper
766 + * @dev: Device from iommu_get_resv_regions()
767 + * @list: Reserved region list from iommu_get_resv_regions()
768 + *
769 + * IOMMU drivers can use this to implement their .get_resv_regions callback
770 + * for general non-IOMMU-specific reservations. Currently, this covers host
771 + * bridge windows for PCI devices.
772 + */
773 +void iommu_dma_get_resv_regions(struct device *dev, struct list_head *list)
774  {
775 -       struct pci_host_bridge *bridge = pci_find_host_bridge(dev->bus);
776 +       struct pci_host_bridge *bridge;
777         struct resource_entry *window;
778 -       unsigned long lo, hi;
779  
780 +       if (!dev_is_pci(dev))
781 +               return;
782 +
783 +       bridge = pci_find_host_bridge(to_pci_dev(dev)->bus);
784         resource_list_for_each_entry(window, &bridge->windows) {
785 +               struct iommu_resv_region *region;
786 +               phys_addr_t start;
787 +               size_t length;
788 +
789                 if (resource_type(window->res) != IORESOURCE_MEM)
790                         continue;
791  
792 -               lo = iova_pfn(iovad, window->res->start - window->offset);
793 -               hi = iova_pfn(iovad, window->res->end - window->offset);
794 +               start = window->res->start - window->offset;
795 +               length = window->res->end - window->res->start + 1;
796 +               region = iommu_alloc_resv_region(start, length, 0,
797 +                               IOMMU_RESV_RESERVED);
798 +               if (!region)
799 +                       return;
800 +
801 +               list_add_tail(&region->list, list);
802 +       }
803 +}
804 +EXPORT_SYMBOL(iommu_dma_get_resv_regions);
805 +
806 +static int cookie_init_hw_msi_region(struct iommu_dma_cookie *cookie,
807 +               phys_addr_t start, phys_addr_t end)
808 +{
809 +       struct iova_domain *iovad = &cookie->iovad;
810 +       struct iommu_dma_msi_page *msi_page;
811 +       int i, num_pages;
812 +
813 +       start -= iova_offset(iovad, start);
814 +       num_pages = iova_align(iovad, end - start) >> iova_shift(iovad);
815 +
816 +       msi_page = kcalloc(num_pages, sizeof(*msi_page), GFP_KERNEL);
817 +       if (!msi_page)
818 +               return -ENOMEM;
819 +
820 +       for (i = 0; i < num_pages; i++) {
821 +               msi_page[i].phys = start;
822 +               msi_page[i].iova = start;
823 +               INIT_LIST_HEAD(&msi_page[i].list);
824 +               list_add(&msi_page[i].list, &cookie->msi_page_list);
825 +               start += iovad->granule;
826 +       }
827 +
828 +       return 0;
829 +}
830 +
831 +static int iova_reserve_iommu_regions(struct device *dev,
832 +               struct iommu_domain *domain)
833 +{
834 +       struct iommu_dma_cookie *cookie = domain->iova_cookie;
835 +       struct iova_domain *iovad = &cookie->iovad;
836 +       struct iommu_resv_region *region;
837 +       LIST_HEAD(resv_regions);
838 +       int ret = 0;
839 +
840 +       iommu_get_resv_regions(dev, &resv_regions);
841 +       list_for_each_entry(region, &resv_regions, list) {
842 +               unsigned long lo, hi;
843 +
844 +               /* We ARE the software that manages these! */
845 +               if (region->type == IOMMU_RESV_SW_MSI)
846 +                       continue;
847 +
848 +               lo = iova_pfn(iovad, region->start);
849 +               hi = iova_pfn(iovad, region->start + region->length - 1);
850                 reserve_iova(iovad, lo, hi);
851 +
852 +               if (region->type == IOMMU_RESV_MSI)
853 +                       ret = cookie_init_hw_msi_region(cookie, region->start,
854 +                                       region->start + region->length);
855 +               if (ret)
856 +                       break;
857         }
858 +       iommu_put_resv_regions(dev, &resv_regions);
859 +
860 +       return ret;
861  }
862  
863  /**
864 @@ -136,11 +277,12 @@ static void iova_reserve_pci_windows(str
865  int iommu_dma_init_domain(struct iommu_domain *domain, dma_addr_t base,
866                 u64 size, struct device *dev)
867  {
868 -       struct iova_domain *iovad = cookie_iovad(domain);
869 +       struct iommu_dma_cookie *cookie = domain->iova_cookie;
870 +       struct iova_domain *iovad = &cookie->iovad;
871         unsigned long order, base_pfn, end_pfn;
872  
873 -       if (!iovad)
874 -               return -ENODEV;
875 +       if (!cookie || cookie->type != IOMMU_DMA_IOVA_COOKIE)
876 +               return -EINVAL;
877  
878         /* Use the smallest supported page size for IOVA granularity */
879         order = __ffs(domain->pgsize_bitmap);
880 @@ -160,22 +302,37 @@ int iommu_dma_init_domain(struct iommu_d
881                 end_pfn = min_t(unsigned long, end_pfn,
882                                 domain->geometry.aperture_end >> order);
883         }
884 +       /*
885 +        * PCI devices may have larger DMA masks, but still prefer allocating
886 +        * within a 32-bit mask to avoid DAC addressing. Such limitations don't
887 +        * apply to the typical platform device, so for those we may as well
888 +        * leave the cache limit at the top of their range to save an rb_last()
889 +        * traversal on every allocation.
890 +        */
891 +       if (dev && dev_is_pci(dev))
892 +               end_pfn &= DMA_BIT_MASK(32) >> order;
893  
894 -       /* All we can safely do with an existing domain is enlarge it */
895 +       /* start_pfn is always nonzero for an already-initialised domain */
896         if (iovad->start_pfn) {
897                 if (1UL << order != iovad->granule ||
898 -                   base_pfn != iovad->start_pfn ||
899 -                   end_pfn < iovad->dma_32bit_pfn) {
900 +                   base_pfn != iovad->start_pfn) {
901                         pr_warn("Incompatible range for DMA domain\n");
902                         return -EFAULT;
903                 }
904 -               iovad->dma_32bit_pfn = end_pfn;
905 -       } else {
906 -               init_iova_domain(iovad, 1UL << order, base_pfn, end_pfn);
907 -               if (dev && dev_is_pci(dev))
908 -                       iova_reserve_pci_windows(to_pci_dev(dev), iovad);
909 +               /*
910 +                * If we have devices with different DMA masks, move the free
911 +                * area cache limit down for the benefit of the smaller one.
912 +                */
913 +               iovad->dma_32bit_pfn = min(end_pfn, iovad->dma_32bit_pfn);
914 +
915 +               return 0;
916         }
917 -       return 0;
918 +
919 +       init_iova_domain(iovad, 1UL << order, base_pfn, end_pfn);
920 +       if (!dev)
921 +               return 0;
922 +
923 +       return iova_reserve_iommu_regions(dev, domain);
924  }
925  EXPORT_SYMBOL(iommu_dma_init_domain);
926  
927 @@ -643,11 +800,12 @@ static struct iommu_dma_msi_page *iommu_
928  {
929         struct iommu_dma_cookie *cookie = domain->iova_cookie;
930         struct iommu_dma_msi_page *msi_page;
931 -       struct iova_domain *iovad = &cookie->iovad;
932 +       struct iova_domain *iovad = cookie_iovad(domain);
933         struct iova *iova;
934         int prot = IOMMU_WRITE | IOMMU_NOEXEC | IOMMU_MMIO;
935 +       size_t size = cookie_msi_granule(cookie);
936  
937 -       msi_addr &= ~(phys_addr_t)iova_mask(iovad);
938 +       msi_addr &= ~(phys_addr_t)(size - 1);
939         list_for_each_entry(msi_page, &cookie->msi_page_list, list)
940                 if (msi_page->phys == msi_addr)
941                         return msi_page;
942 @@ -656,13 +814,18 @@ static struct iommu_dma_msi_page *iommu_
943         if (!msi_page)
944                 return NULL;
945  
946 -       iova = __alloc_iova(domain, iovad->granule, dma_get_mask(dev));
947 -       if (!iova)
948 -               goto out_free_page;
949 -
950         msi_page->phys = msi_addr;
951 -       msi_page->iova = iova_dma_addr(iovad, iova);
952 -       if (iommu_map(domain, msi_page->iova, msi_addr, iovad->granule, prot))
953 +       if (iovad) {
954 +               iova = __alloc_iova(domain, size, dma_get_mask(dev));
955 +               if (!iova)
956 +                       goto out_free_page;
957 +               msi_page->iova = iova_dma_addr(iovad, iova);
958 +       } else {
959 +               msi_page->iova = cookie->msi_iova;
960 +               cookie->msi_iova += size;
961 +       }
962 +
963 +       if (iommu_map(domain, msi_page->iova, msi_addr, size, prot))
964                 goto out_free_iova;
965  
966         INIT_LIST_HEAD(&msi_page->list);
967 @@ -670,7 +833,10 @@ static struct iommu_dma_msi_page *iommu_
968         return msi_page;
969  
970  out_free_iova:
971 -       __free_iova(iovad, iova);
972 +       if (iovad)
973 +               __free_iova(iovad, iova);
974 +       else
975 +               cookie->msi_iova -= size;
976  out_free_page:
977         kfree(msi_page);
978         return NULL;
979 @@ -711,7 +877,7 @@ void iommu_dma_map_msi_msg(int irq, stru
980                 msg->data = ~0U;
981         } else {
982                 msg->address_hi = upper_32_bits(msi_page->iova);
983 -               msg->address_lo &= iova_mask(&cookie->iovad);
984 +               msg->address_lo &= cookie_msi_granule(cookie) - 1;
985                 msg->address_lo += lower_32_bits(msi_page->iova);
986         }
987  }
988 --- a/drivers/iommu/intel-iommu.c
989 +++ b/drivers/iommu/intel-iommu.c
990 @@ -440,6 +440,7 @@ struct dmar_rmrr_unit {
991         u64     end_address;            /* reserved end address */
992         struct dmar_dev_scope *devices; /* target devices */
993         int     devices_cnt;            /* target device count */
994 +       struct iommu_resv_region *resv; /* reserved region handle */
995  };
996  
997  struct dmar_atsr_unit {
998 @@ -4251,27 +4252,40 @@ static inline void init_iommu_pm_ops(voi
999  int __init dmar_parse_one_rmrr(struct acpi_dmar_header *header, void *arg)
1000  {
1001         struct acpi_dmar_reserved_memory *rmrr;
1002 +       int prot = DMA_PTE_READ|DMA_PTE_WRITE;
1003         struct dmar_rmrr_unit *rmrru;
1004 +       size_t length;
1005  
1006         rmrru = kzalloc(sizeof(*rmrru), GFP_KERNEL);
1007         if (!rmrru)
1008 -               return -ENOMEM;
1009 +               goto out;
1010  
1011         rmrru->hdr = header;
1012         rmrr = (struct acpi_dmar_reserved_memory *)header;
1013         rmrru->base_address = rmrr->base_address;
1014         rmrru->end_address = rmrr->end_address;
1015 +
1016 +       length = rmrr->end_address - rmrr->base_address + 1;
1017 +       rmrru->resv = iommu_alloc_resv_region(rmrr->base_address, length, prot,
1018 +                                             IOMMU_RESV_DIRECT);
1019 +       if (!rmrru->resv)
1020 +               goto free_rmrru;
1021 +
1022         rmrru->devices = dmar_alloc_dev_scope((void *)(rmrr + 1),
1023                                 ((void *)rmrr) + rmrr->header.length,
1024                                 &rmrru->devices_cnt);
1025 -       if (rmrru->devices_cnt && rmrru->devices == NULL) {
1026 -               kfree(rmrru);
1027 -               return -ENOMEM;
1028 -       }
1029 +       if (rmrru->devices_cnt && rmrru->devices == NULL)
1030 +               goto free_all;
1031  
1032         list_add(&rmrru->list, &dmar_rmrr_units);
1033  
1034         return 0;
1035 +free_all:
1036 +       kfree(rmrru->resv);
1037 +free_rmrru:
1038 +       kfree(rmrru);
1039 +out:
1040 +       return -ENOMEM;
1041  }
1042  
1043  static struct dmar_atsr_unit *dmar_find_atsr(struct acpi_dmar_atsr *atsr)
1044 @@ -4485,6 +4499,7 @@ static void intel_iommu_free_dmars(void)
1045         list_for_each_entry_safe(rmrru, rmrr_n, &dmar_rmrr_units, list) {
1046                 list_del(&rmrru->list);
1047                 dmar_free_dev_scope(&rmrru->devices, &rmrru->devices_cnt);
1048 +               kfree(rmrru->resv);
1049                 kfree(rmrru);
1050         }
1051  
1052 @@ -5220,6 +5235,45 @@ static void intel_iommu_remove_device(st
1053         iommu_device_unlink(iommu->iommu_dev, dev);
1054  }
1055  
1056 +static void intel_iommu_get_resv_regions(struct device *device,
1057 +                                        struct list_head *head)
1058 +{
1059 +       struct iommu_resv_region *reg;
1060 +       struct dmar_rmrr_unit *rmrr;
1061 +       struct device *i_dev;
1062 +       int i;
1063 +
1064 +       rcu_read_lock();
1065 +       for_each_rmrr_units(rmrr) {
1066 +               for_each_active_dev_scope(rmrr->devices, rmrr->devices_cnt,
1067 +                                         i, i_dev) {
1068 +                       if (i_dev != device)
1069 +                               continue;
1070 +
1071 +                       list_add_tail(&rmrr->resv->list, head);
1072 +               }
1073 +       }
1074 +       rcu_read_unlock();
1075 +
1076 +       reg = iommu_alloc_resv_region(IOAPIC_RANGE_START,
1077 +                                     IOAPIC_RANGE_END - IOAPIC_RANGE_START + 1,
1078 +                                     0, IOMMU_RESV_MSI);
1079 +       if (!reg)
1080 +               return;
1081 +       list_add_tail(&reg->list, head);
1082 +}
1083 +
1084 +static void intel_iommu_put_resv_regions(struct device *dev,
1085 +                                        struct list_head *head)
1086 +{
1087 +       struct iommu_resv_region *entry, *next;
1088 +
1089 +       list_for_each_entry_safe(entry, next, head, list) {
1090 +               if (entry->type == IOMMU_RESV_RESERVED)
1091 +                       kfree(entry);
1092 +       }
1093 +}
1094 +
1095  #ifdef CONFIG_INTEL_IOMMU_SVM
1096  #define MAX_NR_PASID_BITS (20)
1097  static inline unsigned long intel_iommu_get_pts(struct intel_iommu *iommu)
1098 @@ -5350,19 +5404,21 @@ struct intel_iommu *intel_svm_device_to_
1099  #endif /* CONFIG_INTEL_IOMMU_SVM */
1100  
1101  static const struct iommu_ops intel_iommu_ops = {
1102 -       .capable        = intel_iommu_capable,
1103 -       .domain_alloc   = intel_iommu_domain_alloc,
1104 -       .domain_free    = intel_iommu_domain_free,
1105 -       .attach_dev     = intel_iommu_attach_device,
1106 -       .detach_dev     = intel_iommu_detach_device,
1107 -       .map            = intel_iommu_map,
1108 -       .unmap          = intel_iommu_unmap,
1109 -       .map_sg         = default_iommu_map_sg,
1110 -       .iova_to_phys   = intel_iommu_iova_to_phys,
1111 -       .add_device     = intel_iommu_add_device,
1112 -       .remove_device  = intel_iommu_remove_device,
1113 -       .device_group   = pci_device_group,
1114 -       .pgsize_bitmap  = INTEL_IOMMU_PGSIZES,
1115 +       .capable                = intel_iommu_capable,
1116 +       .domain_alloc           = intel_iommu_domain_alloc,
1117 +       .domain_free            = intel_iommu_domain_free,
1118 +       .attach_dev             = intel_iommu_attach_device,
1119 +       .detach_dev             = intel_iommu_detach_device,
1120 +       .map                    = intel_iommu_map,
1121 +       .unmap                  = intel_iommu_unmap,
1122 +       .map_sg                 = default_iommu_map_sg,
1123 +       .iova_to_phys           = intel_iommu_iova_to_phys,
1124 +       .add_device             = intel_iommu_add_device,
1125 +       .remove_device          = intel_iommu_remove_device,
1126 +       .get_resv_regions       = intel_iommu_get_resv_regions,
1127 +       .put_resv_regions       = intel_iommu_put_resv_regions,
1128 +       .device_group           = pci_device_group,
1129 +       .pgsize_bitmap          = INTEL_IOMMU_PGSIZES,
1130  };
1131  
1132  static void quirk_iommu_g4x_gfx(struct pci_dev *dev)
1133 --- a/drivers/iommu/iommu.c
1134 +++ b/drivers/iommu/iommu.c
1135 @@ -36,6 +36,7 @@
1136  
1137  static struct kset *iommu_group_kset;
1138  static DEFINE_IDA(iommu_group_ida);
1139 +static unsigned int iommu_def_domain_type = IOMMU_DOMAIN_DMA;
1140  
1141  struct iommu_callback_data {
1142         const struct iommu_ops *ops;
1143 @@ -68,6 +69,13 @@ struct iommu_group_attribute {
1144                          const char *buf, size_t count);
1145  };
1146  
1147 +static const char * const iommu_group_resv_type_string[] = {
1148 +       [IOMMU_RESV_DIRECT]     = "direct",
1149 +       [IOMMU_RESV_RESERVED]   = "reserved",
1150 +       [IOMMU_RESV_MSI]        = "msi",
1151 +       [IOMMU_RESV_SW_MSI]     = "msi",
1152 +};
1153 +
1154  #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store)          \
1155  struct iommu_group_attribute iommu_group_attr_##_name =                \
1156         __ATTR(_name, _mode, _show, _store)
1157 @@ -86,6 +94,18 @@ static int __iommu_attach_group(struct i
1158  static void __iommu_detach_group(struct iommu_domain *domain,
1159                                  struct iommu_group *group);
1160  
1161 +static int __init iommu_set_def_domain_type(char *str)
1162 +{
1163 +       bool pt;
1164 +
1165 +       if (!str || strtobool(str, &pt))
1166 +               return -EINVAL;
1167 +
1168 +       iommu_def_domain_type = pt ? IOMMU_DOMAIN_IDENTITY : IOMMU_DOMAIN_DMA;
1169 +       return 0;
1170 +}
1171 +early_param("iommu.passthrough", iommu_set_def_domain_type);
1172 +
1173  static ssize_t iommu_group_attr_show(struct kobject *kobj,
1174                                      struct attribute *__attr, char *buf)
1175  {
1176 @@ -133,8 +153,131 @@ static ssize_t iommu_group_show_name(str
1177         return sprintf(buf, "%s\n", group->name);
1178  }
1179  
1180 +/**
1181 + * iommu_insert_resv_region - Insert a new region in the
1182 + * list of reserved regions.
1183 + * @new: new region to insert
1184 + * @regions: list of regions
1185 + *
1186 + * The new element is sorted by address with respect to the other
1187 + * regions of the same type. In case it overlaps with another
1188 + * region of the same type, regions are merged. In case it
1189 + * overlaps with another region of different type, regions are
1190 + * not merged.
1191 + */
1192 +static int iommu_insert_resv_region(struct iommu_resv_region *new,
1193 +                                   struct list_head *regions)
1194 +{
1195 +       struct iommu_resv_region *region;
1196 +       phys_addr_t start = new->start;
1197 +       phys_addr_t end = new->start + new->length - 1;
1198 +       struct list_head *pos = regions->next;
1199 +
1200 +       while (pos != regions) {
1201 +               struct iommu_resv_region *entry =
1202 +                       list_entry(pos, struct iommu_resv_region, list);
1203 +               phys_addr_t a = entry->start;
1204 +               phys_addr_t b = entry->start + entry->length - 1;
1205 +               int type = entry->type;
1206 +
1207 +               if (end < a) {
1208 +                       goto insert;
1209 +               } else if (start > b) {
1210 +                       pos = pos->next;
1211 +               } else if ((start >= a) && (end <= b)) {
1212 +                       if (new->type == type)
1213 +                               goto done;
1214 +                       else
1215 +                               pos = pos->next;
1216 +               } else {
1217 +                       if (new->type == type) {
1218 +                               phys_addr_t new_start = min(a, start);
1219 +                               phys_addr_t new_end = max(b, end);
1220 +
1221 +                               list_del(&entry->list);
1222 +                               entry->start = new_start;
1223 +                               entry->length = new_end - new_start + 1;
1224 +                               iommu_insert_resv_region(entry, regions);
1225 +                       } else {
1226 +                               pos = pos->next;
1227 +                       }
1228 +               }
1229 +       }
1230 +insert:
1231 +       region = iommu_alloc_resv_region(new->start, new->length,
1232 +                                        new->prot, new->type);
1233 +       if (!region)
1234 +               return -ENOMEM;
1235 +
1236 +       list_add_tail(&region->list, pos);
1237 +done:
1238 +       return 0;
1239 +}
1240 +
1241 +static int
1242 +iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
1243 +                                struct list_head *group_resv_regions)
1244 +{
1245 +       struct iommu_resv_region *entry;
1246 +       int ret;
1247 +
1248 +       list_for_each_entry(entry, dev_resv_regions, list) {
1249 +               ret = iommu_insert_resv_region(entry, group_resv_regions);
1250 +               if (ret)
1251 +                       break;
1252 +       }
1253 +       return ret;
1254 +}
1255 +
1256 +int iommu_get_group_resv_regions(struct iommu_group *group,
1257 +                                struct list_head *head)
1258 +{
1259 +       struct iommu_device *device;
1260 +       int ret = 0;
1261 +
1262 +       mutex_lock(&group->mutex);
1263 +       list_for_each_entry(device, &group->devices, list) {
1264 +               struct list_head dev_resv_regions;
1265 +
1266 +               INIT_LIST_HEAD(&dev_resv_regions);
1267 +               iommu_get_resv_regions(device->dev, &dev_resv_regions);
1268 +               ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
1269 +               iommu_put_resv_regions(device->dev, &dev_resv_regions);
1270 +               if (ret)
1271 +                       break;
1272 +       }
1273 +       mutex_unlock(&group->mutex);
1274 +       return ret;
1275 +}
1276 +EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
1277 +
1278 +static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
1279 +                                            char *buf)
1280 +{
1281 +       struct iommu_resv_region *region, *next;
1282 +       struct list_head group_resv_regions;
1283 +       char *str = buf;
1284 +
1285 +       INIT_LIST_HEAD(&group_resv_regions);
1286 +       iommu_get_group_resv_regions(group, &group_resv_regions);
1287 +
1288 +       list_for_each_entry_safe(region, next, &group_resv_regions, list) {
1289 +               str += sprintf(str, "0x%016llx 0x%016llx %s\n",
1290 +                              (long long int)region->start,
1291 +                              (long long int)(region->start +
1292 +                                               region->length - 1),
1293 +                              iommu_group_resv_type_string[region->type]);
1294 +               kfree(region);
1295 +       }
1296 +
1297 +       return (str - buf);
1298 +}
1299 +
1300  static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
1301  
1302 +static IOMMU_GROUP_ATTR(reserved_regions, 0444,
1303 +                       iommu_group_show_resv_regions, NULL);
1304 +
1305  static void iommu_group_release(struct kobject *kobj)
1306  {
1307         struct iommu_group *group = to_iommu_group(kobj);
1308 @@ -212,6 +355,11 @@ struct iommu_group *iommu_group_alloc(vo
1309          */
1310         kobject_put(&group->kobj);
1311  
1312 +       ret = iommu_group_create_file(group,
1313 +                                     &iommu_group_attr_reserved_regions);
1314 +       if (ret)
1315 +               return ERR_PTR(ret);
1316 +
1317         pr_debug("Allocated group %d\n", group->id);
1318  
1319         return group;
1320 @@ -318,7 +466,7 @@ static int iommu_group_create_direct_map
1321                                               struct device *dev)
1322  {
1323         struct iommu_domain *domain = group->default_domain;
1324 -       struct iommu_dm_region *entry;
1325 +       struct iommu_resv_region *entry;
1326         struct list_head mappings;
1327         unsigned long pg_size;
1328         int ret = 0;
1329 @@ -331,18 +479,21 @@ static int iommu_group_create_direct_map
1330         pg_size = 1UL << __ffs(domain->pgsize_bitmap);
1331         INIT_LIST_HEAD(&mappings);
1332  
1333 -       iommu_get_dm_regions(dev, &mappings);
1334 +       iommu_get_resv_regions(dev, &mappings);
1335  
1336         /* We need to consider overlapping regions for different devices */
1337         list_for_each_entry(entry, &mappings, list) {
1338                 dma_addr_t start, end, addr;
1339  
1340 -               if (domain->ops->apply_dm_region)
1341 -                       domain->ops->apply_dm_region(dev, domain, entry);
1342 +               if (domain->ops->apply_resv_region)
1343 +                       domain->ops->apply_resv_region(dev, domain, entry);
1344  
1345                 start = ALIGN(entry->start, pg_size);
1346                 end   = ALIGN(entry->start + entry->length, pg_size);
1347  
1348 +               if (entry->type != IOMMU_RESV_DIRECT)
1349 +                       continue;
1350 +
1351                 for (addr = start; addr < end; addr += pg_size) {
1352                         phys_addr_t phys_addr;
1353  
1354 @@ -358,7 +509,7 @@ static int iommu_group_create_direct_map
1355         }
1356  
1357  out:
1358 -       iommu_put_dm_regions(dev, &mappings);
1359 +       iommu_put_resv_regions(dev, &mappings);
1360  
1361         return ret;
1362  }
1363 @@ -563,6 +714,19 @@ struct iommu_group *iommu_group_get(stru
1364  EXPORT_SYMBOL_GPL(iommu_group_get);
1365  
1366  /**
1367 + * iommu_group_ref_get - Increment reference on a group
1368 + * @group: the group to use, must not be NULL
1369 + *
1370 + * This function is called by iommu drivers to take additional references on an
1371 + * existing group.  Returns the given group for convenience.
1372 + */
1373 +struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
1374 +{
1375 +       kobject_get(group->devices_kobj);
1376 +       return group;
1377 +}
1378 +
1379 +/**
1380   * iommu_group_put - Decrement group reference
1381   * @group: the group to use
1382   *
1383 @@ -845,10 +1009,19 @@ struct iommu_group *iommu_group_get_for_
1384          * IOMMU driver.
1385          */
1386         if (!group->default_domain) {
1387 -               group->default_domain = __iommu_domain_alloc(dev->bus,
1388 -                                                            IOMMU_DOMAIN_DMA);
1389 +               struct iommu_domain *dom;
1390 +
1391 +               dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type);
1392 +               if (!dom && iommu_def_domain_type != IOMMU_DOMAIN_DMA) {
1393 +                       dev_warn(dev,
1394 +                                "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA",
1395 +                                iommu_def_domain_type);
1396 +                       dom = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_DMA);
1397 +               }
1398 +
1399 +               group->default_domain = dom;
1400                 if (!group->domain)
1401 -                       group->domain = group->default_domain;
1402 +                       group->domain = dom;
1403         }
1404  
1405         ret = iommu_group_add_device(group, dev);
1406 @@ -1557,20 +1730,38 @@ int iommu_domain_set_attr(struct iommu_d
1407  }
1408  EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
1409  
1410 -void iommu_get_dm_regions(struct device *dev, struct list_head *list)
1411 +void iommu_get_resv_regions(struct device *dev, struct list_head *list)
1412  {
1413         const struct iommu_ops *ops = dev->bus->iommu_ops;
1414  
1415 -       if (ops && ops->get_dm_regions)
1416 -               ops->get_dm_regions(dev, list);
1417 +       if (ops && ops->get_resv_regions)
1418 +               ops->get_resv_regions(dev, list);
1419  }
1420  
1421 -void iommu_put_dm_regions(struct device *dev, struct list_head *list)
1422 +void iommu_put_resv_regions(struct device *dev, struct list_head *list)
1423  {
1424         const struct iommu_ops *ops = dev->bus->iommu_ops;
1425  
1426 -       if (ops && ops->put_dm_regions)
1427 -               ops->put_dm_regions(dev, list);
1428 +       if (ops && ops->put_resv_regions)
1429 +               ops->put_resv_regions(dev, list);
1430 +}
1431 +
1432 +struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
1433 +                                                 size_t length, int prot,
1434 +                                                 enum iommu_resv_type type)
1435 +{
1436 +       struct iommu_resv_region *region;
1437 +
1438 +       region = kzalloc(sizeof(*region), GFP_KERNEL);
1439 +       if (!region)
1440 +               return NULL;
1441 +
1442 +       INIT_LIST_HEAD(&region->list);
1443 +       region->start = start;
1444 +       region->length = length;
1445 +       region->prot = prot;
1446 +       region->type = type;
1447 +       return region;
1448  }
1449  
1450  /* Request that a device is direct mapped by the IOMMU */
1451 --- a/drivers/iommu/mtk_iommu.c
1452 +++ b/drivers/iommu/mtk_iommu.c
1453 @@ -410,6 +410,8 @@ static struct iommu_group *mtk_iommu_dev
1454                 data->m4u_group = iommu_group_alloc();
1455                 if (IS_ERR(data->m4u_group))
1456                         dev_err(dev, "Failed to allocate M4U IOMMU group\n");
1457 +       } else {
1458 +               iommu_group_ref_get(data->m4u_group);
1459         }
1460         return data->m4u_group;
1461  }
1462 --- a/drivers/iommu/mtk_iommu_v1.c
1463 +++ b/drivers/iommu/mtk_iommu_v1.c
1464 @@ -502,6 +502,8 @@ static struct iommu_group *mtk_iommu_dev
1465                 data->m4u_group = iommu_group_alloc();
1466                 if (IS_ERR(data->m4u_group))
1467                         dev_err(dev, "Failed to allocate M4U IOMMU group\n");
1468 +       } else {
1469 +               iommu_group_ref_get(data->m4u_group);
1470         }
1471         return data->m4u_group;
1472  }
1473 --- a/include/linux/dma-iommu.h
1474 +++ b/include/linux/dma-iommu.h
1475 @@ -27,6 +27,7 @@ int iommu_dma_init(void);
1476  
1477  /* Domain management interface for IOMMU drivers */
1478  int iommu_get_dma_cookie(struct iommu_domain *domain);
1479 +int iommu_get_msi_cookie(struct iommu_domain *domain, dma_addr_t base);
1480  void iommu_put_dma_cookie(struct iommu_domain *domain);
1481  
1482  /* Setup call for arch DMA mapping code */
1483 @@ -66,6 +67,7 @@ int iommu_dma_mapping_error(struct devic
1484  
1485  /* The DMA API isn't _quite_ the whole story, though... */
1486  void iommu_dma_map_msi_msg(int irq, struct msi_msg *msg);
1487 +void iommu_dma_get_resv_regions(struct device *dev, struct list_head *list);
1488  
1489  #else
1490  
1491 @@ -82,6 +84,11 @@ static inline int iommu_get_dma_cookie(s
1492         return -ENODEV;
1493  }
1494  
1495 +static inline int iommu_get_msi_cookie(struct iommu_domain *domain, dma_addr_t base)
1496 +{
1497 +       return -ENODEV;
1498 +}
1499 +
1500  static inline void iommu_put_dma_cookie(struct iommu_domain *domain)
1501  {
1502  }
1503 @@ -90,6 +97,10 @@ static inline void iommu_dma_map_msi_msg
1504  {
1505  }
1506  
1507 +static inline void iommu_dma_get_resv_regions(struct device *dev, struct list_head *list)
1508 +{
1509 +}
1510 +
1511  #endif /* CONFIG_IOMMU_DMA */
1512  #endif /* __KERNEL__ */
1513  #endif /* __DMA_IOMMU_H */
1514 --- a/include/linux/iommu.h
1515 +++ b/include/linux/iommu.h
1516 @@ -117,18 +117,32 @@ enum iommu_attr {
1517         DOMAIN_ATTR_MAX,
1518  };
1519  
1520 +/* These are the possible reserved region types */
1521 +enum iommu_resv_type {
1522 +       /* Memory regions which must be mapped 1:1 at all times */
1523 +       IOMMU_RESV_DIRECT,
1524 +       /* Arbitrary "never map this or give it to a device" address ranges */
1525 +       IOMMU_RESV_RESERVED,
1526 +       /* Hardware MSI region (untranslated) */
1527 +       IOMMU_RESV_MSI,
1528 +       /* Software-managed MSI translation window */
1529 +       IOMMU_RESV_SW_MSI,
1530 +};
1531 +
1532  /**
1533 - * struct iommu_dm_region - descriptor for a direct mapped memory region
1534 + * struct iommu_resv_region - descriptor for a reserved memory region
1535   * @list: Linked list pointers
1536   * @start: System physical start address of the region
1537   * @length: Length of the region in bytes
1538   * @prot: IOMMU Protection flags (READ/WRITE/...)
1539 + * @type: Type of the reserved region
1540   */
1541 -struct iommu_dm_region {
1542 +struct iommu_resv_region {
1543         struct list_head        list;
1544         phys_addr_t             start;
1545         size_t                  length;
1546         int                     prot;
1547 +       enum iommu_resv_type    type;
1548  };
1549  
1550  #ifdef CONFIG_IOMMU_API
1551 @@ -150,9 +164,9 @@ struct iommu_dm_region {
1552   * @device_group: find iommu group for a particular device
1553   * @domain_get_attr: Query domain attributes
1554   * @domain_set_attr: Change domain attributes
1555 - * @get_dm_regions: Request list of direct mapping requirements for a device
1556 - * @put_dm_regions: Free list of direct mapping requirements for a device
1557 - * @apply_dm_region: Temporary helper call-back for iova reserved ranges
1558 + * @get_resv_regions: Request list of reserved regions for a device
1559 + * @put_resv_regions: Free list of reserved regions for a device
1560 + * @apply_resv_region: Temporary helper call-back for iova reserved ranges
1561   * @domain_window_enable: Configure and enable a particular window for a domain
1562   * @domain_window_disable: Disable a particular window for a domain
1563   * @domain_set_windows: Set the number of windows for a domain
1564 @@ -184,11 +198,12 @@ struct iommu_ops {
1565         int (*domain_set_attr)(struct iommu_domain *domain,
1566                                enum iommu_attr attr, void *data);
1567  
1568 -       /* Request/Free a list of direct mapping requirements for a device */
1569 -       void (*get_dm_regions)(struct device *dev, struct list_head *list);
1570 -       void (*put_dm_regions)(struct device *dev, struct list_head *list);
1571 -       void (*apply_dm_region)(struct device *dev, struct iommu_domain *domain,
1572 -                               struct iommu_dm_region *region);
1573 +       /* Request/Free a list of reserved regions for a device */
1574 +       void (*get_resv_regions)(struct device *dev, struct list_head *list);
1575 +       void (*put_resv_regions)(struct device *dev, struct list_head *list);
1576 +       void (*apply_resv_region)(struct device *dev,
1577 +                                 struct iommu_domain *domain,
1578 +                                 struct iommu_resv_region *region);
1579  
1580         /* Window handling functions */
1581         int (*domain_window_enable)(struct iommu_domain *domain, u32 wnd_nr,
1582 @@ -233,9 +248,14 @@ extern phys_addr_t iommu_iova_to_phys(st
1583  extern void iommu_set_fault_handler(struct iommu_domain *domain,
1584                         iommu_fault_handler_t handler, void *token);
1585  
1586 -extern void iommu_get_dm_regions(struct device *dev, struct list_head *list);
1587 -extern void iommu_put_dm_regions(struct device *dev, struct list_head *list);
1588 +extern void iommu_get_resv_regions(struct device *dev, struct list_head *list);
1589 +extern void iommu_put_resv_regions(struct device *dev, struct list_head *list);
1590  extern int iommu_request_dm_for_dev(struct device *dev);
1591 +extern struct iommu_resv_region *
1592 +iommu_alloc_resv_region(phys_addr_t start, size_t length, int prot,
1593 +                       enum iommu_resv_type type);
1594 +extern int iommu_get_group_resv_regions(struct iommu_group *group,
1595 +                                       struct list_head *head);
1596  
1597  extern int iommu_attach_group(struct iommu_domain *domain,
1598                               struct iommu_group *group);
1599 @@ -253,6 +273,7 @@ extern void iommu_group_remove_device(st
1600  extern int iommu_group_for_each_dev(struct iommu_group *group, void *data,
1601                                     int (*fn)(struct device *, void *));
1602  extern struct iommu_group *iommu_group_get(struct device *dev);
1603 +extern struct iommu_group *iommu_group_ref_get(struct iommu_group *group);
1604  extern void iommu_group_put(struct iommu_group *group);
1605  extern int iommu_group_register_notifier(struct iommu_group *group,
1606                                          struct notifier_block *nb);
1607 @@ -439,16 +460,22 @@ static inline void iommu_set_fault_handl
1608  {
1609  }
1610  
1611 -static inline void iommu_get_dm_regions(struct device *dev,
1612 +static inline void iommu_get_resv_regions(struct device *dev,
1613                                         struct list_head *list)
1614  {
1615  }
1616  
1617 -static inline void iommu_put_dm_regions(struct device *dev,
1618 +static inline void iommu_put_resv_regions(struct device *dev,
1619                                         struct list_head *list)
1620  {
1621  }
1622  
1623 +static inline int iommu_get_group_resv_regions(struct iommu_group *group,
1624 +                                              struct list_head *head)
1625 +{
1626 +       return -ENODEV;
1627 +}
1628 +
1629  static inline int iommu_request_dm_for_dev(struct device *dev)
1630  {
1631         return -ENODEV;