~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/dma/swiotlb.c

Version: ~ [ linux-6.11.5 ] ~ [ linux-6.10.14 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.58 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.114 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.169 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.228 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.284 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.322 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-only
  2 /*
  3  * Dynamic DMA mapping support.
  4  *
  5  * This implementation is a fallback for platforms that do not support
  6  * I/O TLBs (aka DMA address translation hardware).
  7  * Copyright (C) 2000 Asit Mallick <Asit.K.Mallick@intel.com>
  8  * Copyright (C) 2000 Goutham Rao <goutham.rao@intel.com>
  9  * Copyright (C) 2000, 2003 Hewlett-Packard Co
 10  *      David Mosberger-Tang <davidm@hpl.hp.com>
 11  *
 12  * 03/05/07 davidm      Switch from PCI-DMA to generic device DMA API.
 13  * 00/12/13 davidm      Rename to swiotlb.c and add mark_clean() to avoid
 14  *                      unnecessary i-cache flushing.
 15  * 04/07/.. ak          Better overflow handling. Assorted fixes.
 16  * 05/09/10 linville    Add support for syncing ranges, support syncing for
 17  *                      DMA_BIDIRECTIONAL mappings, miscellaneous cleanup.
 18  * 08/12/11 beckyb      Add highmem support
 19  */
 20 
 21 #define pr_fmt(fmt) "software IO TLB: " fmt
 22 
 23 #include <linux/cache.h>
 24 #include <linux/cc_platform.h>
 25 #include <linux/ctype.h>
 26 #include <linux/debugfs.h>
 27 #include <linux/dma-direct.h>
 28 #include <linux/dma-map-ops.h>
 29 #include <linux/export.h>
 30 #include <linux/gfp.h>
 31 #include <linux/highmem.h>
 32 #include <linux/io.h>
 33 #include <linux/iommu-helper.h>
 34 #include <linux/init.h>
 35 #include <linux/memblock.h>
 36 #include <linux/mm.h>
 37 #include <linux/pfn.h>
 38 #include <linux/rculist.h>
 39 #include <linux/scatterlist.h>
 40 #include <linux/set_memory.h>
 41 #include <linux/spinlock.h>
 42 #include <linux/string.h>
 43 #include <linux/swiotlb.h>
 44 #include <linux/types.h>
 45 #ifdef CONFIG_DMA_RESTRICTED_POOL
 46 #include <linux/of.h>
 47 #include <linux/of_fdt.h>
 48 #include <linux/of_reserved_mem.h>
 49 #include <linux/slab.h>
 50 #endif
 51 
 52 #define CREATE_TRACE_POINTS
 53 #include <trace/events/swiotlb.h>
 54 
 55 #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT))
 56 
 57 /*
 58  * Minimum IO TLB size to bother booting with.  Systems with mainly
 59  * 64bit capable cards will only lightly use the swiotlb.  If we can't
 60  * allocate a contiguous 1MB, we're probably in trouble anyway.
 61  */
 62 #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT)
 63 
 64 #define INVALID_PHYS_ADDR (~(phys_addr_t)0)
 65 
 66 /**
 67  * struct io_tlb_slot - IO TLB slot descriptor
 68  * @orig_addr:  The original address corresponding to a mapped entry.
 69  * @alloc_size: Size of the allocated buffer.
 70  * @list:       The free list describing the number of free entries available
 71  *              from each index.
 72  * @pad_slots:  Number of preceding padding slots. Valid only in the first
 73  *              allocated non-padding slot.
 74  */
 75 struct io_tlb_slot {
 76         phys_addr_t orig_addr;
 77         size_t alloc_size;
 78         unsigned short list;
 79         unsigned short pad_slots;
 80 };
 81 
 82 static bool swiotlb_force_bounce;
 83 static bool swiotlb_force_disable;
 84 
 85 #ifdef CONFIG_SWIOTLB_DYNAMIC
 86 
 87 static void swiotlb_dyn_alloc(struct work_struct *work);
 88 
 89 static struct io_tlb_mem io_tlb_default_mem = {
 90         .lock = __SPIN_LOCK_UNLOCKED(io_tlb_default_mem.lock),
 91         .pools = LIST_HEAD_INIT(io_tlb_default_mem.pools),
 92         .dyn_alloc = __WORK_INITIALIZER(io_tlb_default_mem.dyn_alloc,
 93                                         swiotlb_dyn_alloc),
 94 };
 95 
 96 #else  /* !CONFIG_SWIOTLB_DYNAMIC */
 97 
 98 static struct io_tlb_mem io_tlb_default_mem;
 99 
100 #endif  /* CONFIG_SWIOTLB_DYNAMIC */
101 
102 static unsigned long default_nslabs = IO_TLB_DEFAULT_SIZE >> IO_TLB_SHIFT;
103 static unsigned long default_nareas;
104 
105 /**
106  * struct io_tlb_area - IO TLB memory area descriptor
107  *
108  * This is a single area with a single lock.
109  *
110  * @used:       The number of used IO TLB block.
111  * @index:      The slot index to start searching in this area for next round.
112  * @lock:       The lock to protect the above data structures in the map and
113  *              unmap calls.
114  */
115 struct io_tlb_area {
116         unsigned long used;
117         unsigned int index;
118         spinlock_t lock;
119 };
120 
121 /*
122  * Round up number of slabs to the next power of 2. The last area is going
123  * be smaller than the rest if default_nslabs is not power of two.
124  * The number of slot in an area should be a multiple of IO_TLB_SEGSIZE,
125  * otherwise a segment may span two or more areas. It conflicts with free
126  * contiguous slots tracking: free slots are treated contiguous no matter
127  * whether they cross an area boundary.
128  *
129  * Return true if default_nslabs is rounded up.
130  */
131 static bool round_up_default_nslabs(void)
132 {
133         if (!default_nareas)
134                 return false;
135 
136         if (default_nslabs < IO_TLB_SEGSIZE * default_nareas)
137                 default_nslabs = IO_TLB_SEGSIZE * default_nareas;
138         else if (is_power_of_2(default_nslabs))
139                 return false;
140         default_nslabs = roundup_pow_of_two(default_nslabs);
141         return true;
142 }
143 
144 /**
145  * swiotlb_adjust_nareas() - adjust the number of areas and slots
146  * @nareas:     Desired number of areas. Zero is treated as 1.
147  *
148  * Adjust the default number of areas in a memory pool.
149  * The default size of the memory pool may also change to meet minimum area
150  * size requirements.
151  */
152 static void swiotlb_adjust_nareas(unsigned int nareas)
153 {
154         if (!nareas)
155                 nareas = 1;
156         else if (!is_power_of_2(nareas))
157                 nareas = roundup_pow_of_two(nareas);
158 
159         default_nareas = nareas;
160 
161         pr_info("area num %d.\n", nareas);
162         if (round_up_default_nslabs())
163                 pr_info("SWIOTLB bounce buffer size roundup to %luMB",
164                         (default_nslabs << IO_TLB_SHIFT) >> 20);
165 }
166 
167 /**
168  * limit_nareas() - get the maximum number of areas for a given memory pool size
169  * @nareas:     Desired number of areas.
170  * @nslots:     Total number of slots in the memory pool.
171  *
172  * Limit the number of areas to the maximum possible number of areas in
173  * a memory pool of the given size.
174  *
175  * Return: Maximum possible number of areas.
176  */
177 static unsigned int limit_nareas(unsigned int nareas, unsigned long nslots)
178 {
179         if (nslots < nareas * IO_TLB_SEGSIZE)
180                 return nslots / IO_TLB_SEGSIZE;
181         return nareas;
182 }
183 
184 static int __init
185 setup_io_tlb_npages(char *str)
186 {
187         if (isdigit(*str)) {
188                 /* avoid tail segment of size < IO_TLB_SEGSIZE */
189                 default_nslabs =
190                         ALIGN(simple_strtoul(str, &str, 0), IO_TLB_SEGSIZE);
191         }
192         if (*str == ',')
193                 ++str;
194         if (isdigit(*str))
195                 swiotlb_adjust_nareas(simple_strtoul(str, &str, 0));
196         if (*str == ',')
197                 ++str;
198         if (!strcmp(str, "force"))
199                 swiotlb_force_bounce = true;
200         else if (!strcmp(str, "noforce"))
201                 swiotlb_force_disable = true;
202 
203         return 0;
204 }
205 early_param("swiotlb", setup_io_tlb_npages);
206 
207 unsigned long swiotlb_size_or_default(void)
208 {
209         return default_nslabs << IO_TLB_SHIFT;
210 }
211 
212 void __init swiotlb_adjust_size(unsigned long size)
213 {
214         /*
215          * If swiotlb parameter has not been specified, give a chance to
216          * architectures such as those supporting memory encryption to
217          * adjust/expand SWIOTLB size for their use.
218          */
219         if (default_nslabs != IO_TLB_DEFAULT_SIZE >> IO_TLB_SHIFT)
220                 return;
221 
222         size = ALIGN(size, IO_TLB_SIZE);
223         default_nslabs = ALIGN(size >> IO_TLB_SHIFT, IO_TLB_SEGSIZE);
224         if (round_up_default_nslabs())
225                 size = default_nslabs << IO_TLB_SHIFT;
226         pr_info("SWIOTLB bounce buffer size adjusted to %luMB", size >> 20);
227 }
228 
229 void swiotlb_print_info(void)
230 {
231         struct io_tlb_pool *mem = &io_tlb_default_mem.defpool;
232 
233         if (!mem->nslabs) {
234                 pr_warn("No low mem\n");
235                 return;
236         }
237 
238         pr_info("mapped [mem %pa-%pa] (%luMB)\n", &mem->start, &mem->end,
239                (mem->nslabs << IO_TLB_SHIFT) >> 20);
240 }
241 
242 static inline unsigned long io_tlb_offset(unsigned long val)
243 {
244         return val & (IO_TLB_SEGSIZE - 1);
245 }
246 
247 static inline unsigned long nr_slots(u64 val)
248 {
249         return DIV_ROUND_UP(val, IO_TLB_SIZE);
250 }
251 
252 /*
253  * Early SWIOTLB allocation may be too early to allow an architecture to
254  * perform the desired operations.  This function allows the architecture to
255  * call SWIOTLB when the operations are possible.  It needs to be called
256  * before the SWIOTLB memory is used.
257  */
258 void __init swiotlb_update_mem_attributes(void)
259 {
260         struct io_tlb_pool *mem = &io_tlb_default_mem.defpool;
261         unsigned long bytes;
262 
263         if (!mem->nslabs || mem->late_alloc)
264                 return;
265         bytes = PAGE_ALIGN(mem->nslabs << IO_TLB_SHIFT);
266         set_memory_decrypted((unsigned long)mem->vaddr, bytes >> PAGE_SHIFT);
267 }
268 
269 static void swiotlb_init_io_tlb_pool(struct io_tlb_pool *mem, phys_addr_t start,
270                 unsigned long nslabs, bool late_alloc, unsigned int nareas)
271 {
272         void *vaddr = phys_to_virt(start);
273         unsigned long bytes = nslabs << IO_TLB_SHIFT, i;
274 
275         mem->nslabs = nslabs;
276         mem->start = start;
277         mem->end = mem->start + bytes;
278         mem->late_alloc = late_alloc;
279         mem->nareas = nareas;
280         mem->area_nslabs = nslabs / mem->nareas;
281 
282         for (i = 0; i < mem->nareas; i++) {
283                 spin_lock_init(&mem->areas[i].lock);
284                 mem->areas[i].index = 0;
285                 mem->areas[i].used = 0;
286         }
287 
288         for (i = 0; i < mem->nslabs; i++) {
289                 mem->slots[i].list = min(IO_TLB_SEGSIZE - io_tlb_offset(i),
290                                          mem->nslabs - i);
291                 mem->slots[i].orig_addr = INVALID_PHYS_ADDR;
292                 mem->slots[i].alloc_size = 0;
293                 mem->slots[i].pad_slots = 0;
294         }
295 
296         memset(vaddr, 0, bytes);
297         mem->vaddr = vaddr;
298         return;
299 }
300 
301 /**
302  * add_mem_pool() - add a memory pool to the allocator
303  * @mem:        Software IO TLB allocator.
304  * @pool:       Memory pool to be added.
305  */
306 static void add_mem_pool(struct io_tlb_mem *mem, struct io_tlb_pool *pool)
307 {
308 #ifdef CONFIG_SWIOTLB_DYNAMIC
309         spin_lock(&mem->lock);
310         list_add_rcu(&pool->node, &mem->pools);
311         mem->nslabs += pool->nslabs;
312         spin_unlock(&mem->lock);
313 #else
314         mem->nslabs = pool->nslabs;
315 #endif
316 }
317 
318 static void __init *swiotlb_memblock_alloc(unsigned long nslabs,
319                 unsigned int flags,
320                 int (*remap)(void *tlb, unsigned long nslabs))
321 {
322         size_t bytes = PAGE_ALIGN(nslabs << IO_TLB_SHIFT);
323         void *tlb;
324 
325         /*
326          * By default allocate the bounce buffer memory from low memory, but
327          * allow to pick a location everywhere for hypervisors with guest
328          * memory encryption.
329          */
330         if (flags & SWIOTLB_ANY)
331                 tlb = memblock_alloc(bytes, PAGE_SIZE);
332         else
333                 tlb = memblock_alloc_low(bytes, PAGE_SIZE);
334 
335         if (!tlb) {
336                 pr_warn("%s: Failed to allocate %zu bytes tlb structure\n",
337                         __func__, bytes);
338                 return NULL;
339         }
340 
341         if (remap && remap(tlb, nslabs) < 0) {
342                 memblock_free(tlb, PAGE_ALIGN(bytes));
343                 pr_warn("%s: Failed to remap %zu bytes\n", __func__, bytes);
344                 return NULL;
345         }
346 
347         return tlb;
348 }
349 
350 /*
351  * Statically reserve bounce buffer space and initialize bounce buffer data
352  * structures for the software IO TLB used to implement the DMA API.
353  */
354 void __init swiotlb_init_remap(bool addressing_limit, unsigned int flags,
355                 int (*remap)(void *tlb, unsigned long nslabs))
356 {
357         struct io_tlb_pool *mem = &io_tlb_default_mem.defpool;
358         unsigned long nslabs;
359         unsigned int nareas;
360         size_t alloc_size;
361         void *tlb;
362 
363         if (!addressing_limit && !swiotlb_force_bounce)
364                 return;
365         if (swiotlb_force_disable)
366                 return;
367 
368         io_tlb_default_mem.force_bounce =
369                 swiotlb_force_bounce || (flags & SWIOTLB_FORCE);
370 
371 #ifdef CONFIG_SWIOTLB_DYNAMIC
372         if (!remap)
373                 io_tlb_default_mem.can_grow = true;
374         if (flags & SWIOTLB_ANY)
375                 io_tlb_default_mem.phys_limit = virt_to_phys(high_memory - 1);
376         else
377                 io_tlb_default_mem.phys_limit = ARCH_LOW_ADDRESS_LIMIT;
378 #endif
379 
380         if (!default_nareas)
381                 swiotlb_adjust_nareas(num_possible_cpus());
382 
383         nslabs = default_nslabs;
384         nareas = limit_nareas(default_nareas, nslabs);
385         while ((tlb = swiotlb_memblock_alloc(nslabs, flags, remap)) == NULL) {
386                 if (nslabs <= IO_TLB_MIN_SLABS)
387                         return;
388                 nslabs = ALIGN(nslabs >> 1, IO_TLB_SEGSIZE);
389                 nareas = limit_nareas(nareas, nslabs);
390         }
391 
392         if (default_nslabs != nslabs) {
393                 pr_info("SWIOTLB bounce buffer size adjusted %lu -> %lu slabs",
394                         default_nslabs, nslabs);
395                 default_nslabs = nslabs;
396         }
397 
398         alloc_size = PAGE_ALIGN(array_size(sizeof(*mem->slots), nslabs));
399         mem->slots = memblock_alloc(alloc_size, PAGE_SIZE);
400         if (!mem->slots) {
401                 pr_warn("%s: Failed to allocate %zu bytes align=0x%lx\n",
402                         __func__, alloc_size, PAGE_SIZE);
403                 return;
404         }
405 
406         mem->areas = memblock_alloc(array_size(sizeof(struct io_tlb_area),
407                 nareas), SMP_CACHE_BYTES);
408         if (!mem->areas) {
409                 pr_warn("%s: Failed to allocate mem->areas.\n", __func__);
410                 return;
411         }
412 
413         swiotlb_init_io_tlb_pool(mem, __pa(tlb), nslabs, false, nareas);
414         add_mem_pool(&io_tlb_default_mem, mem);
415 
416         if (flags & SWIOTLB_VERBOSE)
417                 swiotlb_print_info();
418 }
419 
420 void __init swiotlb_init(bool addressing_limit, unsigned int flags)
421 {
422         swiotlb_init_remap(addressing_limit, flags, NULL);
423 }
424 
425 /*
426  * Systems with larger DMA zones (those that don't support ISA) can
427  * initialize the swiotlb later using the slab allocator if needed.
428  * This should be just like above, but with some error catching.
429  */
430 int swiotlb_init_late(size_t size, gfp_t gfp_mask,
431                 int (*remap)(void *tlb, unsigned long nslabs))
432 {
433         struct io_tlb_pool *mem = &io_tlb_default_mem.defpool;
434         unsigned long nslabs = ALIGN(size >> IO_TLB_SHIFT, IO_TLB_SEGSIZE);
435         unsigned int nareas;
436         unsigned char *vstart = NULL;
437         unsigned int order, area_order;
438         bool retried = false;
439         int rc = 0;
440 
441         if (io_tlb_default_mem.nslabs)
442                 return 0;
443 
444         if (swiotlb_force_disable)
445                 return 0;
446 
447         io_tlb_default_mem.force_bounce = swiotlb_force_bounce;
448 
449 #ifdef CONFIG_SWIOTLB_DYNAMIC
450         if (!remap)
451                 io_tlb_default_mem.can_grow = true;
452         if (IS_ENABLED(CONFIG_ZONE_DMA) && (gfp_mask & __GFP_DMA))
453                 io_tlb_default_mem.phys_limit = DMA_BIT_MASK(zone_dma_bits);
454         else if (IS_ENABLED(CONFIG_ZONE_DMA32) && (gfp_mask & __GFP_DMA32))
455                 io_tlb_default_mem.phys_limit = DMA_BIT_MASK(32);
456         else
457                 io_tlb_default_mem.phys_limit = virt_to_phys(high_memory - 1);
458 #endif
459 
460         if (!default_nareas)
461                 swiotlb_adjust_nareas(num_possible_cpus());
462 
463 retry:
464         order = get_order(nslabs << IO_TLB_SHIFT);
465         nslabs = SLABS_PER_PAGE << order;
466 
467         while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) {
468                 vstart = (void *)__get_free_pages(gfp_mask | __GFP_NOWARN,
469                                                   order);
470                 if (vstart)
471                         break;
472                 order--;
473                 nslabs = SLABS_PER_PAGE << order;
474                 retried = true;
475         }
476 
477         if (!vstart)
478                 return -ENOMEM;
479 
480         if (remap)
481                 rc = remap(vstart, nslabs);
482         if (rc) {
483                 free_pages((unsigned long)vstart, order);
484 
485                 nslabs = ALIGN(nslabs >> 1, IO_TLB_SEGSIZE);
486                 if (nslabs < IO_TLB_MIN_SLABS)
487                         return rc;
488                 retried = true;
489                 goto retry;
490         }
491 
492         if (retried) {
493                 pr_warn("only able to allocate %ld MB\n",
494                         (PAGE_SIZE << order) >> 20);
495         }
496 
497         nareas = limit_nareas(default_nareas, nslabs);
498         area_order = get_order(array_size(sizeof(*mem->areas), nareas));
499         mem->areas = (struct io_tlb_area *)
500                 __get_free_pages(GFP_KERNEL | __GFP_ZERO, area_order);
501         if (!mem->areas)
502                 goto error_area;
503 
504         mem->slots = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
505                 get_order(array_size(sizeof(*mem->slots), nslabs)));
506         if (!mem->slots)
507                 goto error_slots;
508 
509         set_memory_decrypted((unsigned long)vstart,
510                              (nslabs << IO_TLB_SHIFT) >> PAGE_SHIFT);
511         swiotlb_init_io_tlb_pool(mem, virt_to_phys(vstart), nslabs, true,
512                                  nareas);
513         add_mem_pool(&io_tlb_default_mem, mem);
514 
515         swiotlb_print_info();
516         return 0;
517 
518 error_slots:
519         free_pages((unsigned long)mem->areas, area_order);
520 error_area:
521         free_pages((unsigned long)vstart, order);
522         return -ENOMEM;
523 }
524 
525 void __init swiotlb_exit(void)
526 {
527         struct io_tlb_pool *mem = &io_tlb_default_mem.defpool;
528         unsigned long tbl_vaddr;
529         size_t tbl_size, slots_size;
530         unsigned int area_order;
531 
532         if (swiotlb_force_bounce)
533                 return;
534 
535         if (!mem->nslabs)
536                 return;
537 
538         pr_info("tearing down default memory pool\n");
539         tbl_vaddr = (unsigned long)phys_to_virt(mem->start);
540         tbl_size = PAGE_ALIGN(mem->end - mem->start);
541         slots_size = PAGE_ALIGN(array_size(sizeof(*mem->slots), mem->nslabs));
542 
543         set_memory_encrypted(tbl_vaddr, tbl_size >> PAGE_SHIFT);
544         if (mem->late_alloc) {
545                 area_order = get_order(array_size(sizeof(*mem->areas),
546                         mem->nareas));
547                 free_pages((unsigned long)mem->areas, area_order);
548                 free_pages(tbl_vaddr, get_order(tbl_size));
549                 free_pages((unsigned long)mem->slots, get_order(slots_size));
550         } else {
551                 memblock_free_late(__pa(mem->areas),
552                         array_size(sizeof(*mem->areas), mem->nareas));
553                 memblock_free_late(mem->start, tbl_size);
554                 memblock_free_late(__pa(mem->slots), slots_size);
555         }
556 
557         memset(mem, 0, sizeof(*mem));
558 }
559 
560 #ifdef CONFIG_SWIOTLB_DYNAMIC
561 
562 /**
563  * alloc_dma_pages() - allocate pages to be used for DMA
564  * @gfp:        GFP flags for the allocation.
565  * @bytes:      Size of the buffer.
566  * @phys_limit: Maximum allowed physical address of the buffer.
567  *
568  * Allocate pages from the buddy allocator. If successful, make the allocated
569  * pages decrypted that they can be used for DMA.
570  *
571  * Return: Decrypted pages, %NULL on allocation failure, or ERR_PTR(-EAGAIN)
572  * if the allocated physical address was above @phys_limit.
573  */
574 static struct page *alloc_dma_pages(gfp_t gfp, size_t bytes, u64 phys_limit)
575 {
576         unsigned int order = get_order(bytes);
577         struct page *page;
578         phys_addr_t paddr;
579         void *vaddr;
580 
581         page = alloc_pages(gfp, order);
582         if (!page)
583                 return NULL;
584 
585         paddr = page_to_phys(page);
586         if (paddr + bytes - 1 > phys_limit) {
587                 __free_pages(page, order);
588                 return ERR_PTR(-EAGAIN);
589         }
590 
591         vaddr = phys_to_virt(paddr);
592         if (set_memory_decrypted((unsigned long)vaddr, PFN_UP(bytes)))
593                 goto error;
594         return page;
595 
596 error:
597         /* Intentional leak if pages cannot be encrypted again. */
598         if (!set_memory_encrypted((unsigned long)vaddr, PFN_UP(bytes)))
599                 __free_pages(page, order);
600         return NULL;
601 }
602 
603 /**
604  * swiotlb_alloc_tlb() - allocate a dynamic IO TLB buffer
605  * @dev:        Device for which a memory pool is allocated.
606  * @bytes:      Size of the buffer.
607  * @phys_limit: Maximum allowed physical address of the buffer.
608  * @gfp:        GFP flags for the allocation.
609  *
610  * Return: Allocated pages, or %NULL on allocation failure.
611  */
612 static struct page *swiotlb_alloc_tlb(struct device *dev, size_t bytes,
613                 u64 phys_limit, gfp_t gfp)
614 {
615         struct page *page;
616 
617         /*
618          * Allocate from the atomic pools if memory is encrypted and
619          * the allocation is atomic, because decrypting may block.
620          */
621         if (!gfpflags_allow_blocking(gfp) && dev && force_dma_unencrypted(dev)) {
622                 void *vaddr;
623 
624                 if (!IS_ENABLED(CONFIG_DMA_COHERENT_POOL))
625                         return NULL;
626 
627                 return dma_alloc_from_pool(dev, bytes, &vaddr, gfp,
628                                            dma_coherent_ok);
629         }
630 
631         gfp &= ~GFP_ZONEMASK;
632         if (phys_limit <= DMA_BIT_MASK(zone_dma_bits))
633                 gfp |= __GFP_DMA;
634         else if (phys_limit <= DMA_BIT_MASK(32))
635                 gfp |= __GFP_DMA32;
636 
637         while (IS_ERR(page = alloc_dma_pages(gfp, bytes, phys_limit))) {
638                 if (IS_ENABLED(CONFIG_ZONE_DMA32) &&
639                     phys_limit < DMA_BIT_MASK(64) &&
640                     !(gfp & (__GFP_DMA32 | __GFP_DMA)))
641                         gfp |= __GFP_DMA32;
642                 else if (IS_ENABLED(CONFIG_ZONE_DMA) &&
643                          !(gfp & __GFP_DMA))
644                         gfp = (gfp & ~__GFP_DMA32) | __GFP_DMA;
645                 else
646                         return NULL;
647         }
648 
649         return page;
650 }
651 
652 /**
653  * swiotlb_free_tlb() - free a dynamically allocated IO TLB buffer
654  * @vaddr:      Virtual address of the buffer.
655  * @bytes:      Size of the buffer.
656  */
657 static void swiotlb_free_tlb(void *vaddr, size_t bytes)
658 {
659         if (IS_ENABLED(CONFIG_DMA_COHERENT_POOL) &&
660             dma_free_from_pool(NULL, vaddr, bytes))
661                 return;
662 
663         /* Intentional leak if pages cannot be encrypted again. */
664         if (!set_memory_encrypted((unsigned long)vaddr, PFN_UP(bytes)))
665                 __free_pages(virt_to_page(vaddr), get_order(bytes));
666 }
667 
668 /**
669  * swiotlb_alloc_pool() - allocate a new IO TLB memory pool
670  * @dev:        Device for which a memory pool is allocated.
671  * @minslabs:   Minimum number of slabs.
672  * @nslabs:     Desired (maximum) number of slabs.
673  * @nareas:     Number of areas.
674  * @phys_limit: Maximum DMA buffer physical address.
675  * @gfp:        GFP flags for the allocations.
676  *
677  * Allocate and initialize a new IO TLB memory pool. The actual number of
678  * slabs may be reduced if allocation of @nslabs fails. If even
679  * @minslabs cannot be allocated, this function fails.
680  *
681  * Return: New memory pool, or %NULL on allocation failure.
682  */
683 static struct io_tlb_pool *swiotlb_alloc_pool(struct device *dev,
684                 unsigned long minslabs, unsigned long nslabs,
685                 unsigned int nareas, u64 phys_limit, gfp_t gfp)
686 {
687         struct io_tlb_pool *pool;
688         unsigned int slot_order;
689         struct page *tlb;
690         size_t pool_size;
691         size_t tlb_size;
692 
693         if (nslabs > SLABS_PER_PAGE << MAX_PAGE_ORDER) {
694                 nslabs = SLABS_PER_PAGE << MAX_PAGE_ORDER;
695                 nareas = limit_nareas(nareas, nslabs);
696         }
697 
698         pool_size = sizeof(*pool) + array_size(sizeof(*pool->areas), nareas);
699         pool = kzalloc(pool_size, gfp);
700         if (!pool)
701                 goto error;
702         pool->areas = (void *)pool + sizeof(*pool);
703 
704         tlb_size = nslabs << IO_TLB_SHIFT;
705         while (!(tlb = swiotlb_alloc_tlb(dev, tlb_size, phys_limit, gfp))) {
706                 if (nslabs <= minslabs)
707                         goto error_tlb;
708                 nslabs = ALIGN(nslabs >> 1, IO_TLB_SEGSIZE);
709                 nareas = limit_nareas(nareas, nslabs);
710                 tlb_size = nslabs << IO_TLB_SHIFT;
711         }
712 
713         slot_order = get_order(array_size(sizeof(*pool->slots), nslabs));
714         pool->slots = (struct io_tlb_slot *)
715                 __get_free_pages(gfp, slot_order);
716         if (!pool->slots)
717                 goto error_slots;
718 
719         swiotlb_init_io_tlb_pool(pool, page_to_phys(tlb), nslabs, true, nareas);
720         return pool;
721 
722 error_slots:
723         swiotlb_free_tlb(page_address(tlb), tlb_size);
724 error_tlb:
725         kfree(pool);
726 error:
727         return NULL;
728 }
729 
730 /**
731  * swiotlb_dyn_alloc() - dynamic memory pool allocation worker
732  * @work:       Pointer to dyn_alloc in struct io_tlb_mem.
733  */
734 static void swiotlb_dyn_alloc(struct work_struct *work)
735 {
736         struct io_tlb_mem *mem =
737                 container_of(work, struct io_tlb_mem, dyn_alloc);
738         struct io_tlb_pool *pool;
739 
740         pool = swiotlb_alloc_pool(NULL, IO_TLB_MIN_SLABS, default_nslabs,
741                                   default_nareas, mem->phys_limit, GFP_KERNEL);
742         if (!pool) {
743                 pr_warn_ratelimited("Failed to allocate new pool");
744                 return;
745         }
746 
747         add_mem_pool(mem, pool);
748 }
749 
750 /**
751  * swiotlb_dyn_free() - RCU callback to free a memory pool
752  * @rcu:        RCU head in the corresponding struct io_tlb_pool.
753  */
754 static void swiotlb_dyn_free(struct rcu_head *rcu)
755 {
756         struct io_tlb_pool *pool = container_of(rcu, struct io_tlb_pool, rcu);
757         size_t slots_size = array_size(sizeof(*pool->slots), pool->nslabs);
758         size_t tlb_size = pool->end - pool->start;
759 
760         free_pages((unsigned long)pool->slots, get_order(slots_size));
761         swiotlb_free_tlb(pool->vaddr, tlb_size);
762         kfree(pool);
763 }
764 
765 /**
766  * __swiotlb_find_pool() - find the IO TLB pool for a physical address
767  * @dev:        Device which has mapped the DMA buffer.
768  * @paddr:      Physical address within the DMA buffer.
769  *
770  * Find the IO TLB memory pool descriptor which contains the given physical
771  * address, if any. This function is for use only when the dev is known to
772  * be using swiotlb. Use swiotlb_find_pool() for the more general case
773  * when this condition is not met.
774  *
775  * Return: Memory pool which contains @paddr, or %NULL if none.
776  */
777 struct io_tlb_pool *__swiotlb_find_pool(struct device *dev, phys_addr_t paddr)
778 {
779         struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
780         struct io_tlb_pool *pool;
781 
782         rcu_read_lock();
783         list_for_each_entry_rcu(pool, &mem->pools, node) {
784                 if (paddr >= pool->start && paddr < pool->end)
785                         goto out;
786         }
787 
788         list_for_each_entry_rcu(pool, &dev->dma_io_tlb_pools, node) {
789                 if (paddr >= pool->start && paddr < pool->end)
790                         goto out;
791         }
792         pool = NULL;
793 out:
794         rcu_read_unlock();
795         return pool;
796 }
797 
798 /**
799  * swiotlb_del_pool() - remove an IO TLB pool from a device
800  * @dev:        Owning device.
801  * @pool:       Memory pool to be removed.
802  */
803 static void swiotlb_del_pool(struct device *dev, struct io_tlb_pool *pool)
804 {
805         unsigned long flags;
806 
807         spin_lock_irqsave(&dev->dma_io_tlb_lock, flags);
808         list_del_rcu(&pool->node);
809         spin_unlock_irqrestore(&dev->dma_io_tlb_lock, flags);
810 
811         call_rcu(&pool->rcu, swiotlb_dyn_free);
812 }
813 
814 #endif  /* CONFIG_SWIOTLB_DYNAMIC */
815 
816 /**
817  * swiotlb_dev_init() - initialize swiotlb fields in &struct device
818  * @dev:        Device to be initialized.
819  */
820 void swiotlb_dev_init(struct device *dev)
821 {
822         dev->dma_io_tlb_mem = &io_tlb_default_mem;
823 #ifdef CONFIG_SWIOTLB_DYNAMIC
824         INIT_LIST_HEAD(&dev->dma_io_tlb_pools);
825         spin_lock_init(&dev->dma_io_tlb_lock);
826         dev->dma_uses_io_tlb = false;
827 #endif
828 }
829 
830 /**
831  * swiotlb_align_offset() - Get required offset into an IO TLB allocation.
832  * @dev:         Owning device.
833  * @align_mask:  Allocation alignment mask.
834  * @addr:        DMA address.
835  *
836  * Return the minimum offset from the start of an IO TLB allocation which is
837  * required for a given buffer address and allocation alignment to keep the
838  * device happy.
839  *
840  * First, the address bits covered by min_align_mask must be identical in the
841  * original address and the bounce buffer address. High bits are preserved by
842  * choosing a suitable IO TLB slot, but bits below IO_TLB_SHIFT require extra
843  * padding bytes before the bounce buffer.
844  *
845  * Second, @align_mask specifies which bits of the first allocated slot must
846  * be zero. This may require allocating additional padding slots, and then the
847  * offset (in bytes) from the first such padding slot is returned.
848  */
849 static unsigned int swiotlb_align_offset(struct device *dev,
850                                          unsigned int align_mask, u64 addr)
851 {
852         return addr & dma_get_min_align_mask(dev) &
853                 (align_mask | (IO_TLB_SIZE - 1));
854 }
855 
856 /*
857  * Bounce: copy the swiotlb buffer from or back to the original dma location
858  */
859 static void swiotlb_bounce(struct device *dev, phys_addr_t tlb_addr, size_t size,
860                            enum dma_data_direction dir, struct io_tlb_pool *mem)
861 {
862         int index = (tlb_addr - mem->start) >> IO_TLB_SHIFT;
863         phys_addr_t orig_addr = mem->slots[index].orig_addr;
864         size_t alloc_size = mem->slots[index].alloc_size;
865         unsigned long pfn = PFN_DOWN(orig_addr);
866         unsigned char *vaddr = mem->vaddr + tlb_addr - mem->start;
867         int tlb_offset;
868 
869         if (orig_addr == INVALID_PHYS_ADDR)
870                 return;
871 
872         /*
873          * It's valid for tlb_offset to be negative. This can happen when the
874          * "offset" returned by swiotlb_align_offset() is non-zero, and the
875          * tlb_addr is pointing within the first "offset" bytes of the second
876          * or subsequent slots of the allocated swiotlb area. While it's not
877          * valid for tlb_addr to be pointing within the first "offset" bytes
878          * of the first slot, there's no way to check for such an error since
879          * this function can't distinguish the first slot from the second and
880          * subsequent slots.
881          */
882         tlb_offset = (tlb_addr & (IO_TLB_SIZE - 1)) -
883                      swiotlb_align_offset(dev, 0, orig_addr);
884 
885         orig_addr += tlb_offset;
886         alloc_size -= tlb_offset;
887 
888         if (size > alloc_size) {
889                 dev_WARN_ONCE(dev, 1,
890                         "Buffer overflow detected. Allocation size: %zu. Mapping size: %zu.\n",
891                         alloc_size, size);
892                 size = alloc_size;
893         }
894 
895         if (PageHighMem(pfn_to_page(pfn))) {
896                 unsigned int offset = orig_addr & ~PAGE_MASK;
897                 struct page *page;
898                 unsigned int sz = 0;
899                 unsigned long flags;
900 
901                 while (size) {
902                         sz = min_t(size_t, PAGE_SIZE - offset, size);
903 
904                         local_irq_save(flags);
905                         page = pfn_to_page(pfn);
906                         if (dir == DMA_TO_DEVICE)
907                                 memcpy_from_page(vaddr, page, offset, sz);
908                         else
909                                 memcpy_to_page(page, offset, vaddr, sz);
910                         local_irq_restore(flags);
911 
912                         size -= sz;
913                         pfn++;
914                         vaddr += sz;
915                         offset = 0;
916                 }
917         } else if (dir == DMA_TO_DEVICE) {
918                 memcpy(vaddr, phys_to_virt(orig_addr), size);
919         } else {
920                 memcpy(phys_to_virt(orig_addr), vaddr, size);
921         }
922 }
923 
924 static inline phys_addr_t slot_addr(phys_addr_t start, phys_addr_t idx)
925 {
926         return start + (idx << IO_TLB_SHIFT);
927 }
928 
929 /*
930  * Carefully handle integer overflow which can occur when boundary_mask == ~0UL.
931  */
932 static inline unsigned long get_max_slots(unsigned long boundary_mask)
933 {
934         return (boundary_mask >> IO_TLB_SHIFT) + 1;
935 }
936 
937 static unsigned int wrap_area_index(struct io_tlb_pool *mem, unsigned int index)
938 {
939         if (index >= mem->area_nslabs)
940                 return 0;
941         return index;
942 }
943 
944 /*
945  * Track the total used slots with a global atomic value in order to have
946  * correct information to determine the high water mark. The mem_used()
947  * function gives imprecise results because there's no locking across
948  * multiple areas.
949  */
950 #ifdef CONFIG_DEBUG_FS
951 static void inc_used_and_hiwater(struct io_tlb_mem *mem, unsigned int nslots)
952 {
953         unsigned long old_hiwater, new_used;
954 
955         new_used = atomic_long_add_return(nslots, &mem->total_used);
956         old_hiwater = atomic_long_read(&mem->used_hiwater);
957         do {
958                 if (new_used <= old_hiwater)
959                         break;
960         } while (!atomic_long_try_cmpxchg(&mem->used_hiwater,
961                                           &old_hiwater, new_used));
962 }
963 
964 static void dec_used(struct io_tlb_mem *mem, unsigned int nslots)
965 {
966         atomic_long_sub(nslots, &mem->total_used);
967 }
968 
969 #else /* !CONFIG_DEBUG_FS */
970 static void inc_used_and_hiwater(struct io_tlb_mem *mem, unsigned int nslots)
971 {
972 }
973 static void dec_used(struct io_tlb_mem *mem, unsigned int nslots)
974 {
975 }
976 #endif /* CONFIG_DEBUG_FS */
977 
978 #ifdef CONFIG_SWIOTLB_DYNAMIC
979 #ifdef CONFIG_DEBUG_FS
980 static void inc_transient_used(struct io_tlb_mem *mem, unsigned int nslots)
981 {
982         atomic_long_add(nslots, &mem->transient_nslabs);
983 }
984 
985 static void dec_transient_used(struct io_tlb_mem *mem, unsigned int nslots)
986 {
987         atomic_long_sub(nslots, &mem->transient_nslabs);
988 }
989 
990 #else /* !CONFIG_DEBUG_FS */
991 static void inc_transient_used(struct io_tlb_mem *mem, unsigned int nslots)
992 {
993 }
994 static void dec_transient_used(struct io_tlb_mem *mem, unsigned int nslots)
995 {
996 }
997 #endif /* CONFIG_DEBUG_FS */
998 #endif /* CONFIG_SWIOTLB_DYNAMIC */
999 
1000 /**
1001  * swiotlb_search_pool_area() - search one memory area in one pool
1002  * @dev:        Device which maps the buffer.
1003  * @pool:       Memory pool to be searched.
1004  * @area_index: Index of the IO TLB memory area to be searched.
1005  * @orig_addr:  Original (non-bounced) IO buffer address.
1006  * @alloc_size: Total requested size of the bounce buffer,
1007  *              including initial alignment padding.
1008  * @alloc_align_mask:   Required alignment of the allocated buffer.
1009  *
1010  * Find a suitable sequence of IO TLB entries for the request and allocate
1011  * a buffer from the given IO TLB memory area.
1012  * This function takes care of locking.
1013  *
1014  * Return: Index of the first allocated slot, or -1 on error.
1015  */
1016 static int swiotlb_search_pool_area(struct device *dev, struct io_tlb_pool *pool,
1017                 int area_index, phys_addr_t orig_addr, size_t alloc_size,
1018                 unsigned int alloc_align_mask)
1019 {
1020         struct io_tlb_area *area = pool->areas + area_index;
1021         unsigned long boundary_mask = dma_get_seg_boundary(dev);
1022         dma_addr_t tbl_dma_addr =
1023                 phys_to_dma_unencrypted(dev, pool->start) & boundary_mask;
1024         unsigned long max_slots = get_max_slots(boundary_mask);
1025         unsigned int iotlb_align_mask = dma_get_min_align_mask(dev);
1026         unsigned int nslots = nr_slots(alloc_size), stride;
1027         unsigned int offset = swiotlb_align_offset(dev, 0, orig_addr);
1028         unsigned int index, slots_checked, count = 0, i;
1029         unsigned long flags;
1030         unsigned int slot_base;
1031         unsigned int slot_index;
1032 
1033         BUG_ON(!nslots);
1034         BUG_ON(area_index >= pool->nareas);
1035 
1036         /*
1037          * Historically, swiotlb allocations >= PAGE_SIZE were guaranteed to be
1038          * page-aligned in the absence of any other alignment requirements.
1039          * 'alloc_align_mask' was later introduced to specify the alignment
1040          * explicitly, however this is passed as zero for streaming mappings
1041          * and so we preserve the old behaviour there in case any drivers are
1042          * relying on it.
1043          */
1044         if (!alloc_align_mask && !iotlb_align_mask && alloc_size >= PAGE_SIZE)
1045                 alloc_align_mask = PAGE_SIZE - 1;
1046 
1047         /*
1048          * Ensure that the allocation is at least slot-aligned and update
1049          * 'iotlb_align_mask' to ignore bits that will be preserved when
1050          * offsetting into the allocation.
1051          */
1052         alloc_align_mask |= (IO_TLB_SIZE - 1);
1053         iotlb_align_mask &= ~alloc_align_mask;
1054 
1055         /*
1056          * For mappings with an alignment requirement don't bother looping to
1057          * unaligned slots once we found an aligned one.
1058          */
1059         stride = get_max_slots(max(alloc_align_mask, iotlb_align_mask));
1060 
1061         spin_lock_irqsave(&area->lock, flags);
1062         if (unlikely(nslots > pool->area_nslabs - area->used))
1063                 goto not_found;
1064 
1065         slot_base = area_index * pool->area_nslabs;
1066         index = area->index;
1067 
1068         for (slots_checked = 0; slots_checked < pool->area_nslabs; ) {
1069                 phys_addr_t tlb_addr;
1070 
1071                 slot_index = slot_base + index;
1072                 tlb_addr = slot_addr(tbl_dma_addr, slot_index);
1073 
1074                 if ((tlb_addr & alloc_align_mask) ||
1075                     (orig_addr && (tlb_addr & iotlb_align_mask) !=
1076                                   (orig_addr & iotlb_align_mask))) {
1077                         index = wrap_area_index(pool, index + 1);
1078                         slots_checked++;
1079                         continue;
1080                 }
1081 
1082                 if (!iommu_is_span_boundary(slot_index, nslots,
1083                                             nr_slots(tbl_dma_addr),
1084                                             max_slots)) {
1085                         if (pool->slots[slot_index].list >= nslots)
1086                                 goto found;
1087                 }
1088                 index = wrap_area_index(pool, index + stride);
1089                 slots_checked += stride;
1090         }
1091 
1092 not_found:
1093         spin_unlock_irqrestore(&area->lock, flags);
1094         return -1;
1095 
1096 found:
1097         /*
1098          * If we find a slot that indicates we have 'nslots' number of
1099          * contiguous buffers, we allocate the buffers from that slot onwards
1100          * and set the list of free entries to '' indicating unavailable.
1101          */
1102         for (i = slot_index; i < slot_index + nslots; i++) {
1103                 pool->slots[i].list = 0;
1104                 pool->slots[i].alloc_size = alloc_size - (offset +
1105                                 ((i - slot_index) << IO_TLB_SHIFT));
1106         }
1107         for (i = slot_index - 1;
1108              io_tlb_offset(i) != IO_TLB_SEGSIZE - 1 &&
1109              pool->slots[i].list; i--)
1110                 pool->slots[i].list = ++count;
1111 
1112         /*
1113          * Update the indices to avoid searching in the next round.
1114          */
1115         area->index = wrap_area_index(pool, index + nslots);
1116         area->used += nslots;
1117         spin_unlock_irqrestore(&area->lock, flags);
1118 
1119         inc_used_and_hiwater(dev->dma_io_tlb_mem, nslots);
1120         return slot_index;
1121 }
1122 
1123 #ifdef CONFIG_SWIOTLB_DYNAMIC
1124 
1125 /**
1126  * swiotlb_search_area() - search one memory area in all pools
1127  * @dev:        Device which maps the buffer.
1128  * @start_cpu:  Start CPU number.
1129  * @cpu_offset: Offset from @start_cpu.
1130  * @orig_addr:  Original (non-bounced) IO buffer address.
1131  * @alloc_size: Total requested size of the bounce buffer,
1132  *              including initial alignment padding.
1133  * @alloc_align_mask:   Required alignment of the allocated buffer.
1134  * @retpool:    Used memory pool, updated on return.
1135  *
1136  * Search one memory area in all pools for a sequence of slots that match the
1137  * allocation constraints.
1138  *
1139  * Return: Index of the first allocated slot, or -1 on error.
1140  */
1141 static int swiotlb_search_area(struct device *dev, int start_cpu,
1142                 int cpu_offset, phys_addr_t orig_addr, size_t alloc_size,
1143                 unsigned int alloc_align_mask, struct io_tlb_pool **retpool)
1144 {
1145         struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
1146         struct io_tlb_pool *pool;
1147         int area_index;
1148         int index = -1;
1149 
1150         rcu_read_lock();
1151         list_for_each_entry_rcu(pool, &mem->pools, node) {
1152                 if (cpu_offset >= pool->nareas)
1153                         continue;
1154                 area_index = (start_cpu + cpu_offset) & (pool->nareas - 1);
1155                 index = swiotlb_search_pool_area(dev, pool, area_index,
1156                                                  orig_addr, alloc_size,
1157                                                  alloc_align_mask);
1158                 if (index >= 0) {
1159                         *retpool = pool;
1160                         break;
1161                 }
1162         }
1163         rcu_read_unlock();
1164         return index;
1165 }
1166 
1167 /**
1168  * swiotlb_find_slots() - search for slots in the whole swiotlb
1169  * @dev:        Device which maps the buffer.
1170  * @orig_addr:  Original (non-bounced) IO buffer address.
1171  * @alloc_size: Total requested size of the bounce buffer,
1172  *              including initial alignment padding.
1173  * @alloc_align_mask:   Required alignment of the allocated buffer.
1174  * @retpool:    Used memory pool, updated on return.
1175  *
1176  * Search through the whole software IO TLB to find a sequence of slots that
1177  * match the allocation constraints.
1178  *
1179  * Return: Index of the first allocated slot, or -1 on error.
1180  */
1181 static int swiotlb_find_slots(struct device *dev, phys_addr_t orig_addr,
1182                 size_t alloc_size, unsigned int alloc_align_mask,
1183                 struct io_tlb_pool **retpool)
1184 {
1185         struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
1186         struct io_tlb_pool *pool;
1187         unsigned long nslabs;
1188         unsigned long flags;
1189         u64 phys_limit;
1190         int cpu, i;
1191         int index;
1192 
1193         if (alloc_size > IO_TLB_SEGSIZE * IO_TLB_SIZE)
1194                 return -1;
1195 
1196         cpu = raw_smp_processor_id();
1197         for (i = 0; i < default_nareas; ++i) {
1198                 index = swiotlb_search_area(dev, cpu, i, orig_addr, alloc_size,
1199                                             alloc_align_mask, &pool);
1200                 if (index >= 0)
1201                         goto found;
1202         }
1203 
1204         if (!mem->can_grow)
1205                 return -1;
1206 
1207         schedule_work(&mem->dyn_alloc);
1208 
1209         nslabs = nr_slots(alloc_size);
1210         phys_limit = min_not_zero(*dev->dma_mask, dev->bus_dma_limit);
1211         pool = swiotlb_alloc_pool(dev, nslabs, nslabs, 1, phys_limit,
1212                                   GFP_NOWAIT | __GFP_NOWARN);
1213         if (!pool)
1214                 return -1;
1215 
1216         index = swiotlb_search_pool_area(dev, pool, 0, orig_addr,
1217                                          alloc_size, alloc_align_mask);
1218         if (index < 0) {
1219                 swiotlb_dyn_free(&pool->rcu);
1220                 return -1;
1221         }
1222 
1223         pool->transient = true;
1224         spin_lock_irqsave(&dev->dma_io_tlb_lock, flags);
1225         list_add_rcu(&pool->node, &dev->dma_io_tlb_pools);
1226         spin_unlock_irqrestore(&dev->dma_io_tlb_lock, flags);
1227         inc_transient_used(mem, pool->nslabs);
1228 
1229 found:
1230         WRITE_ONCE(dev->dma_uses_io_tlb, true);
1231 
1232         /*
1233          * The general barrier orders reads and writes against a presumed store
1234          * of the SWIOTLB buffer address by a device driver (to a driver private
1235          * data structure). It serves two purposes.
1236          *
1237          * First, the store to dev->dma_uses_io_tlb must be ordered before the
1238          * presumed store. This guarantees that the returned buffer address
1239          * cannot be passed to another CPU before updating dev->dma_uses_io_tlb.
1240          *
1241          * Second, the load from mem->pools must be ordered before the same
1242          * presumed store. This guarantees that the returned buffer address
1243          * cannot be observed by another CPU before an update of the RCU list
1244          * that was made by swiotlb_dyn_alloc() on a third CPU (cf. multicopy
1245          * atomicity).
1246          *
1247          * See also the comment in swiotlb_find_pool().
1248          */
1249         smp_mb();
1250 
1251         *retpool = pool;
1252         return index;
1253 }
1254 
1255 #else  /* !CONFIG_SWIOTLB_DYNAMIC */
1256 
1257 static int swiotlb_find_slots(struct device *dev, phys_addr_t orig_addr,
1258                 size_t alloc_size, unsigned int alloc_align_mask,
1259                 struct io_tlb_pool **retpool)
1260 {
1261         struct io_tlb_pool *pool;
1262         int start, i;
1263         int index;
1264 
1265         *retpool = pool = &dev->dma_io_tlb_mem->defpool;
1266         i = start = raw_smp_processor_id() & (pool->nareas - 1);
1267         do {
1268                 index = swiotlb_search_pool_area(dev, pool, i, orig_addr,
1269                                                  alloc_size, alloc_align_mask);
1270                 if (index >= 0)
1271                         return index;
1272                 if (++i >= pool->nareas)
1273                         i = 0;
1274         } while (i != start);
1275         return -1;
1276 }
1277 
1278 #endif /* CONFIG_SWIOTLB_DYNAMIC */
1279 
1280 #ifdef CONFIG_DEBUG_FS
1281 
1282 /**
1283  * mem_used() - get number of used slots in an allocator
1284  * @mem:        Software IO TLB allocator.
1285  *
1286  * The result is accurate in this version of the function, because an atomic
1287  * counter is available if CONFIG_DEBUG_FS is set.
1288  *
1289  * Return: Number of used slots.
1290  */
1291 static unsigned long mem_used(struct io_tlb_mem *mem)
1292 {
1293         return atomic_long_read(&mem->total_used);
1294 }
1295 
1296 #else /* !CONFIG_DEBUG_FS */
1297 
1298 /**
1299  * mem_pool_used() - get number of used slots in a memory pool
1300  * @pool:       Software IO TLB memory pool.
1301  *
1302  * The result is not accurate, see mem_used().
1303  *
1304  * Return: Approximate number of used slots.
1305  */
1306 static unsigned long mem_pool_used(struct io_tlb_pool *pool)
1307 {
1308         int i;
1309         unsigned long used = 0;
1310 
1311         for (i = 0; i < pool->nareas; i++)
1312                 used += pool->areas[i].used;
1313         return used;
1314 }
1315 
1316 /**
1317  * mem_used() - get number of used slots in an allocator
1318  * @mem:        Software IO TLB allocator.
1319  *
1320  * The result is not accurate, because there is no locking of individual
1321  * areas.
1322  *
1323  * Return: Approximate number of used slots.
1324  */
1325 static unsigned long mem_used(struct io_tlb_mem *mem)
1326 {
1327 #ifdef CONFIG_SWIOTLB_DYNAMIC
1328         struct io_tlb_pool *pool;
1329         unsigned long used = 0;
1330 
1331         rcu_read_lock();
1332         list_for_each_entry_rcu(pool, &mem->pools, node)
1333                 used += mem_pool_used(pool);
1334         rcu_read_unlock();
1335 
1336         return used;
1337 #else
1338         return mem_pool_used(&mem->defpool);
1339 #endif
1340 }
1341 
1342 #endif /* CONFIG_DEBUG_FS */
1343 
1344 /**
1345  * swiotlb_tbl_map_single() - bounce buffer map a single contiguous physical area
1346  * @dev:                Device which maps the buffer.
1347  * @orig_addr:          Original (non-bounced) physical IO buffer address
1348  * @mapping_size:       Requested size of the actual bounce buffer, excluding
1349  *                      any pre- or post-padding for alignment
1350  * @alloc_align_mask:   Required start and end alignment of the allocated buffer
1351  * @dir:                DMA direction
1352  * @attrs:              Optional DMA attributes for the map operation
1353  *
1354  * Find and allocate a suitable sequence of IO TLB slots for the request.
1355  * The allocated space starts at an alignment specified by alloc_align_mask,
1356  * and the size of the allocated space is rounded up so that the total amount
1357  * of allocated space is a multiple of (alloc_align_mask + 1). If
1358  * alloc_align_mask is zero, the allocated space may be at any alignment and
1359  * the size is not rounded up.
1360  *
1361  * The returned address is within the allocated space and matches the bits
1362  * of orig_addr that are specified in the DMA min_align_mask for the device. As
1363  * such, this returned address may be offset from the beginning of the allocated
1364  * space. The bounce buffer space starting at the returned address for
1365  * mapping_size bytes is initialized to the contents of the original IO buffer
1366  * area. Any pre-padding (due to an offset) and any post-padding (due to
1367  * rounding-up the size) is not initialized.
1368  */
1369 phys_addr_t swiotlb_tbl_map_single(struct device *dev, phys_addr_t orig_addr,
1370                 size_t mapping_size, unsigned int alloc_align_mask,
1371                 enum dma_data_direction dir, unsigned long attrs)
1372 {
1373         struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
1374         unsigned int offset;
1375         struct io_tlb_pool *pool;
1376         unsigned int i;
1377         size_t size;
1378         int index;
1379         phys_addr_t tlb_addr;
1380         unsigned short pad_slots;
1381 
1382         if (!mem || !mem->nslabs) {
1383                 dev_warn_ratelimited(dev,
1384                         "Can not allocate SWIOTLB buffer earlier and can't now provide you with the DMA bounce buffer");
1385                 return (phys_addr_t)DMA_MAPPING_ERROR;
1386         }
1387 
1388         if (cc_platform_has(CC_ATTR_MEM_ENCRYPT))
1389                 pr_warn_once("Memory encryption is active and system is using DMA bounce buffers\n");
1390 
1391         /*
1392          * The default swiotlb memory pool is allocated with PAGE_SIZE
1393          * alignment. If a mapping is requested with larger alignment,
1394          * the mapping may be unable to use the initial slot(s) in all
1395          * sets of IO_TLB_SEGSIZE slots. In such case, a mapping request
1396          * of or near the maximum mapping size would always fail.
1397          */
1398         dev_WARN_ONCE(dev, alloc_align_mask > ~PAGE_MASK,
1399                 "Alloc alignment may prevent fulfilling requests with max mapping_size\n");
1400 
1401         offset = swiotlb_align_offset(dev, alloc_align_mask, orig_addr);
1402         size = ALIGN(mapping_size + offset, alloc_align_mask + 1);
1403         index = swiotlb_find_slots(dev, orig_addr, size, alloc_align_mask, &pool);
1404         if (index == -1) {
1405                 if (!(attrs & DMA_ATTR_NO_WARN))
1406                         dev_warn_ratelimited(dev,
1407         "swiotlb buffer is full (sz: %zd bytes), total %lu (slots), used %lu (slots)\n",
1408                                  size, mem->nslabs, mem_used(mem));
1409                 return (phys_addr_t)DMA_MAPPING_ERROR;
1410         }
1411 
1412         /*
1413          * If dma_skip_sync was set, reset it on first SWIOTLB buffer
1414          * mapping to always sync SWIOTLB buffers.
1415          */
1416         dma_reset_need_sync(dev);
1417 
1418         /*
1419          * Save away the mapping from the original address to the DMA address.
1420          * This is needed when we sync the memory.  Then we sync the buffer if
1421          * needed.
1422          */
1423         pad_slots = offset >> IO_TLB_SHIFT;
1424         offset &= (IO_TLB_SIZE - 1);
1425         index += pad_slots;
1426         pool->slots[index].pad_slots = pad_slots;
1427         for (i = 0; i < (nr_slots(size) - pad_slots); i++)
1428                 pool->slots[index + i].orig_addr = slot_addr(orig_addr, i);
1429         tlb_addr = slot_addr(pool->start, index) + offset;
1430         /*
1431          * When the device is writing memory, i.e. dir == DMA_FROM_DEVICE, copy
1432          * the original buffer to the TLB buffer before initiating DMA in order
1433          * to preserve the original's data if the device does a partial write,
1434          * i.e. if the device doesn't overwrite the entire buffer.  Preserving
1435          * the original data, even if it's garbage, is necessary to match
1436          * hardware behavior.  Use of swiotlb is supposed to be transparent,
1437          * i.e. swiotlb must not corrupt memory by clobbering unwritten bytes.
1438          */
1439         swiotlb_bounce(dev, tlb_addr, mapping_size, DMA_TO_DEVICE, pool);
1440         return tlb_addr;
1441 }
1442 
1443 static void swiotlb_release_slots(struct device *dev, phys_addr_t tlb_addr,
1444                                   struct io_tlb_pool *mem)
1445 {
1446         unsigned long flags;
1447         unsigned int offset = swiotlb_align_offset(dev, 0, tlb_addr);
1448         int index, nslots, aindex;
1449         struct io_tlb_area *area;
1450         int count, i;
1451 
1452         index = (tlb_addr - offset - mem->start) >> IO_TLB_SHIFT;
1453         index -= mem->slots[index].pad_slots;
1454         nslots = nr_slots(mem->slots[index].alloc_size + offset);
1455         aindex = index / mem->area_nslabs;
1456         area = &mem->areas[aindex];
1457 
1458         /*
1459          * Return the buffer to the free list by setting the corresponding
1460          * entries to indicate the number of contiguous entries available.
1461          * While returning the entries to the free list, we merge the entries
1462          * with slots below and above the pool being returned.
1463          */
1464         BUG_ON(aindex >= mem->nareas);
1465 
1466         spin_lock_irqsave(&area->lock, flags);
1467         if (index + nslots < ALIGN(index + 1, IO_TLB_SEGSIZE))
1468                 count = mem->slots[index + nslots].list;
1469         else
1470                 count = 0;
1471 
1472         /*
1473          * Step 1: return the slots to the free list, merging the slots with
1474          * superceeding slots
1475          */
1476         for (i = index + nslots - 1; i >= index; i--) {
1477                 mem->slots[i].list = ++count;
1478                 mem->slots[i].orig_addr = INVALID_PHYS_ADDR;
1479                 mem->slots[i].alloc_size = 0;
1480                 mem->slots[i].pad_slots = 0;
1481         }
1482 
1483         /*
1484          * Step 2: merge the returned slots with the preceding slots, if
1485          * available (non zero)
1486          */
1487         for (i = index - 1;
1488              io_tlb_offset(i) != IO_TLB_SEGSIZE - 1 && mem->slots[i].list;
1489              i--)
1490                 mem->slots[i].list = ++count;
1491         area->used -= nslots;
1492         spin_unlock_irqrestore(&area->lock, flags);
1493 
1494         dec_used(dev->dma_io_tlb_mem, nslots);
1495 }
1496 
1497 #ifdef CONFIG_SWIOTLB_DYNAMIC
1498 
1499 /**
1500  * swiotlb_del_transient() - delete a transient memory pool
1501  * @dev:        Device which mapped the buffer.
1502  * @tlb_addr:   Physical address within a bounce buffer.
1503  * @pool:       Pointer to the transient memory pool to be checked and deleted.
1504  *
1505  * Check whether the address belongs to a transient SWIOTLB memory pool.
1506  * If yes, then delete the pool.
1507  *
1508  * Return: %true if @tlb_addr belonged to a transient pool that was released.
1509  */
1510 static bool swiotlb_del_transient(struct device *dev, phys_addr_t tlb_addr,
1511                 struct io_tlb_pool *pool)
1512 {
1513         if (!pool->transient)
1514                 return false;
1515 
1516         dec_used(dev->dma_io_tlb_mem, pool->nslabs);
1517         swiotlb_del_pool(dev, pool);
1518         dec_transient_used(dev->dma_io_tlb_mem, pool->nslabs);
1519         return true;
1520 }
1521 
1522 #else  /* !CONFIG_SWIOTLB_DYNAMIC */
1523 
1524 static inline bool swiotlb_del_transient(struct device *dev,
1525                 phys_addr_t tlb_addr, struct io_tlb_pool *pool)
1526 {
1527         return false;
1528 }
1529 
1530 #endif  /* CONFIG_SWIOTLB_DYNAMIC */
1531 
1532 /*
1533  * tlb_addr is the physical address of the bounce buffer to unmap.
1534  */
1535 void __swiotlb_tbl_unmap_single(struct device *dev, phys_addr_t tlb_addr,
1536                 size_t mapping_size, enum dma_data_direction dir,
1537                 unsigned long attrs, struct io_tlb_pool *pool)
1538 {
1539         /*
1540          * First, sync the memory before unmapping the entry
1541          */
1542         if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC) &&
1543             (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL))
1544                 swiotlb_bounce(dev, tlb_addr, mapping_size,
1545                                                 DMA_FROM_DEVICE, pool);
1546 
1547         if (swiotlb_del_transient(dev, tlb_addr, pool))
1548                 return;
1549         swiotlb_release_slots(dev, tlb_addr, pool);
1550 }
1551 
1552 void __swiotlb_sync_single_for_device(struct device *dev, phys_addr_t tlb_addr,
1553                 size_t size, enum dma_data_direction dir,
1554                 struct io_tlb_pool *pool)
1555 {
1556         if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL)
1557                 swiotlb_bounce(dev, tlb_addr, size, DMA_TO_DEVICE, pool);
1558         else
1559                 BUG_ON(dir != DMA_FROM_DEVICE);
1560 }
1561 
1562 void __swiotlb_sync_single_for_cpu(struct device *dev, phys_addr_t tlb_addr,
1563                 size_t size, enum dma_data_direction dir,
1564                 struct io_tlb_pool *pool)
1565 {
1566         if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
1567                 swiotlb_bounce(dev, tlb_addr, size, DMA_FROM_DEVICE, pool);
1568         else
1569                 BUG_ON(dir != DMA_TO_DEVICE);
1570 }
1571 
1572 /*
1573  * Create a swiotlb mapping for the buffer at @paddr, and in case of DMAing
1574  * to the device copy the data into it as well.
1575  */
1576 dma_addr_t swiotlb_map(struct device *dev, phys_addr_t paddr, size_t size,
1577                 enum dma_data_direction dir, unsigned long attrs)
1578 {
1579         phys_addr_t swiotlb_addr;
1580         dma_addr_t dma_addr;
1581 
1582         trace_swiotlb_bounced(dev, phys_to_dma(dev, paddr), size);
1583 
1584         swiotlb_addr = swiotlb_tbl_map_single(dev, paddr, size, 0, dir, attrs);
1585         if (swiotlb_addr == (phys_addr_t)DMA_MAPPING_ERROR)
1586                 return DMA_MAPPING_ERROR;
1587 
1588         /* Ensure that the address returned is DMA'ble */
1589         dma_addr = phys_to_dma_unencrypted(dev, swiotlb_addr);
1590         if (unlikely(!dma_capable(dev, dma_addr, size, true))) {
1591                 __swiotlb_tbl_unmap_single(dev, swiotlb_addr, size, dir,
1592                         attrs | DMA_ATTR_SKIP_CPU_SYNC,
1593                         swiotlb_find_pool(dev, swiotlb_addr));
1594                 dev_WARN_ONCE(dev, 1,
1595                         "swiotlb addr %pad+%zu overflow (mask %llx, bus limit %llx).\n",
1596                         &dma_addr, size, *dev->dma_mask, dev->bus_dma_limit);
1597                 return DMA_MAPPING_ERROR;
1598         }
1599 
1600         if (!dev_is_dma_coherent(dev) && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
1601                 arch_sync_dma_for_device(swiotlb_addr, size, dir);
1602         return dma_addr;
1603 }
1604 
1605 size_t swiotlb_max_mapping_size(struct device *dev)
1606 {
1607         int min_align_mask = dma_get_min_align_mask(dev);
1608         int min_align = 0;
1609 
1610         /*
1611          * swiotlb_find_slots() skips slots according to
1612          * min align mask. This affects max mapping size.
1613          * Take it into acount here.
1614          */
1615         if (min_align_mask)
1616                 min_align = roundup(min_align_mask, IO_TLB_SIZE);
1617 
1618         return ((size_t)IO_TLB_SIZE) * IO_TLB_SEGSIZE - min_align;
1619 }
1620 
1621 /**
1622  * is_swiotlb_allocated() - check if the default software IO TLB is initialized
1623  */
1624 bool is_swiotlb_allocated(void)
1625 {
1626         return io_tlb_default_mem.nslabs;
1627 }
1628 
1629 bool is_swiotlb_active(struct device *dev)
1630 {
1631         struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
1632 
1633         return mem && mem->nslabs;
1634 }
1635 
1636 /**
1637  * default_swiotlb_base() - get the base address of the default SWIOTLB
1638  *
1639  * Get the lowest physical address used by the default software IO TLB pool.
1640  */
1641 phys_addr_t default_swiotlb_base(void)
1642 {
1643 #ifdef CONFIG_SWIOTLB_DYNAMIC
1644         io_tlb_default_mem.can_grow = false;
1645 #endif
1646         return io_tlb_default_mem.defpool.start;
1647 }
1648 
1649 /**
1650  * default_swiotlb_limit() - get the address limit of the default SWIOTLB
1651  *
1652  * Get the highest physical address used by the default software IO TLB pool.
1653  */
1654 phys_addr_t default_swiotlb_limit(void)
1655 {
1656 #ifdef CONFIG_SWIOTLB_DYNAMIC
1657         return io_tlb_default_mem.phys_limit;
1658 #else
1659         return io_tlb_default_mem.defpool.end - 1;
1660 #endif
1661 }
1662 
1663 #ifdef CONFIG_DEBUG_FS
1664 #ifdef CONFIG_SWIOTLB_DYNAMIC
1665 static unsigned long mem_transient_used(struct io_tlb_mem *mem)
1666 {
1667         return atomic_long_read(&mem->transient_nslabs);
1668 }
1669 
1670 static int io_tlb_transient_used_get(void *data, u64 *val)
1671 {
1672         struct io_tlb_mem *mem = data;
1673 
1674         *val = mem_transient_used(mem);
1675         return 0;
1676 }
1677 
1678 DEFINE_DEBUGFS_ATTRIBUTE(fops_io_tlb_transient_used, io_tlb_transient_used_get,
1679                          NULL, "%llu\n");
1680 #endif /* CONFIG_SWIOTLB_DYNAMIC */
1681 
1682 static int io_tlb_used_get(void *data, u64 *val)
1683 {
1684         struct io_tlb_mem *mem = data;
1685 
1686         *val = mem_used(mem);
1687         return 0;
1688 }
1689 
1690 static int io_tlb_hiwater_get(void *data, u64 *val)
1691 {
1692         struct io_tlb_mem *mem = data;
1693 
1694         *val = atomic_long_read(&mem->used_hiwater);
1695         return 0;
1696 }
1697 
1698 static int io_tlb_hiwater_set(void *data, u64 val)
1699 {
1700         struct io_tlb_mem *mem = data;
1701 
1702         /* Only allow setting to zero */
1703         if (val != 0)
1704                 return -EINVAL;
1705 
1706         atomic_long_set(&mem->used_hiwater, val);
1707         return 0;
1708 }
1709 
1710 DEFINE_DEBUGFS_ATTRIBUTE(fops_io_tlb_used, io_tlb_used_get, NULL, "%llu\n");
1711 DEFINE_DEBUGFS_ATTRIBUTE(fops_io_tlb_hiwater, io_tlb_hiwater_get,
1712                                 io_tlb_hiwater_set, "%llu\n");
1713 
1714 static void swiotlb_create_debugfs_files(struct io_tlb_mem *mem,
1715                                          const char *dirname)
1716 {
1717         mem->debugfs = debugfs_create_dir(dirname, io_tlb_default_mem.debugfs);
1718         if (!mem->nslabs)
1719                 return;
1720 
1721         debugfs_create_ulong("io_tlb_nslabs", 0400, mem->debugfs, &mem->nslabs);
1722         debugfs_create_file("io_tlb_used", 0400, mem->debugfs, mem,
1723                         &fops_io_tlb_used);
1724         debugfs_create_file("io_tlb_used_hiwater", 0600, mem->debugfs, mem,
1725                         &fops_io_tlb_hiwater);
1726 #ifdef CONFIG_SWIOTLB_DYNAMIC
1727         debugfs_create_file("io_tlb_transient_nslabs", 0400, mem->debugfs,
1728                             mem, &fops_io_tlb_transient_used);
1729 #endif
1730 }
1731 
1732 static int __init swiotlb_create_default_debugfs(void)
1733 {
1734         swiotlb_create_debugfs_files(&io_tlb_default_mem, "swiotlb");
1735         return 0;
1736 }
1737 
1738 late_initcall(swiotlb_create_default_debugfs);
1739 
1740 #else  /* !CONFIG_DEBUG_FS */
1741 
1742 static inline void swiotlb_create_debugfs_files(struct io_tlb_mem *mem,
1743                                                 const char *dirname)
1744 {
1745 }
1746 
1747 #endif  /* CONFIG_DEBUG_FS */
1748 
1749 #ifdef CONFIG_DMA_RESTRICTED_POOL
1750 
1751 struct page *swiotlb_alloc(struct device *dev, size_t size)
1752 {
1753         struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
1754         struct io_tlb_pool *pool;
1755         phys_addr_t tlb_addr;
1756         unsigned int align;
1757         int index;
1758 
1759         if (!mem)
1760                 return NULL;
1761 
1762         align = (1 << (get_order(size) + PAGE_SHIFT)) - 1;
1763         index = swiotlb_find_slots(dev, 0, size, align, &pool);
1764         if (index == -1)
1765                 return NULL;
1766 
1767         tlb_addr = slot_addr(pool->start, index);
1768         if (unlikely(!PAGE_ALIGNED(tlb_addr))) {
1769                 dev_WARN_ONCE(dev, 1, "Cannot allocate pages from non page-aligned swiotlb addr 0x%pa.\n",
1770                               &tlb_addr);
1771                 swiotlb_release_slots(dev, tlb_addr, pool);
1772                 return NULL;
1773         }
1774 
1775         return pfn_to_page(PFN_DOWN(tlb_addr));
1776 }
1777 
1778 bool swiotlb_free(struct device *dev, struct page *page, size_t size)
1779 {
1780         phys_addr_t tlb_addr = page_to_phys(page);
1781         struct io_tlb_pool *pool;
1782 
1783         pool = swiotlb_find_pool(dev, tlb_addr);
1784         if (!pool)
1785                 return false;
1786 
1787         swiotlb_release_slots(dev, tlb_addr, pool);
1788 
1789         return true;
1790 }
1791 
1792 static int rmem_swiotlb_device_init(struct reserved_mem *rmem,
1793                                     struct device *dev)
1794 {
1795         struct io_tlb_mem *mem = rmem->priv;
1796         unsigned long nslabs = rmem->size >> IO_TLB_SHIFT;
1797 
1798         /* Set Per-device io tlb area to one */
1799         unsigned int nareas = 1;
1800 
1801         if (PageHighMem(pfn_to_page(PHYS_PFN(rmem->base)))) {
1802                 dev_err(dev, "Restricted DMA pool must be accessible within the linear mapping.");
1803                 return -EINVAL;
1804         }
1805 
1806         /*
1807          * Since multiple devices can share the same pool, the private data,
1808          * io_tlb_mem struct, will be initialized by the first device attached
1809          * to it.
1810          */
1811         if (!mem) {
1812                 struct io_tlb_pool *pool;
1813 
1814                 mem = kzalloc(sizeof(*mem), GFP_KERNEL);
1815                 if (!mem)
1816                         return -ENOMEM;
1817                 pool = &mem->defpool;
1818 
1819                 pool->slots = kcalloc(nslabs, sizeof(*pool->slots), GFP_KERNEL);
1820                 if (!pool->slots) {
1821                         kfree(mem);
1822                         return -ENOMEM;
1823                 }
1824 
1825                 pool->areas = kcalloc(nareas, sizeof(*pool->areas),
1826                                 GFP_KERNEL);
1827                 if (!pool->areas) {
1828                         kfree(pool->slots);
1829                         kfree(mem);
1830                         return -ENOMEM;
1831                 }
1832 
1833                 set_memory_decrypted((unsigned long)phys_to_virt(rmem->base),
1834                                      rmem->size >> PAGE_SHIFT);
1835                 swiotlb_init_io_tlb_pool(pool, rmem->base, nslabs,
1836                                          false, nareas);
1837                 mem->force_bounce = true;
1838                 mem->for_alloc = true;
1839 #ifdef CONFIG_SWIOTLB_DYNAMIC
1840                 spin_lock_init(&mem->lock);
1841                 INIT_LIST_HEAD_RCU(&mem->pools);
1842 #endif
1843                 add_mem_pool(mem, pool);
1844 
1845                 rmem->priv = mem;
1846 
1847                 swiotlb_create_debugfs_files(mem, rmem->name);
1848         }
1849 
1850         dev->dma_io_tlb_mem = mem;
1851 
1852         return 0;
1853 }
1854 
1855 static void rmem_swiotlb_device_release(struct reserved_mem *rmem,
1856                                         struct device *dev)
1857 {
1858         dev->dma_io_tlb_mem = &io_tlb_default_mem;
1859 }
1860 
1861 static const struct reserved_mem_ops rmem_swiotlb_ops = {
1862         .device_init = rmem_swiotlb_device_init,
1863         .device_release = rmem_swiotlb_device_release,
1864 };
1865 
1866 static int __init rmem_swiotlb_setup(struct reserved_mem *rmem)
1867 {
1868         unsigned long node = rmem->fdt_node;
1869 
1870         if (of_get_flat_dt_prop(node, "reusable", NULL) ||
1871             of_get_flat_dt_prop(node, "linux,cma-default", NULL) ||
1872             of_get_flat_dt_prop(node, "linux,dma-default", NULL) ||
1873             of_get_flat_dt_prop(node, "no-map", NULL))
1874                 return -EINVAL;
1875 
1876         rmem->ops = &rmem_swiotlb_ops;
1877         pr_info("Reserved memory: created restricted DMA pool at %pa, size %ld MiB\n",
1878                 &rmem->base, (unsigned long)rmem->size / SZ_1M);
1879         return 0;
1880 }
1881 
1882 RESERVEDMEM_OF_DECLARE(dma, "restricted-dma-pool", rmem_swiotlb_setup);
1883 #endif /* CONFIG_DMA_RESTRICTED_POOL */
1884 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

sflogo.php