This source file includes following definitions.
- xen_phys_to_bus
- xen_bus_to_phys
- xen_virt_to_bus
- range_straddles_page_boundary
- is_xen_swiotlb_buffer
- xen_swiotlb_fixup
- xen_set_nslabs
- xen_swiotlb_error
- xen_swiotlb_init
- xen_swiotlb_alloc_coherent
- xen_swiotlb_free_coherent
- xen_swiotlb_map_page
- xen_swiotlb_unmap_page
- xen_swiotlb_sync_single_for_cpu
- xen_swiotlb_sync_single_for_device
- xen_swiotlb_unmap_sg
- xen_swiotlb_map_sg
- xen_swiotlb_sync_sg_for_cpu
- xen_swiotlb_sync_sg_for_device
- xen_swiotlb_dma_supported
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
28
29 #include <linux/memblock.h>
30 #include <linux/dma-direct.h>
31 #include <linux/dma-noncoherent.h>
32 #include <linux/export.h>
33 #include <xen/swiotlb-xen.h>
34 #include <xen/page.h>
35 #include <xen/xen-ops.h>
36 #include <xen/hvc-console.h>
37
38 #include <asm/dma-mapping.h>
39 #include <asm/xen/page-coherent.h>
40
41 #include <trace/events/swiotlb.h>
42 #define MAX_DMA_BITS 32
43
44
45
46
47
48
49 static char *xen_io_tlb_start, *xen_io_tlb_end;
50 static unsigned long xen_io_tlb_nslabs;
51
52
53
54
55 static u64 start_dma_addr;
56
57
58
59
60
61
62 static inline dma_addr_t xen_phys_to_bus(phys_addr_t paddr)
63 {
64 unsigned long bfn = pfn_to_bfn(XEN_PFN_DOWN(paddr));
65 dma_addr_t dma = (dma_addr_t)bfn << XEN_PAGE_SHIFT;
66
67 dma |= paddr & ~XEN_PAGE_MASK;
68
69 return dma;
70 }
71
72 static inline phys_addr_t xen_bus_to_phys(dma_addr_t baddr)
73 {
74 unsigned long xen_pfn = bfn_to_pfn(XEN_PFN_DOWN(baddr));
75 dma_addr_t dma = (dma_addr_t)xen_pfn << XEN_PAGE_SHIFT;
76 phys_addr_t paddr = dma;
77
78 paddr |= baddr & ~XEN_PAGE_MASK;
79
80 return paddr;
81 }
82
83 static inline dma_addr_t xen_virt_to_bus(void *address)
84 {
85 return xen_phys_to_bus(virt_to_phys(address));
86 }
87
88 static inline int range_straddles_page_boundary(phys_addr_t p, size_t size)
89 {
90 unsigned long next_bfn, xen_pfn = XEN_PFN_DOWN(p);
91 unsigned int i, nr_pages = XEN_PFN_UP(xen_offset_in_page(p) + size);
92
93 next_bfn = pfn_to_bfn(xen_pfn);
94
95 for (i = 1; i < nr_pages; i++)
96 if (pfn_to_bfn(++xen_pfn) != ++next_bfn)
97 return 1;
98
99 return 0;
100 }
101
102 static int is_xen_swiotlb_buffer(dma_addr_t dma_addr)
103 {
104 unsigned long bfn = XEN_PFN_DOWN(dma_addr);
105 unsigned long xen_pfn = bfn_to_local_pfn(bfn);
106 phys_addr_t paddr = XEN_PFN_PHYS(xen_pfn);
107
108
109
110
111
112 if (pfn_valid(PFN_DOWN(paddr))) {
113 return paddr >= virt_to_phys(xen_io_tlb_start) &&
114 paddr < virt_to_phys(xen_io_tlb_end);
115 }
116 return 0;
117 }
118
119 static int
120 xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs)
121 {
122 int i, rc;
123 int dma_bits;
124 dma_addr_t dma_handle;
125 phys_addr_t p = virt_to_phys(buf);
126
127 dma_bits = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT) + PAGE_SHIFT;
128
129 i = 0;
130 do {
131 int slabs = min(nslabs - i, (unsigned long)IO_TLB_SEGSIZE);
132
133 do {
134 rc = xen_create_contiguous_region(
135 p + (i << IO_TLB_SHIFT),
136 get_order(slabs << IO_TLB_SHIFT),
137 dma_bits, &dma_handle);
138 } while (rc && dma_bits++ < MAX_DMA_BITS);
139 if (rc)
140 return rc;
141
142 i += slabs;
143 } while (i < nslabs);
144 return 0;
145 }
146 static unsigned long xen_set_nslabs(unsigned long nr_tbl)
147 {
148 if (!nr_tbl) {
149 xen_io_tlb_nslabs = (64 * 1024 * 1024 >> IO_TLB_SHIFT);
150 xen_io_tlb_nslabs = ALIGN(xen_io_tlb_nslabs, IO_TLB_SEGSIZE);
151 } else
152 xen_io_tlb_nslabs = nr_tbl;
153
154 return xen_io_tlb_nslabs << IO_TLB_SHIFT;
155 }
156
157 enum xen_swiotlb_err {
158 XEN_SWIOTLB_UNKNOWN = 0,
159 XEN_SWIOTLB_ENOMEM,
160 XEN_SWIOTLB_EFIXUP
161 };
162
163 static const char *xen_swiotlb_error(enum xen_swiotlb_err err)
164 {
165 switch (err) {
166 case XEN_SWIOTLB_ENOMEM:
167 return "Cannot allocate Xen-SWIOTLB buffer\n";
168 case XEN_SWIOTLB_EFIXUP:
169 return "Failed to get contiguous memory for DMA from Xen!\n"\
170 "You either: don't have the permissions, do not have"\
171 " enough free memory under 4GB, or the hypervisor memory"\
172 " is too fragmented!";
173 default:
174 break;
175 }
176 return "";
177 }
178 int __ref xen_swiotlb_init(int verbose, bool early)
179 {
180 unsigned long bytes, order;
181 int rc = -ENOMEM;
182 enum xen_swiotlb_err m_ret = XEN_SWIOTLB_UNKNOWN;
183 unsigned int repeat = 3;
184
185 xen_io_tlb_nslabs = swiotlb_nr_tbl();
186 retry:
187 bytes = xen_set_nslabs(xen_io_tlb_nslabs);
188 order = get_order(xen_io_tlb_nslabs << IO_TLB_SHIFT);
189
190
191
192
193 if (io_tlb_start != 0) {
194 xen_io_tlb_start = phys_to_virt(io_tlb_start);
195 goto end;
196 }
197
198
199
200
201 if (early) {
202 xen_io_tlb_start = memblock_alloc(PAGE_ALIGN(bytes),
203 PAGE_SIZE);
204 if (!xen_io_tlb_start)
205 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
206 __func__, PAGE_ALIGN(bytes), PAGE_SIZE);
207 } else {
208 #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT))
209 #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT)
210 while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) {
211 xen_io_tlb_start = (void *)xen_get_swiotlb_free_pages(order);
212 if (xen_io_tlb_start)
213 break;
214 order--;
215 }
216 if (order != get_order(bytes)) {
217 pr_warn("Warning: only able to allocate %ld MB for software IO TLB\n",
218 (PAGE_SIZE << order) >> 20);
219 xen_io_tlb_nslabs = SLABS_PER_PAGE << order;
220 bytes = xen_io_tlb_nslabs << IO_TLB_SHIFT;
221 }
222 }
223 if (!xen_io_tlb_start) {
224 m_ret = XEN_SWIOTLB_ENOMEM;
225 goto error;
226 }
227
228
229
230 rc = xen_swiotlb_fixup(xen_io_tlb_start,
231 bytes,
232 xen_io_tlb_nslabs);
233 if (rc) {
234 if (early)
235 memblock_free(__pa(xen_io_tlb_start),
236 PAGE_ALIGN(bytes));
237 else {
238 free_pages((unsigned long)xen_io_tlb_start, order);
239 xen_io_tlb_start = NULL;
240 }
241 m_ret = XEN_SWIOTLB_EFIXUP;
242 goto error;
243 }
244 start_dma_addr = xen_virt_to_bus(xen_io_tlb_start);
245 if (early) {
246 if (swiotlb_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs,
247 verbose))
248 panic("Cannot allocate SWIOTLB buffer");
249 rc = 0;
250 } else
251 rc = swiotlb_late_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs);
252
253 end:
254 xen_io_tlb_end = xen_io_tlb_start + bytes;
255 if (!rc)
256 swiotlb_set_max_segment(PAGE_SIZE);
257
258 return rc;
259 error:
260 if (repeat--) {
261 xen_io_tlb_nslabs = max(1024UL,
262 (xen_io_tlb_nslabs >> 1));
263 pr_info("Lowering to %luMB\n",
264 (xen_io_tlb_nslabs << IO_TLB_SHIFT) >> 20);
265 goto retry;
266 }
267 pr_err("%s (rc:%d)\n", xen_swiotlb_error(m_ret), rc);
268 if (early)
269 panic("%s (rc:%d)", xen_swiotlb_error(m_ret), rc);
270 else
271 free_pages((unsigned long)xen_io_tlb_start, order);
272 return rc;
273 }
274
275 static void *
276 xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size,
277 dma_addr_t *dma_handle, gfp_t flags,
278 unsigned long attrs)
279 {
280 void *ret;
281 int order = get_order(size);
282 u64 dma_mask = DMA_BIT_MASK(32);
283 phys_addr_t phys;
284 dma_addr_t dev_addr;
285
286
287
288
289
290
291
292 flags &= ~(__GFP_DMA | __GFP_HIGHMEM);
293
294
295 size = 1UL << (order + XEN_PAGE_SHIFT);
296
297
298
299
300
301
302 ret = xen_alloc_coherent_pages(hwdev, size, dma_handle, flags, attrs);
303
304 if (!ret)
305 return ret;
306
307 if (hwdev && hwdev->coherent_dma_mask)
308 dma_mask = hwdev->coherent_dma_mask;
309
310
311
312
313
314 phys = *dma_handle;
315 dev_addr = xen_phys_to_bus(phys);
316 if (((dev_addr + size - 1 <= dma_mask)) &&
317 !range_straddles_page_boundary(phys, size))
318 *dma_handle = dev_addr;
319 else {
320 if (xen_create_contiguous_region(phys, order,
321 fls64(dma_mask), dma_handle) != 0) {
322 xen_free_coherent_pages(hwdev, size, ret, (dma_addr_t)phys, attrs);
323 return NULL;
324 }
325 SetPageXenRemapped(virt_to_page(ret));
326 }
327 memset(ret, 0, size);
328 return ret;
329 }
330
331 static void
332 xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr,
333 dma_addr_t dev_addr, unsigned long attrs)
334 {
335 int order = get_order(size);
336 phys_addr_t phys;
337 u64 dma_mask = DMA_BIT_MASK(32);
338
339 if (hwdev && hwdev->coherent_dma_mask)
340 dma_mask = hwdev->coherent_dma_mask;
341
342
343
344 phys = xen_bus_to_phys(dev_addr);
345
346
347 size = 1UL << (order + XEN_PAGE_SHIFT);
348
349 if (!WARN_ON((dev_addr + size - 1 > dma_mask) ||
350 range_straddles_page_boundary(phys, size)) &&
351 TestClearPageXenRemapped(virt_to_page(vaddr)))
352 xen_destroy_contiguous_region(phys, order);
353
354 xen_free_coherent_pages(hwdev, size, vaddr, (dma_addr_t)phys, attrs);
355 }
356
357
358
359
360
361
362
363
364 static dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page,
365 unsigned long offset, size_t size,
366 enum dma_data_direction dir,
367 unsigned long attrs)
368 {
369 phys_addr_t map, phys = page_to_phys(page) + offset;
370 dma_addr_t dev_addr = xen_phys_to_bus(phys);
371
372 BUG_ON(dir == DMA_NONE);
373
374
375
376
377
378 if (dma_capable(dev, dev_addr, size) &&
379 !range_straddles_page_boundary(phys, size) &&
380 !xen_arch_need_swiotlb(dev, phys, dev_addr) &&
381 swiotlb_force != SWIOTLB_FORCE)
382 goto done;
383
384
385
386
387 trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force);
388
389 map = swiotlb_tbl_map_single(dev, start_dma_addr, phys,
390 size, size, dir, attrs);
391 if (map == (phys_addr_t)DMA_MAPPING_ERROR)
392 return DMA_MAPPING_ERROR;
393
394 phys = map;
395 dev_addr = xen_phys_to_bus(map);
396
397
398
399
400 if (unlikely(!dma_capable(dev, dev_addr, size))) {
401 swiotlb_tbl_unmap_single(dev, map, size, size, dir,
402 attrs | DMA_ATTR_SKIP_CPU_SYNC);
403 return DMA_MAPPING_ERROR;
404 }
405
406 done:
407 if (!dev_is_dma_coherent(dev) && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
408 xen_dma_sync_for_device(dev, dev_addr, phys, size, dir);
409 return dev_addr;
410 }
411
412
413
414
415
416
417
418
419
420 static void xen_swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr,
421 size_t size, enum dma_data_direction dir, unsigned long attrs)
422 {
423 phys_addr_t paddr = xen_bus_to_phys(dev_addr);
424
425 BUG_ON(dir == DMA_NONE);
426
427 if (!dev_is_dma_coherent(hwdev) && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
428 xen_dma_sync_for_cpu(hwdev, dev_addr, paddr, size, dir);
429
430
431 if (is_xen_swiotlb_buffer(dev_addr))
432 swiotlb_tbl_unmap_single(hwdev, paddr, size, size, dir, attrs);
433 }
434
435 static void
436 xen_swiotlb_sync_single_for_cpu(struct device *dev, dma_addr_t dma_addr,
437 size_t size, enum dma_data_direction dir)
438 {
439 phys_addr_t paddr = xen_bus_to_phys(dma_addr);
440
441 if (!dev_is_dma_coherent(dev))
442 xen_dma_sync_for_cpu(dev, dma_addr, paddr, size, dir);
443
444 if (is_xen_swiotlb_buffer(dma_addr))
445 swiotlb_tbl_sync_single(dev, paddr, size, dir, SYNC_FOR_CPU);
446 }
447
448 static void
449 xen_swiotlb_sync_single_for_device(struct device *dev, dma_addr_t dma_addr,
450 size_t size, enum dma_data_direction dir)
451 {
452 phys_addr_t paddr = xen_bus_to_phys(dma_addr);
453
454 if (is_xen_swiotlb_buffer(dma_addr))
455 swiotlb_tbl_sync_single(dev, paddr, size, dir, SYNC_FOR_DEVICE);
456
457 if (!dev_is_dma_coherent(dev))
458 xen_dma_sync_for_device(dev, dma_addr, paddr, size, dir);
459 }
460
461
462
463
464
465 static void
466 xen_swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sgl, int nelems,
467 enum dma_data_direction dir, unsigned long attrs)
468 {
469 struct scatterlist *sg;
470 int i;
471
472 BUG_ON(dir == DMA_NONE);
473
474 for_each_sg(sgl, sg, nelems, i)
475 xen_swiotlb_unmap_page(hwdev, sg->dma_address, sg_dma_len(sg),
476 dir, attrs);
477
478 }
479
480 static int
481 xen_swiotlb_map_sg(struct device *dev, struct scatterlist *sgl, int nelems,
482 enum dma_data_direction dir, unsigned long attrs)
483 {
484 struct scatterlist *sg;
485 int i;
486
487 BUG_ON(dir == DMA_NONE);
488
489 for_each_sg(sgl, sg, nelems, i) {
490 sg->dma_address = xen_swiotlb_map_page(dev, sg_page(sg),
491 sg->offset, sg->length, dir, attrs);
492 if (sg->dma_address == DMA_MAPPING_ERROR)
493 goto out_unmap;
494 sg_dma_len(sg) = sg->length;
495 }
496
497 return nelems;
498 out_unmap:
499 xen_swiotlb_unmap_sg(dev, sgl, i, dir, attrs | DMA_ATTR_SKIP_CPU_SYNC);
500 sg_dma_len(sgl) = 0;
501 return 0;
502 }
503
504 static void
505 xen_swiotlb_sync_sg_for_cpu(struct device *dev, struct scatterlist *sgl,
506 int nelems, enum dma_data_direction dir)
507 {
508 struct scatterlist *sg;
509 int i;
510
511 for_each_sg(sgl, sg, nelems, i) {
512 xen_swiotlb_sync_single_for_cpu(dev, sg->dma_address,
513 sg->length, dir);
514 }
515 }
516
517 static void
518 xen_swiotlb_sync_sg_for_device(struct device *dev, struct scatterlist *sgl,
519 int nelems, enum dma_data_direction dir)
520 {
521 struct scatterlist *sg;
522 int i;
523
524 for_each_sg(sgl, sg, nelems, i) {
525 xen_swiotlb_sync_single_for_device(dev, sg->dma_address,
526 sg->length, dir);
527 }
528 }
529
530
531
532
533
534
535
536 static int
537 xen_swiotlb_dma_supported(struct device *hwdev, u64 mask)
538 {
539 return xen_virt_to_bus(xen_io_tlb_end - 1) <= mask;
540 }
541
542 const struct dma_map_ops xen_swiotlb_dma_ops = {
543 .alloc = xen_swiotlb_alloc_coherent,
544 .free = xen_swiotlb_free_coherent,
545 .sync_single_for_cpu = xen_swiotlb_sync_single_for_cpu,
546 .sync_single_for_device = xen_swiotlb_sync_single_for_device,
547 .sync_sg_for_cpu = xen_swiotlb_sync_sg_for_cpu,
548 .sync_sg_for_device = xen_swiotlb_sync_sg_for_device,
549 .map_sg = xen_swiotlb_map_sg,
550 .unmap_sg = xen_swiotlb_unmap_sg,
551 .map_page = xen_swiotlb_map_page,
552 .unmap_page = xen_swiotlb_unmap_page,
553 .dma_supported = xen_swiotlb_dma_supported,
554 .mmap = dma_common_mmap,
555 .get_sgtable = dma_common_get_sgtable,
556 };