]>
Commit | Line | Data |
---|---|---|
b097186f KRW |
1 | /* |
2 | * Copyright 2010 | |
3 | * by Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> | |
4 | * | |
5 | * This code provides a IOMMU for Xen PV guests with PCI passthrough. | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License v2.0 as published by | |
9 | * the Free Software Foundation | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * PV guests under Xen are running in an non-contiguous memory architecture. | |
17 | * | |
18 | * When PCI pass-through is utilized, this necessitates an IOMMU for | |
19 | * translating bus (DMA) to virtual and vice-versa and also providing a | |
20 | * mechanism to have contiguous pages for device drivers operations (say DMA | |
21 | * operations). | |
22 | * | |
23 | * Specifically, under Xen the Linux idea of pages is an illusion. It | |
24 | * assumes that pages start at zero and go up to the available memory. To | |
25 | * help with that, the Linux Xen MMU provides a lookup mechanism to | |
26 | * translate the page frame numbers (PFN) to machine frame numbers (MFN) | |
27 | * and vice-versa. The MFN are the "real" frame numbers. Furthermore | |
28 | * memory is not contiguous. Xen hypervisor stitches memory for guests | |
29 | * from different pools, which means there is no guarantee that PFN==MFN | |
30 | * and PFN+1==MFN+1. Lastly with Xen 4.0, pages (in debug mode) are | |
31 | * allocated in descending order (high to low), meaning the guest might | |
32 | * never get any MFN's under the 4GB mark. | |
33 | * | |
34 | */ | |
35 | ||
283c0972 JP |
36 | #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt |
37 | ||
b097186f KRW |
38 | #include <linux/bootmem.h> |
39 | #include <linux/dma-mapping.h> | |
63c9744b | 40 | #include <linux/export.h> |
b097186f KRW |
41 | #include <xen/swiotlb-xen.h> |
42 | #include <xen/page.h> | |
43 | #include <xen/xen-ops.h> | |
f4b2f07b | 44 | #include <xen/hvc-console.h> |
b097186f KRW |
45 | /* |
46 | * Used to do a quick range check in swiotlb_tbl_unmap_single and | |
47 | * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this | |
48 | * API. | |
49 | */ | |
50 | ||
51 | static char *xen_io_tlb_start, *xen_io_tlb_end; | |
52 | static unsigned long xen_io_tlb_nslabs; | |
53 | /* | |
54 | * Quick lookup value of the bus address of the IOTLB. | |
55 | */ | |
56 | ||
b8b0f559 | 57 | static u64 start_dma_addr; |
b097186f KRW |
58 | |
59 | static dma_addr_t xen_phys_to_bus(phys_addr_t paddr) | |
60 | { | |
6eab04a8 | 61 | return phys_to_machine(XPADDR(paddr)).maddr; |
b097186f KRW |
62 | } |
63 | ||
64 | static phys_addr_t xen_bus_to_phys(dma_addr_t baddr) | |
65 | { | |
66 | return machine_to_phys(XMADDR(baddr)).paddr; | |
67 | } | |
68 | ||
69 | static dma_addr_t xen_virt_to_bus(void *address) | |
70 | { | |
71 | return xen_phys_to_bus(virt_to_phys(address)); | |
72 | } | |
73 | ||
74 | static int check_pages_physically_contiguous(unsigned long pfn, | |
75 | unsigned int offset, | |
76 | size_t length) | |
77 | { | |
78 | unsigned long next_mfn; | |
79 | int i; | |
80 | int nr_pages; | |
81 | ||
82 | next_mfn = pfn_to_mfn(pfn); | |
83 | nr_pages = (offset + length + PAGE_SIZE-1) >> PAGE_SHIFT; | |
84 | ||
85 | for (i = 1; i < nr_pages; i++) { | |
86 | if (pfn_to_mfn(++pfn) != ++next_mfn) | |
87 | return 0; | |
88 | } | |
89 | return 1; | |
90 | } | |
91 | ||
92 | static int range_straddles_page_boundary(phys_addr_t p, size_t size) | |
93 | { | |
94 | unsigned long pfn = PFN_DOWN(p); | |
95 | unsigned int offset = p & ~PAGE_MASK; | |
96 | ||
97 | if (offset + size <= PAGE_SIZE) | |
98 | return 0; | |
99 | if (check_pages_physically_contiguous(pfn, offset, size)) | |
100 | return 0; | |
101 | return 1; | |
102 | } | |
103 | ||
104 | static int is_xen_swiotlb_buffer(dma_addr_t dma_addr) | |
105 | { | |
106 | unsigned long mfn = PFN_DOWN(dma_addr); | |
107 | unsigned long pfn = mfn_to_local_pfn(mfn); | |
108 | phys_addr_t paddr; | |
109 | ||
110 | /* If the address is outside our domain, it CAN | |
111 | * have the same virtual address as another address | |
112 | * in our domain. Therefore _only_ check address within our domain. | |
113 | */ | |
114 | if (pfn_valid(pfn)) { | |
115 | paddr = PFN_PHYS(pfn); | |
116 | return paddr >= virt_to_phys(xen_io_tlb_start) && | |
117 | paddr < virt_to_phys(xen_io_tlb_end); | |
118 | } | |
119 | return 0; | |
120 | } | |
121 | ||
122 | static int max_dma_bits = 32; | |
123 | ||
124 | static int | |
125 | xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs) | |
126 | { | |
127 | int i, rc; | |
128 | int dma_bits; | |
129 | ||
130 | dma_bits = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT) + PAGE_SHIFT; | |
131 | ||
132 | i = 0; | |
133 | do { | |
134 | int slabs = min(nslabs - i, (unsigned long)IO_TLB_SEGSIZE); | |
135 | ||
136 | do { | |
137 | rc = xen_create_contiguous_region( | |
138 | (unsigned long)buf + (i << IO_TLB_SHIFT), | |
139 | get_order(slabs << IO_TLB_SHIFT), | |
140 | dma_bits); | |
141 | } while (rc && dma_bits++ < max_dma_bits); | |
142 | if (rc) | |
143 | return rc; | |
144 | ||
145 | i += slabs; | |
146 | } while (i < nslabs); | |
147 | return 0; | |
148 | } | |
1cef36a5 KRW |
149 | static unsigned long xen_set_nslabs(unsigned long nr_tbl) |
150 | { | |
151 | if (!nr_tbl) { | |
152 | xen_io_tlb_nslabs = (64 * 1024 * 1024 >> IO_TLB_SHIFT); | |
153 | xen_io_tlb_nslabs = ALIGN(xen_io_tlb_nslabs, IO_TLB_SEGSIZE); | |
154 | } else | |
155 | xen_io_tlb_nslabs = nr_tbl; | |
b097186f | 156 | |
1cef36a5 KRW |
157 | return xen_io_tlb_nslabs << IO_TLB_SHIFT; |
158 | } | |
b097186f | 159 | |
5bab7864 KRW |
160 | enum xen_swiotlb_err { |
161 | XEN_SWIOTLB_UNKNOWN = 0, | |
162 | XEN_SWIOTLB_ENOMEM, | |
163 | XEN_SWIOTLB_EFIXUP | |
164 | }; | |
165 | ||
166 | static const char *xen_swiotlb_error(enum xen_swiotlb_err err) | |
167 | { | |
168 | switch (err) { | |
169 | case XEN_SWIOTLB_ENOMEM: | |
170 | return "Cannot allocate Xen-SWIOTLB buffer\n"; | |
171 | case XEN_SWIOTLB_EFIXUP: | |
172 | return "Failed to get contiguous memory for DMA from Xen!\n"\ | |
173 | "You either: don't have the permissions, do not have"\ | |
174 | " enough free memory under 4GB, or the hypervisor memory"\ | |
175 | " is too fragmented!"; | |
176 | default: | |
177 | break; | |
178 | } | |
179 | return ""; | |
180 | } | |
b8277600 | 181 | int __ref xen_swiotlb_init(int verbose, bool early) |
b097186f | 182 | { |
b8277600 | 183 | unsigned long bytes, order; |
f4b2f07b | 184 | int rc = -ENOMEM; |
5bab7864 | 185 | enum xen_swiotlb_err m_ret = XEN_SWIOTLB_UNKNOWN; |
f4b2f07b | 186 | unsigned int repeat = 3; |
5f98ecdb | 187 | |
1cef36a5 | 188 | xen_io_tlb_nslabs = swiotlb_nr_tbl(); |
f4b2f07b | 189 | retry: |
1cef36a5 | 190 | bytes = xen_set_nslabs(xen_io_tlb_nslabs); |
b8277600 | 191 | order = get_order(xen_io_tlb_nslabs << IO_TLB_SHIFT); |
b097186f KRW |
192 | /* |
193 | * Get IO TLB memory from any location. | |
194 | */ | |
b8277600 KRW |
195 | if (early) |
196 | xen_io_tlb_start = alloc_bootmem_pages(PAGE_ALIGN(bytes)); | |
197 | else { | |
198 | #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT)) | |
199 | #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT) | |
200 | while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) { | |
201 | xen_io_tlb_start = (void *)__get_free_pages(__GFP_NOWARN, order); | |
202 | if (xen_io_tlb_start) | |
203 | break; | |
204 | order--; | |
205 | } | |
206 | if (order != get_order(bytes)) { | |
283c0972 JP |
207 | pr_warn("Warning: only able to allocate %ld MB for software IO TLB\n", |
208 | (PAGE_SIZE << order) >> 20); | |
b8277600 KRW |
209 | xen_io_tlb_nslabs = SLABS_PER_PAGE << order; |
210 | bytes = xen_io_tlb_nslabs << IO_TLB_SHIFT; | |
211 | } | |
212 | } | |
f4b2f07b | 213 | if (!xen_io_tlb_start) { |
5bab7864 | 214 | m_ret = XEN_SWIOTLB_ENOMEM; |
f4b2f07b KRW |
215 | goto error; |
216 | } | |
b097186f KRW |
217 | xen_io_tlb_end = xen_io_tlb_start + bytes; |
218 | /* | |
219 | * And replace that memory with pages under 4GB. | |
220 | */ | |
221 | rc = xen_swiotlb_fixup(xen_io_tlb_start, | |
222 | bytes, | |
223 | xen_io_tlb_nslabs); | |
f4b2f07b | 224 | if (rc) { |
b8277600 KRW |
225 | if (early) |
226 | free_bootmem(__pa(xen_io_tlb_start), PAGE_ALIGN(bytes)); | |
227 | else { | |
228 | free_pages((unsigned long)xen_io_tlb_start, order); | |
229 | xen_io_tlb_start = NULL; | |
230 | } | |
5bab7864 | 231 | m_ret = XEN_SWIOTLB_EFIXUP; |
b097186f | 232 | goto error; |
f4b2f07b | 233 | } |
b097186f | 234 | start_dma_addr = xen_virt_to_bus(xen_io_tlb_start); |
c468bdee | 235 | if (early) { |
ac2cbab2 YL |
236 | if (swiotlb_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs, |
237 | verbose)) | |
238 | panic("Cannot allocate SWIOTLB buffer"); | |
c468bdee KRW |
239 | rc = 0; |
240 | } else | |
b8277600 KRW |
241 | rc = swiotlb_late_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs); |
242 | return rc; | |
b097186f | 243 | error: |
f4b2f07b KRW |
244 | if (repeat--) { |
245 | xen_io_tlb_nslabs = max(1024UL, /* Min is 2MB */ | |
246 | (xen_io_tlb_nslabs >> 1)); | |
283c0972 JP |
247 | pr_info("Lowering to %luMB\n", |
248 | (xen_io_tlb_nslabs << IO_TLB_SHIFT) >> 20); | |
f4b2f07b KRW |
249 | goto retry; |
250 | } | |
283c0972 | 251 | pr_err("%s (rc:%d)\n", xen_swiotlb_error(m_ret), rc); |
b8277600 KRW |
252 | if (early) |
253 | panic("%s (rc:%d)", xen_swiotlb_error(m_ret), rc); | |
254 | else | |
255 | free_pages((unsigned long)xen_io_tlb_start, order); | |
256 | return rc; | |
b097186f | 257 | } |
b097186f KRW |
258 | void * |
259 | xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size, | |
baa676fc AP |
260 | dma_addr_t *dma_handle, gfp_t flags, |
261 | struct dma_attrs *attrs) | |
b097186f KRW |
262 | { |
263 | void *ret; | |
264 | int order = get_order(size); | |
265 | u64 dma_mask = DMA_BIT_MASK(32); | |
266 | unsigned long vstart; | |
6810df88 KRW |
267 | phys_addr_t phys; |
268 | dma_addr_t dev_addr; | |
b097186f KRW |
269 | |
270 | /* | |
271 | * Ignore region specifiers - the kernel's ideas of | |
272 | * pseudo-phys memory layout has nothing to do with the | |
273 | * machine physical layout. We can't allocate highmem | |
274 | * because we can't return a pointer to it. | |
275 | */ | |
276 | flags &= ~(__GFP_DMA | __GFP_HIGHMEM); | |
277 | ||
278 | if (dma_alloc_from_coherent(hwdev, size, dma_handle, &ret)) | |
279 | return ret; | |
280 | ||
281 | vstart = __get_free_pages(flags, order); | |
282 | ret = (void *)vstart; | |
283 | ||
6810df88 KRW |
284 | if (!ret) |
285 | return ret; | |
286 | ||
b097186f | 287 | if (hwdev && hwdev->coherent_dma_mask) |
b5031ed1 | 288 | dma_mask = dma_alloc_coherent_mask(hwdev, flags); |
b097186f | 289 | |
6810df88 KRW |
290 | phys = virt_to_phys(ret); |
291 | dev_addr = xen_phys_to_bus(phys); | |
292 | if (((dev_addr + size - 1 <= dma_mask)) && | |
293 | !range_straddles_page_boundary(phys, size)) | |
294 | *dma_handle = dev_addr; | |
295 | else { | |
b097186f KRW |
296 | if (xen_create_contiguous_region(vstart, order, |
297 | fls64(dma_mask)) != 0) { | |
298 | free_pages(vstart, order); | |
299 | return NULL; | |
300 | } | |
b097186f KRW |
301 | *dma_handle = virt_to_machine(ret).maddr; |
302 | } | |
6810df88 | 303 | memset(ret, 0, size); |
b097186f KRW |
304 | return ret; |
305 | } | |
306 | EXPORT_SYMBOL_GPL(xen_swiotlb_alloc_coherent); | |
307 | ||
308 | void | |
309 | xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr, | |
baa676fc | 310 | dma_addr_t dev_addr, struct dma_attrs *attrs) |
b097186f KRW |
311 | { |
312 | int order = get_order(size); | |
6810df88 KRW |
313 | phys_addr_t phys; |
314 | u64 dma_mask = DMA_BIT_MASK(32); | |
b097186f KRW |
315 | |
316 | if (dma_release_from_coherent(hwdev, order, vaddr)) | |
317 | return; | |
318 | ||
6810df88 KRW |
319 | if (hwdev && hwdev->coherent_dma_mask) |
320 | dma_mask = hwdev->coherent_dma_mask; | |
321 | ||
322 | phys = virt_to_phys(vaddr); | |
323 | ||
324 | if (((dev_addr + size - 1 > dma_mask)) || | |
325 | range_straddles_page_boundary(phys, size)) | |
326 | xen_destroy_contiguous_region((unsigned long)vaddr, order); | |
327 | ||
b097186f KRW |
328 | free_pages((unsigned long)vaddr, order); |
329 | } | |
330 | EXPORT_SYMBOL_GPL(xen_swiotlb_free_coherent); | |
331 | ||
332 | ||
333 | /* | |
334 | * Map a single buffer of the indicated size for DMA in streaming mode. The | |
335 | * physical address to use is returned. | |
336 | * | |
337 | * Once the device is given the dma address, the device owns this memory until | |
338 | * either xen_swiotlb_unmap_page or xen_swiotlb_dma_sync_single is performed. | |
339 | */ | |
340 | dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page, | |
341 | unsigned long offset, size_t size, | |
342 | enum dma_data_direction dir, | |
343 | struct dma_attrs *attrs) | |
344 | { | |
e05ed4d1 | 345 | phys_addr_t map, phys = page_to_phys(page) + offset; |
b097186f | 346 | dma_addr_t dev_addr = xen_phys_to_bus(phys); |
b097186f KRW |
347 | |
348 | BUG_ON(dir == DMA_NONE); | |
349 | /* | |
350 | * If the address happens to be in the device's DMA window, | |
351 | * we can safely return the device addr and not worry about bounce | |
352 | * buffering it. | |
353 | */ | |
354 | if (dma_capable(dev, dev_addr, size) && | |
355 | !range_straddles_page_boundary(phys, size) && !swiotlb_force) | |
356 | return dev_addr; | |
357 | ||
358 | /* | |
359 | * Oh well, have to allocate and map a bounce buffer. | |
360 | */ | |
361 | map = swiotlb_tbl_map_single(dev, start_dma_addr, phys, size, dir); | |
e05ed4d1 | 362 | if (map == SWIOTLB_MAP_ERROR) |
b097186f KRW |
363 | return DMA_ERROR_CODE; |
364 | ||
e05ed4d1 | 365 | dev_addr = xen_phys_to_bus(map); |
b097186f KRW |
366 | |
367 | /* | |
368 | * Ensure that the address returned is DMA'ble | |
369 | */ | |
ab2a47bd | 370 | if (!dma_capable(dev, dev_addr, size)) { |
61ca08c3 | 371 | swiotlb_tbl_unmap_single(dev, map, size, dir); |
ab2a47bd KRW |
372 | dev_addr = 0; |
373 | } | |
b097186f KRW |
374 | return dev_addr; |
375 | } | |
376 | EXPORT_SYMBOL_GPL(xen_swiotlb_map_page); | |
377 | ||
378 | /* | |
379 | * Unmap a single streaming mode DMA translation. The dma_addr and size must | |
380 | * match what was provided for in a previous xen_swiotlb_map_page call. All | |
381 | * other usages are undefined. | |
382 | * | |
383 | * After this call, reads by the cpu to the buffer are guaranteed to see | |
384 | * whatever the device wrote there. | |
385 | */ | |
386 | static void xen_unmap_single(struct device *hwdev, dma_addr_t dev_addr, | |
387 | size_t size, enum dma_data_direction dir) | |
388 | { | |
389 | phys_addr_t paddr = xen_bus_to_phys(dev_addr); | |
390 | ||
391 | BUG_ON(dir == DMA_NONE); | |
392 | ||
393 | /* NOTE: We use dev_addr here, not paddr! */ | |
394 | if (is_xen_swiotlb_buffer(dev_addr)) { | |
61ca08c3 | 395 | swiotlb_tbl_unmap_single(hwdev, paddr, size, dir); |
b097186f KRW |
396 | return; |
397 | } | |
398 | ||
399 | if (dir != DMA_FROM_DEVICE) | |
400 | return; | |
401 | ||
402 | /* | |
403 | * phys_to_virt doesn't work with hihgmem page but we could | |
404 | * call dma_mark_clean() with hihgmem page here. However, we | |
405 | * are fine since dma_mark_clean() is null on POWERPC. We can | |
406 | * make dma_mark_clean() take a physical address if necessary. | |
407 | */ | |
408 | dma_mark_clean(phys_to_virt(paddr), size); | |
409 | } | |
410 | ||
411 | void xen_swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr, | |
412 | size_t size, enum dma_data_direction dir, | |
413 | struct dma_attrs *attrs) | |
414 | { | |
415 | xen_unmap_single(hwdev, dev_addr, size, dir); | |
416 | } | |
417 | EXPORT_SYMBOL_GPL(xen_swiotlb_unmap_page); | |
418 | ||
419 | /* | |
420 | * Make physical memory consistent for a single streaming mode DMA translation | |
421 | * after a transfer. | |
422 | * | |
423 | * If you perform a xen_swiotlb_map_page() but wish to interrogate the buffer | |
424 | * using the cpu, yet do not wish to teardown the dma mapping, you must | |
425 | * call this function before doing so. At the next point you give the dma | |
426 | * address back to the card, you must first perform a | |
427 | * xen_swiotlb_dma_sync_for_device, and then the device again owns the buffer | |
428 | */ | |
429 | static void | |
430 | xen_swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr, | |
431 | size_t size, enum dma_data_direction dir, | |
432 | enum dma_sync_target target) | |
433 | { | |
434 | phys_addr_t paddr = xen_bus_to_phys(dev_addr); | |
435 | ||
436 | BUG_ON(dir == DMA_NONE); | |
437 | ||
438 | /* NOTE: We use dev_addr here, not paddr! */ | |
439 | if (is_xen_swiotlb_buffer(dev_addr)) { | |
fbfda893 | 440 | swiotlb_tbl_sync_single(hwdev, paddr, size, dir, target); |
b097186f KRW |
441 | return; |
442 | } | |
443 | ||
444 | if (dir != DMA_FROM_DEVICE) | |
445 | return; | |
446 | ||
447 | dma_mark_clean(phys_to_virt(paddr), size); | |
448 | } | |
449 | ||
450 | void | |
451 | xen_swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr, | |
452 | size_t size, enum dma_data_direction dir) | |
453 | { | |
454 | xen_swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_CPU); | |
455 | } | |
456 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_single_for_cpu); | |
457 | ||
458 | void | |
459 | xen_swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr, | |
460 | size_t size, enum dma_data_direction dir) | |
461 | { | |
462 | xen_swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_DEVICE); | |
463 | } | |
464 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_single_for_device); | |
465 | ||
466 | /* | |
467 | * Map a set of buffers described by scatterlist in streaming mode for DMA. | |
468 | * This is the scatter-gather version of the above xen_swiotlb_map_page | |
469 | * interface. Here the scatter gather list elements are each tagged with the | |
470 | * appropriate dma address and length. They are obtained via | |
471 | * sg_dma_{address,length}(SG). | |
472 | * | |
473 | * NOTE: An implementation may be able to use a smaller number of | |
474 | * DMA address/length pairs than there are SG table elements. | |
475 | * (for example via virtual mapping capabilities) | |
476 | * The routine returns the number of addr/length pairs actually | |
477 | * used, at most nents. | |
478 | * | |
479 | * Device ownership issues as mentioned above for xen_swiotlb_map_page are the | |
480 | * same here. | |
481 | */ | |
482 | int | |
483 | xen_swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, | |
484 | int nelems, enum dma_data_direction dir, | |
485 | struct dma_attrs *attrs) | |
486 | { | |
487 | struct scatterlist *sg; | |
488 | int i; | |
489 | ||
490 | BUG_ON(dir == DMA_NONE); | |
491 | ||
492 | for_each_sg(sgl, sg, nelems, i) { | |
493 | phys_addr_t paddr = sg_phys(sg); | |
494 | dma_addr_t dev_addr = xen_phys_to_bus(paddr); | |
495 | ||
496 | if (swiotlb_force || | |
497 | !dma_capable(hwdev, dev_addr, sg->length) || | |
498 | range_straddles_page_boundary(paddr, sg->length)) { | |
e05ed4d1 AD |
499 | phys_addr_t map = swiotlb_tbl_map_single(hwdev, |
500 | start_dma_addr, | |
501 | sg_phys(sg), | |
502 | sg->length, | |
503 | dir); | |
504 | if (map == SWIOTLB_MAP_ERROR) { | |
b097186f KRW |
505 | /* Don't panic here, we expect map_sg users |
506 | to do proper error handling. */ | |
507 | xen_swiotlb_unmap_sg_attrs(hwdev, sgl, i, dir, | |
508 | attrs); | |
781575cd | 509 | sg_dma_len(sgl) = 0; |
b097186f KRW |
510 | return DMA_ERROR_CODE; |
511 | } | |
e05ed4d1 | 512 | sg->dma_address = xen_phys_to_bus(map); |
b097186f KRW |
513 | } else |
514 | sg->dma_address = dev_addr; | |
781575cd | 515 | sg_dma_len(sg) = sg->length; |
b097186f KRW |
516 | } |
517 | return nelems; | |
518 | } | |
519 | EXPORT_SYMBOL_GPL(xen_swiotlb_map_sg_attrs); | |
520 | ||
b097186f KRW |
521 | /* |
522 | * Unmap a set of streaming mode DMA translations. Again, cpu read rules | |
523 | * concerning calls here are the same as for swiotlb_unmap_page() above. | |
524 | */ | |
525 | void | |
526 | xen_swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl, | |
527 | int nelems, enum dma_data_direction dir, | |
528 | struct dma_attrs *attrs) | |
529 | { | |
530 | struct scatterlist *sg; | |
531 | int i; | |
532 | ||
533 | BUG_ON(dir == DMA_NONE); | |
534 | ||
535 | for_each_sg(sgl, sg, nelems, i) | |
781575cd | 536 | xen_unmap_single(hwdev, sg->dma_address, sg_dma_len(sg), dir); |
b097186f KRW |
537 | |
538 | } | |
539 | EXPORT_SYMBOL_GPL(xen_swiotlb_unmap_sg_attrs); | |
540 | ||
b097186f KRW |
541 | /* |
542 | * Make physical memory consistent for a set of streaming mode DMA translations | |
543 | * after a transfer. | |
544 | * | |
545 | * The same as swiotlb_sync_single_* but for a scatter-gather list, same rules | |
546 | * and usage. | |
547 | */ | |
548 | static void | |
549 | xen_swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl, | |
550 | int nelems, enum dma_data_direction dir, | |
551 | enum dma_sync_target target) | |
552 | { | |
553 | struct scatterlist *sg; | |
554 | int i; | |
555 | ||
556 | for_each_sg(sgl, sg, nelems, i) | |
557 | xen_swiotlb_sync_single(hwdev, sg->dma_address, | |
781575cd | 558 | sg_dma_len(sg), dir, target); |
b097186f KRW |
559 | } |
560 | ||
561 | void | |
562 | xen_swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg, | |
563 | int nelems, enum dma_data_direction dir) | |
564 | { | |
565 | xen_swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_CPU); | |
566 | } | |
567 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_sg_for_cpu); | |
568 | ||
569 | void | |
570 | xen_swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, | |
571 | int nelems, enum dma_data_direction dir) | |
572 | { | |
573 | xen_swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_DEVICE); | |
574 | } | |
575 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_sg_for_device); | |
576 | ||
577 | int | |
578 | xen_swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr) | |
579 | { | |
580 | return !dma_addr; | |
581 | } | |
582 | EXPORT_SYMBOL_GPL(xen_swiotlb_dma_mapping_error); | |
583 | ||
584 | /* | |
585 | * Return whether the given device DMA address mask can be supported | |
586 | * properly. For example, if your device can only drive the low 24-bits | |
587 | * during bus mastering, then you would pass 0x00ffffff as the mask to | |
588 | * this function. | |
589 | */ | |
590 | int | |
591 | xen_swiotlb_dma_supported(struct device *hwdev, u64 mask) | |
592 | { | |
593 | return xen_virt_to_bus(xen_io_tlb_end - 1) <= mask; | |
594 | } | |
595 | EXPORT_SYMBOL_GPL(xen_swiotlb_dma_supported); |