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> |
83862ccf | 45 | #include <asm/dma-mapping.h> |
b097186f KRW |
46 | /* |
47 | * Used to do a quick range check in swiotlb_tbl_unmap_single and | |
48 | * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this | |
49 | * API. | |
50 | */ | |
51 | ||
83862ccf SS |
52 | #ifndef CONFIG_X86 |
53 | static unsigned long dma_alloc_coherent_mask(struct device *dev, | |
54 | gfp_t gfp) | |
55 | { | |
56 | unsigned long dma_mask = 0; | |
57 | ||
58 | dma_mask = dev->coherent_dma_mask; | |
59 | if (!dma_mask) | |
60 | dma_mask = (gfp & GFP_DMA) ? DMA_BIT_MASK(24) : DMA_BIT_MASK(32); | |
61 | ||
62 | return dma_mask; | |
63 | } | |
64 | #endif | |
65 | ||
b097186f KRW |
66 | static char *xen_io_tlb_start, *xen_io_tlb_end; |
67 | static unsigned long xen_io_tlb_nslabs; | |
68 | /* | |
69 | * Quick lookup value of the bus address of the IOTLB. | |
70 | */ | |
71 | ||
b8b0f559 | 72 | static u64 start_dma_addr; |
b097186f KRW |
73 | |
74 | static dma_addr_t xen_phys_to_bus(phys_addr_t paddr) | |
75 | { | |
6eab04a8 | 76 | return phys_to_machine(XPADDR(paddr)).maddr; |
b097186f KRW |
77 | } |
78 | ||
79 | static phys_addr_t xen_bus_to_phys(dma_addr_t baddr) | |
80 | { | |
81 | return machine_to_phys(XMADDR(baddr)).paddr; | |
82 | } | |
83 | ||
84 | static dma_addr_t xen_virt_to_bus(void *address) | |
85 | { | |
86 | return xen_phys_to_bus(virt_to_phys(address)); | |
87 | } | |
88 | ||
89 | static int check_pages_physically_contiguous(unsigned long pfn, | |
90 | unsigned int offset, | |
91 | size_t length) | |
92 | { | |
93 | unsigned long next_mfn; | |
94 | int i; | |
95 | int nr_pages; | |
96 | ||
97 | next_mfn = pfn_to_mfn(pfn); | |
98 | nr_pages = (offset + length + PAGE_SIZE-1) >> PAGE_SHIFT; | |
99 | ||
100 | for (i = 1; i < nr_pages; i++) { | |
101 | if (pfn_to_mfn(++pfn) != ++next_mfn) | |
102 | return 0; | |
103 | } | |
104 | return 1; | |
105 | } | |
106 | ||
107 | static int range_straddles_page_boundary(phys_addr_t p, size_t size) | |
108 | { | |
109 | unsigned long pfn = PFN_DOWN(p); | |
110 | unsigned int offset = p & ~PAGE_MASK; | |
111 | ||
112 | if (offset + size <= PAGE_SIZE) | |
113 | return 0; | |
114 | if (check_pages_physically_contiguous(pfn, offset, size)) | |
115 | return 0; | |
116 | return 1; | |
117 | } | |
118 | ||
119 | static int is_xen_swiotlb_buffer(dma_addr_t dma_addr) | |
120 | { | |
121 | unsigned long mfn = PFN_DOWN(dma_addr); | |
122 | unsigned long pfn = mfn_to_local_pfn(mfn); | |
123 | phys_addr_t paddr; | |
124 | ||
125 | /* If the address is outside our domain, it CAN | |
126 | * have the same virtual address as another address | |
127 | * in our domain. Therefore _only_ check address within our domain. | |
128 | */ | |
129 | if (pfn_valid(pfn)) { | |
130 | paddr = PFN_PHYS(pfn); | |
131 | return paddr >= virt_to_phys(xen_io_tlb_start) && | |
132 | paddr < virt_to_phys(xen_io_tlb_end); | |
133 | } | |
134 | return 0; | |
135 | } | |
136 | ||
137 | static int max_dma_bits = 32; | |
138 | ||
139 | static int | |
140 | xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs) | |
141 | { | |
142 | int i, rc; | |
143 | int dma_bits; | |
69908907 | 144 | dma_addr_t dma_handle; |
b097186f KRW |
145 | |
146 | dma_bits = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT) + PAGE_SHIFT; | |
147 | ||
148 | i = 0; | |
149 | do { | |
150 | int slabs = min(nslabs - i, (unsigned long)IO_TLB_SEGSIZE); | |
151 | ||
152 | do { | |
153 | rc = xen_create_contiguous_region( | |
154 | (unsigned long)buf + (i << IO_TLB_SHIFT), | |
155 | get_order(slabs << IO_TLB_SHIFT), | |
69908907 | 156 | dma_bits, &dma_handle); |
b097186f KRW |
157 | } while (rc && dma_bits++ < max_dma_bits); |
158 | if (rc) | |
159 | return rc; | |
160 | ||
161 | i += slabs; | |
162 | } while (i < nslabs); | |
163 | return 0; | |
164 | } | |
1cef36a5 KRW |
165 | static unsigned long xen_set_nslabs(unsigned long nr_tbl) |
166 | { | |
167 | if (!nr_tbl) { | |
168 | xen_io_tlb_nslabs = (64 * 1024 * 1024 >> IO_TLB_SHIFT); | |
169 | xen_io_tlb_nslabs = ALIGN(xen_io_tlb_nslabs, IO_TLB_SEGSIZE); | |
170 | } else | |
171 | xen_io_tlb_nslabs = nr_tbl; | |
b097186f | 172 | |
1cef36a5 KRW |
173 | return xen_io_tlb_nslabs << IO_TLB_SHIFT; |
174 | } | |
b097186f | 175 | |
5bab7864 KRW |
176 | enum xen_swiotlb_err { |
177 | XEN_SWIOTLB_UNKNOWN = 0, | |
178 | XEN_SWIOTLB_ENOMEM, | |
179 | XEN_SWIOTLB_EFIXUP | |
180 | }; | |
181 | ||
182 | static const char *xen_swiotlb_error(enum xen_swiotlb_err err) | |
183 | { | |
184 | switch (err) { | |
185 | case XEN_SWIOTLB_ENOMEM: | |
186 | return "Cannot allocate Xen-SWIOTLB buffer\n"; | |
187 | case XEN_SWIOTLB_EFIXUP: | |
188 | return "Failed to get contiguous memory for DMA from Xen!\n"\ | |
189 | "You either: don't have the permissions, do not have"\ | |
190 | " enough free memory under 4GB, or the hypervisor memory"\ | |
191 | " is too fragmented!"; | |
192 | default: | |
193 | break; | |
194 | } | |
195 | return ""; | |
196 | } | |
b8277600 | 197 | int __ref xen_swiotlb_init(int verbose, bool early) |
b097186f | 198 | { |
b8277600 | 199 | unsigned long bytes, order; |
f4b2f07b | 200 | int rc = -ENOMEM; |
5bab7864 | 201 | enum xen_swiotlb_err m_ret = XEN_SWIOTLB_UNKNOWN; |
f4b2f07b | 202 | unsigned int repeat = 3; |
5f98ecdb | 203 | |
1cef36a5 | 204 | xen_io_tlb_nslabs = swiotlb_nr_tbl(); |
f4b2f07b | 205 | retry: |
1cef36a5 | 206 | bytes = xen_set_nslabs(xen_io_tlb_nslabs); |
b8277600 | 207 | order = get_order(xen_io_tlb_nslabs << IO_TLB_SHIFT); |
b097186f KRW |
208 | /* |
209 | * Get IO TLB memory from any location. | |
210 | */ | |
b8277600 KRW |
211 | if (early) |
212 | xen_io_tlb_start = alloc_bootmem_pages(PAGE_ALIGN(bytes)); | |
213 | else { | |
214 | #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT)) | |
215 | #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT) | |
216 | while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) { | |
217 | xen_io_tlb_start = (void *)__get_free_pages(__GFP_NOWARN, order); | |
218 | if (xen_io_tlb_start) | |
219 | break; | |
220 | order--; | |
221 | } | |
222 | if (order != get_order(bytes)) { | |
283c0972 JP |
223 | pr_warn("Warning: only able to allocate %ld MB for software IO TLB\n", |
224 | (PAGE_SIZE << order) >> 20); | |
b8277600 KRW |
225 | xen_io_tlb_nslabs = SLABS_PER_PAGE << order; |
226 | bytes = xen_io_tlb_nslabs << IO_TLB_SHIFT; | |
227 | } | |
228 | } | |
f4b2f07b | 229 | if (!xen_io_tlb_start) { |
5bab7864 | 230 | m_ret = XEN_SWIOTLB_ENOMEM; |
f4b2f07b KRW |
231 | goto error; |
232 | } | |
b097186f KRW |
233 | xen_io_tlb_end = xen_io_tlb_start + bytes; |
234 | /* | |
235 | * And replace that memory with pages under 4GB. | |
236 | */ | |
237 | rc = xen_swiotlb_fixup(xen_io_tlb_start, | |
238 | bytes, | |
239 | xen_io_tlb_nslabs); | |
f4b2f07b | 240 | if (rc) { |
b8277600 KRW |
241 | if (early) |
242 | free_bootmem(__pa(xen_io_tlb_start), PAGE_ALIGN(bytes)); | |
243 | else { | |
244 | free_pages((unsigned long)xen_io_tlb_start, order); | |
245 | xen_io_tlb_start = NULL; | |
246 | } | |
5bab7864 | 247 | m_ret = XEN_SWIOTLB_EFIXUP; |
b097186f | 248 | goto error; |
f4b2f07b | 249 | } |
b097186f | 250 | start_dma_addr = xen_virt_to_bus(xen_io_tlb_start); |
c468bdee | 251 | if (early) { |
ac2cbab2 YL |
252 | if (swiotlb_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs, |
253 | verbose)) | |
254 | panic("Cannot allocate SWIOTLB buffer"); | |
c468bdee KRW |
255 | rc = 0; |
256 | } else | |
b8277600 KRW |
257 | rc = swiotlb_late_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs); |
258 | return rc; | |
b097186f | 259 | error: |
f4b2f07b KRW |
260 | if (repeat--) { |
261 | xen_io_tlb_nslabs = max(1024UL, /* Min is 2MB */ | |
262 | (xen_io_tlb_nslabs >> 1)); | |
283c0972 JP |
263 | pr_info("Lowering to %luMB\n", |
264 | (xen_io_tlb_nslabs << IO_TLB_SHIFT) >> 20); | |
f4b2f07b KRW |
265 | goto retry; |
266 | } | |
283c0972 | 267 | pr_err("%s (rc:%d)\n", xen_swiotlb_error(m_ret), rc); |
b8277600 KRW |
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; | |
b097186f | 273 | } |
b097186f KRW |
274 | void * |
275 | xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size, | |
baa676fc AP |
276 | dma_addr_t *dma_handle, gfp_t flags, |
277 | struct dma_attrs *attrs) | |
b097186f KRW |
278 | { |
279 | void *ret; | |
280 | int order = get_order(size); | |
281 | u64 dma_mask = DMA_BIT_MASK(32); | |
282 | unsigned long vstart; | |
6810df88 KRW |
283 | phys_addr_t phys; |
284 | dma_addr_t dev_addr; | |
b097186f KRW |
285 | |
286 | /* | |
287 | * Ignore region specifiers - the kernel's ideas of | |
288 | * pseudo-phys memory layout has nothing to do with the | |
289 | * machine physical layout. We can't allocate highmem | |
290 | * because we can't return a pointer to it. | |
291 | */ | |
292 | flags &= ~(__GFP_DMA | __GFP_HIGHMEM); | |
293 | ||
294 | if (dma_alloc_from_coherent(hwdev, size, dma_handle, &ret)) | |
295 | return ret; | |
296 | ||
297 | vstart = __get_free_pages(flags, order); | |
298 | ret = (void *)vstart; | |
299 | ||
6810df88 KRW |
300 | if (!ret) |
301 | return ret; | |
302 | ||
b097186f | 303 | if (hwdev && hwdev->coherent_dma_mask) |
b5031ed1 | 304 | dma_mask = dma_alloc_coherent_mask(hwdev, flags); |
b097186f | 305 | |
6810df88 KRW |
306 | phys = virt_to_phys(ret); |
307 | dev_addr = xen_phys_to_bus(phys); | |
308 | if (((dev_addr + size - 1 <= dma_mask)) && | |
309 | !range_straddles_page_boundary(phys, size)) | |
310 | *dma_handle = dev_addr; | |
311 | else { | |
b097186f | 312 | if (xen_create_contiguous_region(vstart, order, |
69908907 | 313 | fls64(dma_mask), dma_handle) != 0) { |
b097186f KRW |
314 | free_pages(vstart, order); |
315 | return NULL; | |
316 | } | |
b097186f | 317 | } |
6810df88 | 318 | memset(ret, 0, size); |
b097186f KRW |
319 | return ret; |
320 | } | |
321 | EXPORT_SYMBOL_GPL(xen_swiotlb_alloc_coherent); | |
322 | ||
323 | void | |
324 | xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr, | |
baa676fc | 325 | dma_addr_t dev_addr, struct dma_attrs *attrs) |
b097186f KRW |
326 | { |
327 | int order = get_order(size); | |
6810df88 KRW |
328 | phys_addr_t phys; |
329 | u64 dma_mask = DMA_BIT_MASK(32); | |
b097186f KRW |
330 | |
331 | if (dma_release_from_coherent(hwdev, order, vaddr)) | |
332 | return; | |
333 | ||
6810df88 KRW |
334 | if (hwdev && hwdev->coherent_dma_mask) |
335 | dma_mask = hwdev->coherent_dma_mask; | |
336 | ||
337 | phys = virt_to_phys(vaddr); | |
338 | ||
339 | if (((dev_addr + size - 1 > dma_mask)) || | |
340 | range_straddles_page_boundary(phys, size)) | |
341 | xen_destroy_contiguous_region((unsigned long)vaddr, order); | |
342 | ||
b097186f KRW |
343 | free_pages((unsigned long)vaddr, order); |
344 | } | |
345 | EXPORT_SYMBOL_GPL(xen_swiotlb_free_coherent); | |
346 | ||
347 | ||
348 | /* | |
349 | * Map a single buffer of the indicated size for DMA in streaming mode. The | |
350 | * physical address to use is returned. | |
351 | * | |
352 | * Once the device is given the dma address, the device owns this memory until | |
353 | * either xen_swiotlb_unmap_page or xen_swiotlb_dma_sync_single is performed. | |
354 | */ | |
355 | dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page, | |
356 | unsigned long offset, size_t size, | |
357 | enum dma_data_direction dir, | |
358 | struct dma_attrs *attrs) | |
359 | { | |
e05ed4d1 | 360 | phys_addr_t map, phys = page_to_phys(page) + offset; |
b097186f | 361 | dma_addr_t dev_addr = xen_phys_to_bus(phys); |
b097186f KRW |
362 | |
363 | BUG_ON(dir == DMA_NONE); | |
364 | /* | |
365 | * If the address happens to be in the device's DMA window, | |
366 | * we can safely return the device addr and not worry about bounce | |
367 | * buffering it. | |
368 | */ | |
369 | if (dma_capable(dev, dev_addr, size) && | |
370 | !range_straddles_page_boundary(phys, size) && !swiotlb_force) | |
371 | return dev_addr; | |
372 | ||
373 | /* | |
374 | * Oh well, have to allocate and map a bounce buffer. | |
375 | */ | |
376 | map = swiotlb_tbl_map_single(dev, start_dma_addr, phys, size, dir); | |
e05ed4d1 | 377 | if (map == SWIOTLB_MAP_ERROR) |
b097186f KRW |
378 | return DMA_ERROR_CODE; |
379 | ||
e05ed4d1 | 380 | dev_addr = xen_phys_to_bus(map); |
b097186f KRW |
381 | |
382 | /* | |
383 | * Ensure that the address returned is DMA'ble | |
384 | */ | |
ab2a47bd | 385 | if (!dma_capable(dev, dev_addr, size)) { |
61ca08c3 | 386 | swiotlb_tbl_unmap_single(dev, map, size, dir); |
ab2a47bd KRW |
387 | dev_addr = 0; |
388 | } | |
b097186f KRW |
389 | return dev_addr; |
390 | } | |
391 | EXPORT_SYMBOL_GPL(xen_swiotlb_map_page); | |
392 | ||
393 | /* | |
394 | * Unmap a single streaming mode DMA translation. The dma_addr and size must | |
395 | * match what was provided for in a previous xen_swiotlb_map_page call. All | |
396 | * other usages are undefined. | |
397 | * | |
398 | * After this call, reads by the cpu to the buffer are guaranteed to see | |
399 | * whatever the device wrote there. | |
400 | */ | |
401 | static void xen_unmap_single(struct device *hwdev, dma_addr_t dev_addr, | |
402 | size_t size, enum dma_data_direction dir) | |
403 | { | |
404 | phys_addr_t paddr = xen_bus_to_phys(dev_addr); | |
405 | ||
406 | BUG_ON(dir == DMA_NONE); | |
407 | ||
408 | /* NOTE: We use dev_addr here, not paddr! */ | |
409 | if (is_xen_swiotlb_buffer(dev_addr)) { | |
61ca08c3 | 410 | swiotlb_tbl_unmap_single(hwdev, paddr, size, dir); |
b097186f KRW |
411 | return; |
412 | } | |
413 | ||
414 | if (dir != DMA_FROM_DEVICE) | |
415 | return; | |
416 | ||
417 | /* | |
418 | * phys_to_virt doesn't work with hihgmem page but we could | |
419 | * call dma_mark_clean() with hihgmem page here. However, we | |
420 | * are fine since dma_mark_clean() is null on POWERPC. We can | |
421 | * make dma_mark_clean() take a physical address if necessary. | |
422 | */ | |
423 | dma_mark_clean(phys_to_virt(paddr), size); | |
424 | } | |
425 | ||
426 | void xen_swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr, | |
427 | size_t size, enum dma_data_direction dir, | |
428 | struct dma_attrs *attrs) | |
429 | { | |
430 | xen_unmap_single(hwdev, dev_addr, size, dir); | |
431 | } | |
432 | EXPORT_SYMBOL_GPL(xen_swiotlb_unmap_page); | |
433 | ||
434 | /* | |
435 | * Make physical memory consistent for a single streaming mode DMA translation | |
436 | * after a transfer. | |
437 | * | |
438 | * If you perform a xen_swiotlb_map_page() but wish to interrogate the buffer | |
439 | * using the cpu, yet do not wish to teardown the dma mapping, you must | |
440 | * call this function before doing so. At the next point you give the dma | |
441 | * address back to the card, you must first perform a | |
442 | * xen_swiotlb_dma_sync_for_device, and then the device again owns the buffer | |
443 | */ | |
444 | static void | |
445 | xen_swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr, | |
446 | size_t size, enum dma_data_direction dir, | |
447 | enum dma_sync_target target) | |
448 | { | |
449 | phys_addr_t paddr = xen_bus_to_phys(dev_addr); | |
450 | ||
451 | BUG_ON(dir == DMA_NONE); | |
452 | ||
453 | /* NOTE: We use dev_addr here, not paddr! */ | |
454 | if (is_xen_swiotlb_buffer(dev_addr)) { | |
fbfda893 | 455 | swiotlb_tbl_sync_single(hwdev, paddr, size, dir, target); |
b097186f KRW |
456 | return; |
457 | } | |
458 | ||
459 | if (dir != DMA_FROM_DEVICE) | |
460 | return; | |
461 | ||
462 | dma_mark_clean(phys_to_virt(paddr), size); | |
463 | } | |
464 | ||
465 | void | |
466 | xen_swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr, | |
467 | size_t size, enum dma_data_direction dir) | |
468 | { | |
469 | xen_swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_CPU); | |
470 | } | |
471 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_single_for_cpu); | |
472 | ||
473 | void | |
474 | xen_swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr, | |
475 | size_t size, enum dma_data_direction dir) | |
476 | { | |
477 | xen_swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_DEVICE); | |
478 | } | |
479 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_single_for_device); | |
480 | ||
481 | /* | |
482 | * Map a set of buffers described by scatterlist in streaming mode for DMA. | |
483 | * This is the scatter-gather version of the above xen_swiotlb_map_page | |
484 | * interface. Here the scatter gather list elements are each tagged with the | |
485 | * appropriate dma address and length. They are obtained via | |
486 | * sg_dma_{address,length}(SG). | |
487 | * | |
488 | * NOTE: An implementation may be able to use a smaller number of | |
489 | * DMA address/length pairs than there are SG table elements. | |
490 | * (for example via virtual mapping capabilities) | |
491 | * The routine returns the number of addr/length pairs actually | |
492 | * used, at most nents. | |
493 | * | |
494 | * Device ownership issues as mentioned above for xen_swiotlb_map_page are the | |
495 | * same here. | |
496 | */ | |
497 | int | |
498 | xen_swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, | |
499 | int nelems, enum dma_data_direction dir, | |
500 | struct dma_attrs *attrs) | |
501 | { | |
502 | struct scatterlist *sg; | |
503 | int i; | |
504 | ||
505 | BUG_ON(dir == DMA_NONE); | |
506 | ||
507 | for_each_sg(sgl, sg, nelems, i) { | |
508 | phys_addr_t paddr = sg_phys(sg); | |
509 | dma_addr_t dev_addr = xen_phys_to_bus(paddr); | |
510 | ||
511 | if (swiotlb_force || | |
512 | !dma_capable(hwdev, dev_addr, sg->length) || | |
513 | range_straddles_page_boundary(paddr, sg->length)) { | |
e05ed4d1 AD |
514 | phys_addr_t map = swiotlb_tbl_map_single(hwdev, |
515 | start_dma_addr, | |
516 | sg_phys(sg), | |
517 | sg->length, | |
518 | dir); | |
519 | if (map == SWIOTLB_MAP_ERROR) { | |
b097186f KRW |
520 | /* Don't panic here, we expect map_sg users |
521 | to do proper error handling. */ | |
522 | xen_swiotlb_unmap_sg_attrs(hwdev, sgl, i, dir, | |
523 | attrs); | |
781575cd | 524 | sg_dma_len(sgl) = 0; |
b097186f KRW |
525 | return DMA_ERROR_CODE; |
526 | } | |
e05ed4d1 | 527 | sg->dma_address = xen_phys_to_bus(map); |
b097186f KRW |
528 | } else |
529 | sg->dma_address = dev_addr; | |
781575cd | 530 | sg_dma_len(sg) = sg->length; |
b097186f KRW |
531 | } |
532 | return nelems; | |
533 | } | |
534 | EXPORT_SYMBOL_GPL(xen_swiotlb_map_sg_attrs); | |
535 | ||
b097186f KRW |
536 | /* |
537 | * Unmap a set of streaming mode DMA translations. Again, cpu read rules | |
538 | * concerning calls here are the same as for swiotlb_unmap_page() above. | |
539 | */ | |
540 | void | |
541 | xen_swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl, | |
542 | int nelems, enum dma_data_direction dir, | |
543 | struct dma_attrs *attrs) | |
544 | { | |
545 | struct scatterlist *sg; | |
546 | int i; | |
547 | ||
548 | BUG_ON(dir == DMA_NONE); | |
549 | ||
550 | for_each_sg(sgl, sg, nelems, i) | |
781575cd | 551 | xen_unmap_single(hwdev, sg->dma_address, sg_dma_len(sg), dir); |
b097186f KRW |
552 | |
553 | } | |
554 | EXPORT_SYMBOL_GPL(xen_swiotlb_unmap_sg_attrs); | |
555 | ||
b097186f KRW |
556 | /* |
557 | * Make physical memory consistent for a set of streaming mode DMA translations | |
558 | * after a transfer. | |
559 | * | |
560 | * The same as swiotlb_sync_single_* but for a scatter-gather list, same rules | |
561 | * and usage. | |
562 | */ | |
563 | static void | |
564 | xen_swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl, | |
565 | int nelems, enum dma_data_direction dir, | |
566 | enum dma_sync_target target) | |
567 | { | |
568 | struct scatterlist *sg; | |
569 | int i; | |
570 | ||
571 | for_each_sg(sgl, sg, nelems, i) | |
572 | xen_swiotlb_sync_single(hwdev, sg->dma_address, | |
781575cd | 573 | sg_dma_len(sg), dir, target); |
b097186f KRW |
574 | } |
575 | ||
576 | void | |
577 | xen_swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg, | |
578 | int nelems, enum dma_data_direction dir) | |
579 | { | |
580 | xen_swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_CPU); | |
581 | } | |
582 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_sg_for_cpu); | |
583 | ||
584 | void | |
585 | xen_swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, | |
586 | int nelems, enum dma_data_direction dir) | |
587 | { | |
588 | xen_swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_DEVICE); | |
589 | } | |
590 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_sg_for_device); | |
591 | ||
592 | int | |
593 | xen_swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr) | |
594 | { | |
595 | return !dma_addr; | |
596 | } | |
597 | EXPORT_SYMBOL_GPL(xen_swiotlb_dma_mapping_error); | |
598 | ||
599 | /* | |
600 | * Return whether the given device DMA address mask can be supported | |
601 | * properly. For example, if your device can only drive the low 24-bits | |
602 | * during bus mastering, then you would pass 0x00ffffff as the mask to | |
603 | * this function. | |
604 | */ | |
605 | int | |
606 | xen_swiotlb_dma_supported(struct device *hwdev, u64 mask) | |
607 | { | |
608 | return xen_virt_to_bus(xen_io_tlb_end - 1) <= mask; | |
609 | } | |
610 | EXPORT_SYMBOL_GPL(xen_swiotlb_dma_supported); | |
eb1ddc00 SS |
611 | |
612 | int | |
613 | xen_swiotlb_set_dma_mask(struct device *dev, u64 dma_mask) | |
614 | { | |
615 | if (!dev->dma_mask || !xen_swiotlb_dma_supported(dev, dma_mask)) | |
616 | return -EIO; | |
617 | ||
618 | *dev->dma_mask = dma_mask; | |
619 | ||
620 | return 0; | |
621 | } | |
622 | EXPORT_SYMBOL_GPL(xen_swiotlb_set_dma_mask); |