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 | ||
2013288f | 38 | #include <linux/memblock.h> |
ea8c64ac | 39 | #include <linux/dma-direct.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> |
2b2b614d | 45 | |
83862ccf | 46 | #include <asm/dma-mapping.h> |
1b65c4e5 | 47 | #include <asm/xen/page-coherent.h> |
e1d8f62a | 48 | |
2b2b614d | 49 | #include <trace/events/swiotlb.h> |
b097186f KRW |
50 | /* |
51 | * Used to do a quick range check in swiotlb_tbl_unmap_single and | |
52 | * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this | |
53 | * API. | |
54 | */ | |
55 | ||
56 | static char *xen_io_tlb_start, *xen_io_tlb_end; | |
57 | static unsigned long xen_io_tlb_nslabs; | |
58 | /* | |
59 | * Quick lookup value of the bus address of the IOTLB. | |
60 | */ | |
61 | ||
b8b0f559 | 62 | static u64 start_dma_addr; |
b097186f | 63 | |
e17b2f11 | 64 | /* |
9435cce8 | 65 | * Both of these functions should avoid XEN_PFN_PHYS because phys_addr_t |
e17b2f11 IC |
66 | * can be 32bit when dma_addr_t is 64bit leading to a loss in |
67 | * information if the shift is done before casting to 64bit. | |
68 | */ | |
6b42a7ea | 69 | static inline dma_addr_t xen_phys_to_bus(phys_addr_t paddr) |
b097186f | 70 | { |
9435cce8 JG |
71 | unsigned long bfn = pfn_to_bfn(XEN_PFN_DOWN(paddr)); |
72 | dma_addr_t dma = (dma_addr_t)bfn << XEN_PAGE_SHIFT; | |
e17b2f11 | 73 | |
9435cce8 | 74 | dma |= paddr & ~XEN_PAGE_MASK; |
e17b2f11 IC |
75 | |
76 | return dma; | |
b097186f KRW |
77 | } |
78 | ||
6b42a7ea | 79 | static inline phys_addr_t xen_bus_to_phys(dma_addr_t baddr) |
b097186f | 80 | { |
9435cce8 JG |
81 | unsigned long xen_pfn = bfn_to_pfn(XEN_PFN_DOWN(baddr)); |
82 | dma_addr_t dma = (dma_addr_t)xen_pfn << XEN_PAGE_SHIFT; | |
e17b2f11 IC |
83 | phys_addr_t paddr = dma; |
84 | ||
9435cce8 | 85 | paddr |= baddr & ~XEN_PAGE_MASK; |
e17b2f11 IC |
86 | |
87 | return paddr; | |
b097186f KRW |
88 | } |
89 | ||
6b42a7ea | 90 | static inline dma_addr_t xen_virt_to_bus(void *address) |
b097186f KRW |
91 | { |
92 | return xen_phys_to_bus(virt_to_phys(address)); | |
93 | } | |
94 | ||
9435cce8 | 95 | static int check_pages_physically_contiguous(unsigned long xen_pfn, |
b097186f KRW |
96 | unsigned int offset, |
97 | size_t length) | |
98 | { | |
32e09870 | 99 | unsigned long next_bfn; |
b097186f KRW |
100 | int i; |
101 | int nr_pages; | |
102 | ||
9435cce8 JG |
103 | next_bfn = pfn_to_bfn(xen_pfn); |
104 | nr_pages = (offset + length + XEN_PAGE_SIZE-1) >> XEN_PAGE_SHIFT; | |
b097186f KRW |
105 | |
106 | for (i = 1; i < nr_pages; i++) { | |
9435cce8 | 107 | if (pfn_to_bfn(++xen_pfn) != ++next_bfn) |
b097186f KRW |
108 | return 0; |
109 | } | |
110 | return 1; | |
111 | } | |
112 | ||
6b42a7ea | 113 | static inline int range_straddles_page_boundary(phys_addr_t p, size_t size) |
b097186f | 114 | { |
9435cce8 JG |
115 | unsigned long xen_pfn = XEN_PFN_DOWN(p); |
116 | unsigned int offset = p & ~XEN_PAGE_MASK; | |
b097186f | 117 | |
9435cce8 | 118 | if (offset + size <= XEN_PAGE_SIZE) |
b097186f | 119 | return 0; |
9435cce8 | 120 | if (check_pages_physically_contiguous(xen_pfn, offset, size)) |
b097186f KRW |
121 | return 0; |
122 | return 1; | |
123 | } | |
124 | ||
125 | static int is_xen_swiotlb_buffer(dma_addr_t dma_addr) | |
126 | { | |
9435cce8 JG |
127 | unsigned long bfn = XEN_PFN_DOWN(dma_addr); |
128 | unsigned long xen_pfn = bfn_to_local_pfn(bfn); | |
129 | phys_addr_t paddr = XEN_PFN_PHYS(xen_pfn); | |
b097186f KRW |
130 | |
131 | /* If the address is outside our domain, it CAN | |
132 | * have the same virtual address as another address | |
133 | * in our domain. Therefore _only_ check address within our domain. | |
134 | */ | |
9435cce8 | 135 | if (pfn_valid(PFN_DOWN(paddr))) { |
b097186f KRW |
136 | return paddr >= virt_to_phys(xen_io_tlb_start) && |
137 | paddr < virt_to_phys(xen_io_tlb_end); | |
138 | } | |
139 | return 0; | |
140 | } | |
141 | ||
142 | static int max_dma_bits = 32; | |
143 | ||
144 | static int | |
145 | xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs) | |
146 | { | |
147 | int i, rc; | |
148 | int dma_bits; | |
69908907 | 149 | dma_addr_t dma_handle; |
1b65c4e5 | 150 | phys_addr_t p = virt_to_phys(buf); |
b097186f KRW |
151 | |
152 | dma_bits = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT) + PAGE_SHIFT; | |
153 | ||
154 | i = 0; | |
155 | do { | |
156 | int slabs = min(nslabs - i, (unsigned long)IO_TLB_SEGSIZE); | |
157 | ||
158 | do { | |
159 | rc = xen_create_contiguous_region( | |
1b65c4e5 | 160 | p + (i << IO_TLB_SHIFT), |
b097186f | 161 | get_order(slabs << IO_TLB_SHIFT), |
69908907 | 162 | dma_bits, &dma_handle); |
b097186f KRW |
163 | } while (rc && dma_bits++ < max_dma_bits); |
164 | if (rc) | |
165 | return rc; | |
166 | ||
167 | i += slabs; | |
168 | } while (i < nslabs); | |
169 | return 0; | |
170 | } | |
1cef36a5 KRW |
171 | static unsigned long xen_set_nslabs(unsigned long nr_tbl) |
172 | { | |
173 | if (!nr_tbl) { | |
174 | xen_io_tlb_nslabs = (64 * 1024 * 1024 >> IO_TLB_SHIFT); | |
175 | xen_io_tlb_nslabs = ALIGN(xen_io_tlb_nslabs, IO_TLB_SEGSIZE); | |
176 | } else | |
177 | xen_io_tlb_nslabs = nr_tbl; | |
b097186f | 178 | |
1cef36a5 KRW |
179 | return xen_io_tlb_nslabs << IO_TLB_SHIFT; |
180 | } | |
b097186f | 181 | |
5bab7864 KRW |
182 | enum xen_swiotlb_err { |
183 | XEN_SWIOTLB_UNKNOWN = 0, | |
184 | XEN_SWIOTLB_ENOMEM, | |
185 | XEN_SWIOTLB_EFIXUP | |
186 | }; | |
187 | ||
188 | static const char *xen_swiotlb_error(enum xen_swiotlb_err err) | |
189 | { | |
190 | switch (err) { | |
191 | case XEN_SWIOTLB_ENOMEM: | |
192 | return "Cannot allocate Xen-SWIOTLB buffer\n"; | |
193 | case XEN_SWIOTLB_EFIXUP: | |
194 | return "Failed to get contiguous memory for DMA from Xen!\n"\ | |
195 | "You either: don't have the permissions, do not have"\ | |
196 | " enough free memory under 4GB, or the hypervisor memory"\ | |
197 | " is too fragmented!"; | |
198 | default: | |
199 | break; | |
200 | } | |
201 | return ""; | |
202 | } | |
b8277600 | 203 | int __ref xen_swiotlb_init(int verbose, bool early) |
b097186f | 204 | { |
b8277600 | 205 | unsigned long bytes, order; |
f4b2f07b | 206 | int rc = -ENOMEM; |
5bab7864 | 207 | enum xen_swiotlb_err m_ret = XEN_SWIOTLB_UNKNOWN; |
f4b2f07b | 208 | unsigned int repeat = 3; |
5f98ecdb | 209 | |
1cef36a5 | 210 | xen_io_tlb_nslabs = swiotlb_nr_tbl(); |
f4b2f07b | 211 | retry: |
1cef36a5 | 212 | bytes = xen_set_nslabs(xen_io_tlb_nslabs); |
b8277600 | 213 | order = get_order(xen_io_tlb_nslabs << IO_TLB_SHIFT); |
b097186f KRW |
214 | /* |
215 | * Get IO TLB memory from any location. | |
216 | */ | |
8a7f97b9 | 217 | if (early) { |
15c3c114 MR |
218 | xen_io_tlb_start = memblock_alloc(PAGE_ALIGN(bytes), |
219 | PAGE_SIZE); | |
8a7f97b9 MR |
220 | if (!xen_io_tlb_start) |
221 | panic("%s: Failed to allocate %lu bytes align=0x%lx\n", | |
222 | __func__, PAGE_ALIGN(bytes), PAGE_SIZE); | |
223 | } else { | |
b8277600 KRW |
224 | #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT)) |
225 | #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT) | |
226 | while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) { | |
8746515d | 227 | xen_io_tlb_start = (void *)xen_get_swiotlb_free_pages(order); |
b8277600 KRW |
228 | if (xen_io_tlb_start) |
229 | break; | |
230 | order--; | |
231 | } | |
232 | if (order != get_order(bytes)) { | |
283c0972 JP |
233 | pr_warn("Warning: only able to allocate %ld MB for software IO TLB\n", |
234 | (PAGE_SIZE << order) >> 20); | |
b8277600 KRW |
235 | xen_io_tlb_nslabs = SLABS_PER_PAGE << order; |
236 | bytes = xen_io_tlb_nslabs << IO_TLB_SHIFT; | |
237 | } | |
238 | } | |
f4b2f07b | 239 | if (!xen_io_tlb_start) { |
5bab7864 | 240 | m_ret = XEN_SWIOTLB_ENOMEM; |
f4b2f07b KRW |
241 | goto error; |
242 | } | |
b097186f KRW |
243 | xen_io_tlb_end = xen_io_tlb_start + bytes; |
244 | /* | |
245 | * And replace that memory with pages under 4GB. | |
246 | */ | |
247 | rc = xen_swiotlb_fixup(xen_io_tlb_start, | |
248 | bytes, | |
249 | xen_io_tlb_nslabs); | |
f4b2f07b | 250 | if (rc) { |
b8277600 | 251 | if (early) |
2013288f MR |
252 | memblock_free(__pa(xen_io_tlb_start), |
253 | PAGE_ALIGN(bytes)); | |
b8277600 KRW |
254 | else { |
255 | free_pages((unsigned long)xen_io_tlb_start, order); | |
256 | xen_io_tlb_start = NULL; | |
257 | } | |
5bab7864 | 258 | m_ret = XEN_SWIOTLB_EFIXUP; |
b097186f | 259 | goto error; |
f4b2f07b | 260 | } |
b097186f | 261 | start_dma_addr = xen_virt_to_bus(xen_io_tlb_start); |
c468bdee | 262 | if (early) { |
ac2cbab2 YL |
263 | if (swiotlb_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs, |
264 | verbose)) | |
265 | panic("Cannot allocate SWIOTLB buffer"); | |
c468bdee KRW |
266 | rc = 0; |
267 | } else | |
b8277600 | 268 | rc = swiotlb_late_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs); |
7453c549 KRW |
269 | |
270 | if (!rc) | |
271 | swiotlb_set_max_segment(PAGE_SIZE); | |
272 | ||
b8277600 | 273 | return rc; |
b097186f | 274 | error: |
f4b2f07b KRW |
275 | if (repeat--) { |
276 | xen_io_tlb_nslabs = max(1024UL, /* Min is 2MB */ | |
277 | (xen_io_tlb_nslabs >> 1)); | |
283c0972 JP |
278 | pr_info("Lowering to %luMB\n", |
279 | (xen_io_tlb_nslabs << IO_TLB_SHIFT) >> 20); | |
f4b2f07b KRW |
280 | goto retry; |
281 | } | |
283c0972 | 282 | pr_err("%s (rc:%d)\n", xen_swiotlb_error(m_ret), rc); |
b8277600 KRW |
283 | if (early) |
284 | panic("%s (rc:%d)", xen_swiotlb_error(m_ret), rc); | |
285 | else | |
286 | free_pages((unsigned long)xen_io_tlb_start, order); | |
287 | return rc; | |
b097186f | 288 | } |
dceb1a68 CH |
289 | |
290 | static void * | |
b097186f | 291 | xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size, |
baa676fc | 292 | dma_addr_t *dma_handle, gfp_t flags, |
00085f1e | 293 | unsigned long attrs) |
b097186f KRW |
294 | { |
295 | void *ret; | |
296 | int order = get_order(size); | |
297 | u64 dma_mask = DMA_BIT_MASK(32); | |
6810df88 KRW |
298 | phys_addr_t phys; |
299 | dma_addr_t dev_addr; | |
b097186f KRW |
300 | |
301 | /* | |
302 | * Ignore region specifiers - the kernel's ideas of | |
303 | * pseudo-phys memory layout has nothing to do with the | |
304 | * machine physical layout. We can't allocate highmem | |
305 | * because we can't return a pointer to it. | |
306 | */ | |
307 | flags &= ~(__GFP_DMA | __GFP_HIGHMEM); | |
308 | ||
7250f422 JJ |
309 | /* Convert the size to actually allocated. */ |
310 | size = 1UL << (order + XEN_PAGE_SHIFT); | |
311 | ||
1b65c4e5 SS |
312 | /* On ARM this function returns an ioremap'ped virtual address for |
313 | * which virt_to_phys doesn't return the corresponding physical | |
314 | * address. In fact on ARM virt_to_phys only works for kernel direct | |
315 | * mapped RAM memory. Also see comment below. | |
316 | */ | |
317 | ret = xen_alloc_coherent_pages(hwdev, size, dma_handle, flags, attrs); | |
b097186f | 318 | |
6810df88 KRW |
319 | if (!ret) |
320 | return ret; | |
321 | ||
b097186f | 322 | if (hwdev && hwdev->coherent_dma_mask) |
038d07a2 | 323 | dma_mask = hwdev->coherent_dma_mask; |
b097186f | 324 | |
1b65c4e5 SS |
325 | /* At this point dma_handle is the physical address, next we are |
326 | * going to set it to the machine address. | |
327 | * Do not use virt_to_phys(ret) because on ARM it doesn't correspond | |
328 | * to *dma_handle. */ | |
329 | phys = *dma_handle; | |
6810df88 KRW |
330 | dev_addr = xen_phys_to_bus(phys); |
331 | if (((dev_addr + size - 1 <= dma_mask)) && | |
332 | !range_straddles_page_boundary(phys, size)) | |
333 | *dma_handle = dev_addr; | |
334 | else { | |
1b65c4e5 | 335 | if (xen_create_contiguous_region(phys, order, |
69908907 | 336 | fls64(dma_mask), dma_handle) != 0) { |
1b65c4e5 | 337 | xen_free_coherent_pages(hwdev, size, ret, (dma_addr_t)phys, attrs); |
b097186f KRW |
338 | return NULL; |
339 | } | |
b097186f | 340 | } |
6810df88 | 341 | memset(ret, 0, size); |
b097186f KRW |
342 | return ret; |
343 | } | |
b097186f | 344 | |
dceb1a68 | 345 | static void |
b097186f | 346 | xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr, |
00085f1e | 347 | dma_addr_t dev_addr, unsigned long attrs) |
b097186f KRW |
348 | { |
349 | int order = get_order(size); | |
6810df88 KRW |
350 | phys_addr_t phys; |
351 | u64 dma_mask = DMA_BIT_MASK(32); | |
b097186f | 352 | |
6810df88 KRW |
353 | if (hwdev && hwdev->coherent_dma_mask) |
354 | dma_mask = hwdev->coherent_dma_mask; | |
355 | ||
1b65c4e5 SS |
356 | /* do not use virt_to_phys because on ARM it doesn't return you the |
357 | * physical address */ | |
358 | phys = xen_bus_to_phys(dev_addr); | |
6810df88 | 359 | |
7250f422 JJ |
360 | /* Convert the size to actually allocated. */ |
361 | size = 1UL << (order + XEN_PAGE_SHIFT); | |
362 | ||
4855c92d | 363 | if (((dev_addr + size - 1 <= dma_mask)) || |
6810df88 | 364 | range_straddles_page_boundary(phys, size)) |
1b65c4e5 | 365 | xen_destroy_contiguous_region(phys, order); |
6810df88 | 366 | |
1b65c4e5 | 367 | xen_free_coherent_pages(hwdev, size, vaddr, (dma_addr_t)phys, attrs); |
b097186f | 368 | } |
b097186f KRW |
369 | |
370 | /* | |
371 | * Map a single buffer of the indicated size for DMA in streaming mode. The | |
372 | * physical address to use is returned. | |
373 | * | |
374 | * Once the device is given the dma address, the device owns this memory until | |
375 | * either xen_swiotlb_unmap_page or xen_swiotlb_dma_sync_single is performed. | |
376 | */ | |
dceb1a68 | 377 | static dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page, |
b097186f KRW |
378 | unsigned long offset, size_t size, |
379 | enum dma_data_direction dir, | |
00085f1e | 380 | unsigned long attrs) |
b097186f | 381 | { |
e05ed4d1 | 382 | phys_addr_t map, phys = page_to_phys(page) + offset; |
b097186f | 383 | dma_addr_t dev_addr = xen_phys_to_bus(phys); |
b097186f KRW |
384 | |
385 | BUG_ON(dir == DMA_NONE); | |
386 | /* | |
387 | * If the address happens to be in the device's DMA window, | |
388 | * we can safely return the device addr and not worry about bounce | |
389 | * buffering it. | |
390 | */ | |
391 | if (dma_capable(dev, dev_addr, size) && | |
a4dba130 | 392 | !range_straddles_page_boundary(phys, size) && |
291be10f | 393 | !xen_arch_need_swiotlb(dev, phys, dev_addr) && |
063b8271 CH |
394 | swiotlb_force != SWIOTLB_FORCE) |
395 | goto done; | |
b097186f KRW |
396 | |
397 | /* | |
398 | * Oh well, have to allocate and map a bounce buffer. | |
399 | */ | |
2b2b614d ZK |
400 | trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force); |
401 | ||
0443fa00 AD |
402 | map = swiotlb_tbl_map_single(dev, start_dma_addr, phys, size, dir, |
403 | attrs); | |
b907e205 | 404 | if (map == DMA_MAPPING_ERROR) |
a4abe0ad | 405 | return DMA_MAPPING_ERROR; |
b097186f | 406 | |
f1225ee4 | 407 | dev_addr = xen_phys_to_bus(map); |
b097186f KRW |
408 | |
409 | /* | |
410 | * Ensure that the address returned is DMA'ble | |
411 | */ | |
063b8271 CH |
412 | if (unlikely(!dma_capable(dev, dev_addr, size))) { |
413 | swiotlb_tbl_unmap_single(dev, map, size, dir, | |
414 | attrs | DMA_ATTR_SKIP_CPU_SYNC); | |
415 | return DMA_MAPPING_ERROR; | |
416 | } | |
76418421 | 417 | |
063b8271 CH |
418 | page = pfn_to_page(map >> PAGE_SHIFT); |
419 | offset = map & ~PAGE_MASK; | |
420 | done: | |
421 | /* | |
422 | * we are not interested in the dma_addr returned by xen_dma_map_page, | |
423 | * only in the potential cache flushes executed by the function. | |
424 | */ | |
425 | xen_dma_map_page(dev, page, dev_addr, offset, size, dir, attrs); | |
426 | return dev_addr; | |
b097186f | 427 | } |
b097186f KRW |
428 | |
429 | /* | |
430 | * Unmap a single streaming mode DMA translation. The dma_addr and size must | |
431 | * match what was provided for in a previous xen_swiotlb_map_page call. All | |
432 | * other usages are undefined. | |
433 | * | |
434 | * After this call, reads by the cpu to the buffer are guaranteed to see | |
435 | * whatever the device wrote there. | |
436 | */ | |
437 | static void xen_unmap_single(struct device *hwdev, dma_addr_t dev_addr, | |
6cf05463 | 438 | size_t size, enum dma_data_direction dir, |
00085f1e | 439 | unsigned long attrs) |
b097186f KRW |
440 | { |
441 | phys_addr_t paddr = xen_bus_to_phys(dev_addr); | |
442 | ||
443 | BUG_ON(dir == DMA_NONE); | |
444 | ||
d6883e6f | 445 | xen_dma_unmap_page(hwdev, dev_addr, size, dir, attrs); |
6cf05463 | 446 | |
b097186f | 447 | /* NOTE: We use dev_addr here, not paddr! */ |
68c60834 | 448 | if (is_xen_swiotlb_buffer(dev_addr)) |
0443fa00 | 449 | swiotlb_tbl_unmap_single(hwdev, paddr, size, dir, attrs); |
b097186f KRW |
450 | } |
451 | ||
dceb1a68 | 452 | static void xen_swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr, |
b097186f | 453 | size_t size, enum dma_data_direction dir, |
00085f1e | 454 | unsigned long attrs) |
b097186f | 455 | { |
6cf05463 | 456 | xen_unmap_single(hwdev, dev_addr, size, dir, attrs); |
b097186f | 457 | } |
b097186f | 458 | |
b097186f | 459 | static void |
2e12dcee CH |
460 | xen_swiotlb_sync_single_for_cpu(struct device *dev, dma_addr_t dma_addr, |
461 | size_t size, enum dma_data_direction dir) | |
b097186f | 462 | { |
2e12dcee | 463 | phys_addr_t paddr = xen_bus_to_phys(dma_addr); |
6cf05463 | 464 | |
2e12dcee | 465 | xen_dma_sync_single_for_cpu(dev, dma_addr, size, dir); |
6cf05463 | 466 | |
2e12dcee CH |
467 | if (is_xen_swiotlb_buffer(dma_addr)) |
468 | swiotlb_tbl_sync_single(dev, paddr, size, dir, SYNC_FOR_CPU); | |
b097186f KRW |
469 | } |
470 | ||
2e12dcee CH |
471 | static void |
472 | xen_swiotlb_sync_single_for_device(struct device *dev, dma_addr_t dma_addr, | |
473 | size_t size, enum dma_data_direction dir) | |
b097186f | 474 | { |
2e12dcee | 475 | phys_addr_t paddr = xen_bus_to_phys(dma_addr); |
b097186f | 476 | |
2e12dcee CH |
477 | if (is_xen_swiotlb_buffer(dma_addr)) |
478 | swiotlb_tbl_sync_single(dev, paddr, size, dir, SYNC_FOR_DEVICE); | |
479 | ||
480 | xen_dma_sync_single_for_device(dev, dma_addr, size, dir); | |
b097186f | 481 | } |
dceb1a68 CH |
482 | |
483 | /* | |
484 | * Unmap a set of streaming mode DMA translations. Again, cpu read rules | |
485 | * concerning calls here are the same as for swiotlb_unmap_page() above. | |
486 | */ | |
487 | static void | |
aca351cc CH |
488 | xen_swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sgl, int nelems, |
489 | enum dma_data_direction dir, unsigned long attrs) | |
dceb1a68 CH |
490 | { |
491 | struct scatterlist *sg; | |
492 | int i; | |
493 | ||
494 | BUG_ON(dir == DMA_NONE); | |
495 | ||
496 | for_each_sg(sgl, sg, nelems, i) | |
497 | xen_unmap_single(hwdev, sg->dma_address, sg_dma_len(sg), dir, attrs); | |
498 | ||
499 | } | |
b097186f | 500 | |
dceb1a68 | 501 | static int |
8b35d9fe | 502 | xen_swiotlb_map_sg(struct device *dev, struct scatterlist *sgl, int nelems, |
aca351cc | 503 | enum dma_data_direction dir, unsigned long attrs) |
b097186f KRW |
504 | { |
505 | struct scatterlist *sg; | |
506 | int i; | |
507 | ||
508 | BUG_ON(dir == DMA_NONE); | |
509 | ||
510 | for_each_sg(sgl, sg, nelems, i) { | |
8b35d9fe CH |
511 | sg->dma_address = xen_swiotlb_map_page(dev, sg_page(sg), |
512 | sg->offset, sg->length, dir, attrs); | |
513 | if (sg->dma_address == DMA_MAPPING_ERROR) | |
514 | goto out_unmap; | |
781575cd | 515 | sg_dma_len(sg) = sg->length; |
b097186f | 516 | } |
8b35d9fe | 517 | |
b097186f | 518 | return nelems; |
8b35d9fe CH |
519 | out_unmap: |
520 | xen_swiotlb_unmap_sg(dev, sgl, i, dir, attrs | DMA_ATTR_SKIP_CPU_SYNC); | |
521 | sg_dma_len(sgl) = 0; | |
522 | return 0; | |
b097186f | 523 | } |
b097186f | 524 | |
b097186f | 525 | static void |
2e12dcee CH |
526 | xen_swiotlb_sync_sg_for_cpu(struct device *dev, struct scatterlist *sgl, |
527 | int nelems, enum dma_data_direction dir) | |
b097186f KRW |
528 | { |
529 | struct scatterlist *sg; | |
530 | int i; | |
531 | ||
2e12dcee CH |
532 | for_each_sg(sgl, sg, nelems, i) { |
533 | xen_swiotlb_sync_single_for_cpu(dev, sg->dma_address, | |
534 | sg->length, dir); | |
535 | } | |
b097186f | 536 | } |
b097186f | 537 | |
dceb1a68 | 538 | static void |
2e12dcee | 539 | xen_swiotlb_sync_sg_for_device(struct device *dev, struct scatterlist *sgl, |
b097186f KRW |
540 | int nelems, enum dma_data_direction dir) |
541 | { | |
2e12dcee CH |
542 | struct scatterlist *sg; |
543 | int i; | |
544 | ||
545 | for_each_sg(sgl, sg, nelems, i) { | |
546 | xen_swiotlb_sync_single_for_device(dev, sg->dma_address, | |
547 | sg->length, dir); | |
548 | } | |
b097186f | 549 | } |
b097186f | 550 | |
b097186f KRW |
551 | /* |
552 | * Return whether the given device DMA address mask can be supported | |
553 | * properly. For example, if your device can only drive the low 24-bits | |
554 | * during bus mastering, then you would pass 0x00ffffff as the mask to | |
555 | * this function. | |
556 | */ | |
dceb1a68 | 557 | static int |
b097186f KRW |
558 | xen_swiotlb_dma_supported(struct device *hwdev, u64 mask) |
559 | { | |
560 | return xen_virt_to_bus(xen_io_tlb_end - 1) <= mask; | |
561 | } | |
eb1ddc00 | 562 | |
7e91c7df SS |
563 | /* |
564 | * Create userspace mapping for the DMA-coherent memory. | |
565 | * This function should be called with the pages from the current domain only, | |
566 | * passing pages mapped from other domains would lead to memory corruption. | |
567 | */ | |
dceb1a68 | 568 | static int |
7e91c7df SS |
569 | xen_swiotlb_dma_mmap(struct device *dev, struct vm_area_struct *vma, |
570 | void *cpu_addr, dma_addr_t dma_addr, size_t size, | |
571 | unsigned long attrs) | |
572 | { | |
60d8cd57 | 573 | #ifdef CONFIG_ARM |
d5ff5061 SS |
574 | if (xen_get_dma_ops(dev)->mmap) |
575 | return xen_get_dma_ops(dev)->mmap(dev, vma, cpu_addr, | |
7e91c7df SS |
576 | dma_addr, size, attrs); |
577 | #endif | |
58b04406 | 578 | return dma_common_mmap(dev, vma, cpu_addr, dma_addr, size, attrs); |
7e91c7df | 579 | } |
69369f52 AA |
580 | |
581 | /* | |
582 | * This function should be called with the pages from the current domain only, | |
583 | * passing pages mapped from other domains would lead to memory corruption. | |
584 | */ | |
dceb1a68 | 585 | static int |
69369f52 AA |
586 | xen_swiotlb_get_sgtable(struct device *dev, struct sg_table *sgt, |
587 | void *cpu_addr, dma_addr_t handle, size_t size, | |
588 | unsigned long attrs) | |
589 | { | |
60d8cd57 | 590 | #ifdef CONFIG_ARM |
d5ff5061 | 591 | if (xen_get_dma_ops(dev)->get_sgtable) { |
69369f52 AA |
592 | #if 0 |
593 | /* | |
594 | * This check verifies that the page belongs to the current domain and | |
595 | * is not one mapped from another domain. | |
596 | * This check is for debug only, and should not go to production build | |
597 | */ | |
598 | unsigned long bfn = PHYS_PFN(dma_to_phys(dev, handle)); | |
599 | BUG_ON (!page_is_ram(bfn)); | |
600 | #endif | |
d5ff5061 | 601 | return xen_get_dma_ops(dev)->get_sgtable(dev, sgt, cpu_addr, |
69369f52 AA |
602 | handle, size, attrs); |
603 | } | |
604 | #endif | |
9406a49f | 605 | return dma_common_get_sgtable(dev, sgt, cpu_addr, handle, size, attrs); |
69369f52 | 606 | } |
dceb1a68 CH |
607 | |
608 | const struct dma_map_ops xen_swiotlb_dma_ops = { | |
609 | .alloc = xen_swiotlb_alloc_coherent, | |
610 | .free = xen_swiotlb_free_coherent, | |
611 | .sync_single_for_cpu = xen_swiotlb_sync_single_for_cpu, | |
612 | .sync_single_for_device = xen_swiotlb_sync_single_for_device, | |
613 | .sync_sg_for_cpu = xen_swiotlb_sync_sg_for_cpu, | |
614 | .sync_sg_for_device = xen_swiotlb_sync_sg_for_device, | |
aca351cc CH |
615 | .map_sg = xen_swiotlb_map_sg, |
616 | .unmap_sg = xen_swiotlb_unmap_sg, | |
dceb1a68 CH |
617 | .map_page = xen_swiotlb_map_page, |
618 | .unmap_page = xen_swiotlb_unmap_page, | |
619 | .dma_supported = xen_swiotlb_dma_supported, | |
dceb1a68 CH |
620 | .mmap = xen_swiotlb_dma_mmap, |
621 | .get_sgtable = xen_swiotlb_get_sgtable, | |
622 | }; |