]>
Commit | Line | Data |
---|---|---|
b097186f KRW |
1 | /* |
2 | * Copyright 2010 | |
3 | * by Konrad Rzeszutek Wilk <[email protected]> | |
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> |
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 | ||
83862ccf SS |
56 | #ifndef CONFIG_X86 |
57 | static unsigned long dma_alloc_coherent_mask(struct device *dev, | |
58 | gfp_t gfp) | |
59 | { | |
60 | unsigned long dma_mask = 0; | |
61 | ||
62 | dma_mask = dev->coherent_dma_mask; | |
63 | if (!dma_mask) | |
64 | dma_mask = (gfp & GFP_DMA) ? DMA_BIT_MASK(24) : DMA_BIT_MASK(32); | |
65 | ||
66 | return dma_mask; | |
67 | } | |
68 | #endif | |
69 | ||
b097186f KRW |
70 | static char *xen_io_tlb_start, *xen_io_tlb_end; |
71 | static unsigned long xen_io_tlb_nslabs; | |
72 | /* | |
73 | * Quick lookup value of the bus address of the IOTLB. | |
74 | */ | |
75 | ||
b8b0f559 | 76 | static u64 start_dma_addr; |
b097186f | 77 | |
e17b2f11 | 78 | /* |
9435cce8 | 79 | * Both of these functions should avoid XEN_PFN_PHYS because phys_addr_t |
e17b2f11 IC |
80 | * can be 32bit when dma_addr_t is 64bit leading to a loss in |
81 | * information if the shift is done before casting to 64bit. | |
82 | */ | |
6b42a7ea | 83 | static inline dma_addr_t xen_phys_to_bus(phys_addr_t paddr) |
b097186f | 84 | { |
9435cce8 JG |
85 | unsigned long bfn = pfn_to_bfn(XEN_PFN_DOWN(paddr)); |
86 | dma_addr_t dma = (dma_addr_t)bfn << XEN_PAGE_SHIFT; | |
e17b2f11 | 87 | |
9435cce8 | 88 | dma |= paddr & ~XEN_PAGE_MASK; |
e17b2f11 IC |
89 | |
90 | return dma; | |
b097186f KRW |
91 | } |
92 | ||
6b42a7ea | 93 | static inline phys_addr_t xen_bus_to_phys(dma_addr_t baddr) |
b097186f | 94 | { |
9435cce8 JG |
95 | unsigned long xen_pfn = bfn_to_pfn(XEN_PFN_DOWN(baddr)); |
96 | dma_addr_t dma = (dma_addr_t)xen_pfn << XEN_PAGE_SHIFT; | |
e17b2f11 IC |
97 | phys_addr_t paddr = dma; |
98 | ||
9435cce8 | 99 | paddr |= baddr & ~XEN_PAGE_MASK; |
e17b2f11 IC |
100 | |
101 | return paddr; | |
b097186f KRW |
102 | } |
103 | ||
6b42a7ea | 104 | static inline dma_addr_t xen_virt_to_bus(void *address) |
b097186f KRW |
105 | { |
106 | return xen_phys_to_bus(virt_to_phys(address)); | |
107 | } | |
108 | ||
9435cce8 | 109 | static int check_pages_physically_contiguous(unsigned long xen_pfn, |
b097186f KRW |
110 | unsigned int offset, |
111 | size_t length) | |
112 | { | |
32e09870 | 113 | unsigned long next_bfn; |
b097186f KRW |
114 | int i; |
115 | int nr_pages; | |
116 | ||
9435cce8 JG |
117 | next_bfn = pfn_to_bfn(xen_pfn); |
118 | nr_pages = (offset + length + XEN_PAGE_SIZE-1) >> XEN_PAGE_SHIFT; | |
b097186f KRW |
119 | |
120 | for (i = 1; i < nr_pages; i++) { | |
9435cce8 | 121 | if (pfn_to_bfn(++xen_pfn) != ++next_bfn) |
b097186f KRW |
122 | return 0; |
123 | } | |
124 | return 1; | |
125 | } | |
126 | ||
6b42a7ea | 127 | static inline int range_straddles_page_boundary(phys_addr_t p, size_t size) |
b097186f | 128 | { |
9435cce8 JG |
129 | unsigned long xen_pfn = XEN_PFN_DOWN(p); |
130 | unsigned int offset = p & ~XEN_PAGE_MASK; | |
b097186f | 131 | |
9435cce8 | 132 | if (offset + size <= XEN_PAGE_SIZE) |
b097186f | 133 | return 0; |
9435cce8 | 134 | if (check_pages_physically_contiguous(xen_pfn, offset, size)) |
b097186f KRW |
135 | return 0; |
136 | return 1; | |
137 | } | |
138 | ||
139 | static int is_xen_swiotlb_buffer(dma_addr_t dma_addr) | |
140 | { | |
9435cce8 JG |
141 | unsigned long bfn = XEN_PFN_DOWN(dma_addr); |
142 | unsigned long xen_pfn = bfn_to_local_pfn(bfn); | |
143 | phys_addr_t paddr = XEN_PFN_PHYS(xen_pfn); | |
b097186f KRW |
144 | |
145 | /* If the address is outside our domain, it CAN | |
146 | * have the same virtual address as another address | |
147 | * in our domain. Therefore _only_ check address within our domain. | |
148 | */ | |
9435cce8 | 149 | if (pfn_valid(PFN_DOWN(paddr))) { |
b097186f KRW |
150 | return paddr >= virt_to_phys(xen_io_tlb_start) && |
151 | paddr < virt_to_phys(xen_io_tlb_end); | |
152 | } | |
153 | return 0; | |
154 | } | |
155 | ||
156 | static int max_dma_bits = 32; | |
157 | ||
158 | static int | |
159 | xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs) | |
160 | { | |
161 | int i, rc; | |
162 | int dma_bits; | |
69908907 | 163 | dma_addr_t dma_handle; |
1b65c4e5 | 164 | phys_addr_t p = virt_to_phys(buf); |
b097186f KRW |
165 | |
166 | dma_bits = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT) + PAGE_SHIFT; | |
167 | ||
168 | i = 0; | |
169 | do { | |
170 | int slabs = min(nslabs - i, (unsigned long)IO_TLB_SEGSIZE); | |
171 | ||
172 | do { | |
173 | rc = xen_create_contiguous_region( | |
1b65c4e5 | 174 | p + (i << IO_TLB_SHIFT), |
b097186f | 175 | get_order(slabs << IO_TLB_SHIFT), |
69908907 | 176 | dma_bits, &dma_handle); |
b097186f KRW |
177 | } while (rc && dma_bits++ < max_dma_bits); |
178 | if (rc) | |
179 | return rc; | |
180 | ||
181 | i += slabs; | |
182 | } while (i < nslabs); | |
183 | return 0; | |
184 | } | |
1cef36a5 KRW |
185 | static unsigned long xen_set_nslabs(unsigned long nr_tbl) |
186 | { | |
187 | if (!nr_tbl) { | |
188 | xen_io_tlb_nslabs = (64 * 1024 * 1024 >> IO_TLB_SHIFT); | |
189 | xen_io_tlb_nslabs = ALIGN(xen_io_tlb_nslabs, IO_TLB_SEGSIZE); | |
190 | } else | |
191 | xen_io_tlb_nslabs = nr_tbl; | |
b097186f | 192 | |
1cef36a5 KRW |
193 | return xen_io_tlb_nslabs << IO_TLB_SHIFT; |
194 | } | |
b097186f | 195 | |
5bab7864 KRW |
196 | enum xen_swiotlb_err { |
197 | XEN_SWIOTLB_UNKNOWN = 0, | |
198 | XEN_SWIOTLB_ENOMEM, | |
199 | XEN_SWIOTLB_EFIXUP | |
200 | }; | |
201 | ||
202 | static const char *xen_swiotlb_error(enum xen_swiotlb_err err) | |
203 | { | |
204 | switch (err) { | |
205 | case XEN_SWIOTLB_ENOMEM: | |
206 | return "Cannot allocate Xen-SWIOTLB buffer\n"; | |
207 | case XEN_SWIOTLB_EFIXUP: | |
208 | return "Failed to get contiguous memory for DMA from Xen!\n"\ | |
209 | "You either: don't have the permissions, do not have"\ | |
210 | " enough free memory under 4GB, or the hypervisor memory"\ | |
211 | " is too fragmented!"; | |
212 | default: | |
213 | break; | |
214 | } | |
215 | return ""; | |
216 | } | |
b8277600 | 217 | int __ref xen_swiotlb_init(int verbose, bool early) |
b097186f | 218 | { |
b8277600 | 219 | unsigned long bytes, order; |
f4b2f07b | 220 | int rc = -ENOMEM; |
5bab7864 | 221 | enum xen_swiotlb_err m_ret = XEN_SWIOTLB_UNKNOWN; |
f4b2f07b | 222 | unsigned int repeat = 3; |
5f98ecdb | 223 | |
1cef36a5 | 224 | xen_io_tlb_nslabs = swiotlb_nr_tbl(); |
f4b2f07b | 225 | retry: |
1cef36a5 | 226 | bytes = xen_set_nslabs(xen_io_tlb_nslabs); |
b8277600 | 227 | order = get_order(xen_io_tlb_nslabs << IO_TLB_SHIFT); |
b097186f KRW |
228 | /* |
229 | * Get IO TLB memory from any location. | |
230 | */ | |
b8277600 KRW |
231 | if (early) |
232 | xen_io_tlb_start = alloc_bootmem_pages(PAGE_ALIGN(bytes)); | |
233 | else { | |
234 | #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT)) | |
235 | #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT) | |
236 | while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) { | |
8746515d | 237 | xen_io_tlb_start = (void *)xen_get_swiotlb_free_pages(order); |
b8277600 KRW |
238 | if (xen_io_tlb_start) |
239 | break; | |
240 | order--; | |
241 | } | |
242 | if (order != get_order(bytes)) { | |
283c0972 JP |
243 | pr_warn("Warning: only able to allocate %ld MB for software IO TLB\n", |
244 | (PAGE_SIZE << order) >> 20); | |
b8277600 KRW |
245 | xen_io_tlb_nslabs = SLABS_PER_PAGE << order; |
246 | bytes = xen_io_tlb_nslabs << IO_TLB_SHIFT; | |
247 | } | |
248 | } | |
f4b2f07b | 249 | if (!xen_io_tlb_start) { |
5bab7864 | 250 | m_ret = XEN_SWIOTLB_ENOMEM; |
f4b2f07b KRW |
251 | goto error; |
252 | } | |
b097186f KRW |
253 | xen_io_tlb_end = xen_io_tlb_start + bytes; |
254 | /* | |
255 | * And replace that memory with pages under 4GB. | |
256 | */ | |
257 | rc = xen_swiotlb_fixup(xen_io_tlb_start, | |
258 | bytes, | |
259 | xen_io_tlb_nslabs); | |
f4b2f07b | 260 | if (rc) { |
b8277600 KRW |
261 | if (early) |
262 | free_bootmem(__pa(xen_io_tlb_start), PAGE_ALIGN(bytes)); | |
263 | else { | |
264 | free_pages((unsigned long)xen_io_tlb_start, order); | |
265 | xen_io_tlb_start = NULL; | |
266 | } | |
5bab7864 | 267 | m_ret = XEN_SWIOTLB_EFIXUP; |
b097186f | 268 | goto error; |
f4b2f07b | 269 | } |
b097186f | 270 | start_dma_addr = xen_virt_to_bus(xen_io_tlb_start); |
c468bdee | 271 | if (early) { |
ac2cbab2 YL |
272 | if (swiotlb_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs, |
273 | verbose)) | |
274 | panic("Cannot allocate SWIOTLB buffer"); | |
c468bdee KRW |
275 | rc = 0; |
276 | } else | |
b8277600 | 277 | rc = swiotlb_late_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs); |
7453c549 KRW |
278 | |
279 | if (!rc) | |
280 | swiotlb_set_max_segment(PAGE_SIZE); | |
281 | ||
b8277600 | 282 | return rc; |
b097186f | 283 | error: |
f4b2f07b KRW |
284 | if (repeat--) { |
285 | xen_io_tlb_nslabs = max(1024UL, /* Min is 2MB */ | |
286 | (xen_io_tlb_nslabs >> 1)); | |
283c0972 JP |
287 | pr_info("Lowering to %luMB\n", |
288 | (xen_io_tlb_nslabs << IO_TLB_SHIFT) >> 20); | |
f4b2f07b KRW |
289 | goto retry; |
290 | } | |
283c0972 | 291 | pr_err("%s (rc:%d)\n", xen_swiotlb_error(m_ret), rc); |
b8277600 KRW |
292 | if (early) |
293 | panic("%s (rc:%d)", xen_swiotlb_error(m_ret), rc); | |
294 | else | |
295 | free_pages((unsigned long)xen_io_tlb_start, order); | |
296 | return rc; | |
b097186f | 297 | } |
b097186f KRW |
298 | void * |
299 | xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size, | |
baa676fc | 300 | dma_addr_t *dma_handle, gfp_t flags, |
00085f1e | 301 | unsigned long attrs) |
b097186f KRW |
302 | { |
303 | void *ret; | |
304 | int order = get_order(size); | |
305 | u64 dma_mask = DMA_BIT_MASK(32); | |
6810df88 KRW |
306 | phys_addr_t phys; |
307 | dma_addr_t dev_addr; | |
b097186f KRW |
308 | |
309 | /* | |
310 | * Ignore region specifiers - the kernel's ideas of | |
311 | * pseudo-phys memory layout has nothing to do with the | |
312 | * machine physical layout. We can't allocate highmem | |
313 | * because we can't return a pointer to it. | |
314 | */ | |
315 | flags &= ~(__GFP_DMA | __GFP_HIGHMEM); | |
316 | ||
1b65c4e5 SS |
317 | /* On ARM this function returns an ioremap'ped virtual address for |
318 | * which virt_to_phys doesn't return the corresponding physical | |
319 | * address. In fact on ARM virt_to_phys only works for kernel direct | |
320 | * mapped RAM memory. Also see comment below. | |
321 | */ | |
322 | ret = xen_alloc_coherent_pages(hwdev, size, dma_handle, flags, attrs); | |
b097186f | 323 | |
6810df88 KRW |
324 | if (!ret) |
325 | return ret; | |
326 | ||
b097186f | 327 | if (hwdev && hwdev->coherent_dma_mask) |
b5031ed1 | 328 | dma_mask = dma_alloc_coherent_mask(hwdev, flags); |
b097186f | 329 | |
1b65c4e5 SS |
330 | /* At this point dma_handle is the physical address, next we are |
331 | * going to set it to the machine address. | |
332 | * Do not use virt_to_phys(ret) because on ARM it doesn't correspond | |
333 | * to *dma_handle. */ | |
334 | phys = *dma_handle; | |
6810df88 KRW |
335 | dev_addr = xen_phys_to_bus(phys); |
336 | if (((dev_addr + size - 1 <= dma_mask)) && | |
337 | !range_straddles_page_boundary(phys, size)) | |
338 | *dma_handle = dev_addr; | |
339 | else { | |
1b65c4e5 | 340 | if (xen_create_contiguous_region(phys, order, |
69908907 | 341 | fls64(dma_mask), dma_handle) != 0) { |
1b65c4e5 | 342 | xen_free_coherent_pages(hwdev, size, ret, (dma_addr_t)phys, attrs); |
b097186f KRW |
343 | return NULL; |
344 | } | |
b097186f | 345 | } |
6810df88 | 346 | memset(ret, 0, size); |
b097186f KRW |
347 | return ret; |
348 | } | |
349 | EXPORT_SYMBOL_GPL(xen_swiotlb_alloc_coherent); | |
350 | ||
351 | void | |
352 | xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr, | |
00085f1e | 353 | dma_addr_t dev_addr, unsigned long attrs) |
b097186f KRW |
354 | { |
355 | int order = get_order(size); | |
6810df88 KRW |
356 | phys_addr_t phys; |
357 | u64 dma_mask = DMA_BIT_MASK(32); | |
b097186f | 358 | |
6810df88 KRW |
359 | if (hwdev && hwdev->coherent_dma_mask) |
360 | dma_mask = hwdev->coherent_dma_mask; | |
361 | ||
1b65c4e5 SS |
362 | /* do not use virt_to_phys because on ARM it doesn't return you the |
363 | * physical address */ | |
364 | phys = xen_bus_to_phys(dev_addr); | |
6810df88 KRW |
365 | |
366 | if (((dev_addr + size - 1 > dma_mask)) || | |
367 | range_straddles_page_boundary(phys, size)) | |
1b65c4e5 | 368 | xen_destroy_contiguous_region(phys, order); |
6810df88 | 369 | |
1b65c4e5 | 370 | xen_free_coherent_pages(hwdev, size, vaddr, (dma_addr_t)phys, attrs); |
b097186f KRW |
371 | } |
372 | EXPORT_SYMBOL_GPL(xen_swiotlb_free_coherent); | |
373 | ||
374 | ||
375 | /* | |
376 | * Map a single buffer of the indicated size for DMA in streaming mode. The | |
377 | * physical address to use is returned. | |
378 | * | |
379 | * Once the device is given the dma address, the device owns this memory until | |
380 | * either xen_swiotlb_unmap_page or xen_swiotlb_dma_sync_single is performed. | |
381 | */ | |
382 | dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page, | |
383 | unsigned long offset, size_t size, | |
384 | enum dma_data_direction dir, | |
00085f1e | 385 | unsigned long attrs) |
b097186f | 386 | { |
e05ed4d1 | 387 | phys_addr_t map, phys = page_to_phys(page) + offset; |
b097186f | 388 | dma_addr_t dev_addr = xen_phys_to_bus(phys); |
b097186f KRW |
389 | |
390 | BUG_ON(dir == DMA_NONE); | |
391 | /* | |
392 | * If the address happens to be in the device's DMA window, | |
393 | * we can safely return the device addr and not worry about bounce | |
394 | * buffering it. | |
395 | */ | |
396 | if (dma_capable(dev, dev_addr, size) && | |
a4dba130 | 397 | !range_straddles_page_boundary(phys, size) && |
291be10f | 398 | !xen_arch_need_swiotlb(dev, phys, dev_addr) && |
ae7871be | 399 | (swiotlb_force != SWIOTLB_FORCE)) { |
6cf05463 SS |
400 | /* we are not interested in the dma_addr returned by |
401 | * xen_dma_map_page, only in the potential cache flushes executed | |
402 | * by the function. */ | |
a0f2dee0 | 403 | xen_dma_map_page(dev, page, dev_addr, offset, size, dir, attrs); |
b097186f | 404 | return dev_addr; |
6cf05463 | 405 | } |
b097186f KRW |
406 | |
407 | /* | |
408 | * Oh well, have to allocate and map a bounce buffer. | |
409 | */ | |
2b2b614d ZK |
410 | trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force); |
411 | ||
0443fa00 AD |
412 | map = swiotlb_tbl_map_single(dev, start_dma_addr, phys, size, dir, |
413 | attrs); | |
e05ed4d1 | 414 | if (map == SWIOTLB_MAP_ERROR) |
b097186f KRW |
415 | return DMA_ERROR_CODE; |
416 | ||
f1225ee4 | 417 | dev_addr = xen_phys_to_bus(map); |
6cf05463 | 418 | xen_dma_map_page(dev, pfn_to_page(map >> PAGE_SHIFT), |
a0f2dee0 | 419 | dev_addr, map & ~PAGE_MASK, size, dir, attrs); |
b097186f KRW |
420 | |
421 | /* | |
422 | * Ensure that the address returned is DMA'ble | |
423 | */ | |
76418421 AD |
424 | if (dma_capable(dev, dev_addr, size)) |
425 | return dev_addr; | |
426 | ||
d29fa0cb AD |
427 | attrs |= DMA_ATTR_SKIP_CPU_SYNC; |
428 | swiotlb_tbl_unmap_single(dev, map, size, dir, attrs); | |
76418421 AD |
429 | |
430 | return DMA_ERROR_CODE; | |
b097186f KRW |
431 | } |
432 | EXPORT_SYMBOL_GPL(xen_swiotlb_map_page); | |
433 | ||
434 | /* | |
435 | * Unmap a single streaming mode DMA translation. The dma_addr and size must | |
436 | * match what was provided for in a previous xen_swiotlb_map_page call. All | |
437 | * other usages are undefined. | |
438 | * | |
439 | * After this call, reads by the cpu to the buffer are guaranteed to see | |
440 | * whatever the device wrote there. | |
441 | */ | |
442 | static void xen_unmap_single(struct device *hwdev, dma_addr_t dev_addr, | |
6cf05463 | 443 | size_t size, enum dma_data_direction dir, |
00085f1e | 444 | unsigned long attrs) |
b097186f KRW |
445 | { |
446 | phys_addr_t paddr = xen_bus_to_phys(dev_addr); | |
447 | ||
448 | BUG_ON(dir == DMA_NONE); | |
449 | ||
d6883e6f | 450 | xen_dma_unmap_page(hwdev, dev_addr, size, dir, attrs); |
6cf05463 | 451 | |
b097186f KRW |
452 | /* NOTE: We use dev_addr here, not paddr! */ |
453 | if (is_xen_swiotlb_buffer(dev_addr)) { | |
0443fa00 | 454 | swiotlb_tbl_unmap_single(hwdev, paddr, size, dir, attrs); |
b097186f KRW |
455 | return; |
456 | } | |
457 | ||
458 | if (dir != DMA_FROM_DEVICE) | |
459 | return; | |
460 | ||
461 | /* | |
462 | * phys_to_virt doesn't work with hihgmem page but we could | |
463 | * call dma_mark_clean() with hihgmem page here. However, we | |
464 | * are fine since dma_mark_clean() is null on POWERPC. We can | |
465 | * make dma_mark_clean() take a physical address if necessary. | |
466 | */ | |
467 | dma_mark_clean(phys_to_virt(paddr), size); | |
468 | } | |
469 | ||
470 | void xen_swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr, | |
471 | size_t size, enum dma_data_direction dir, | |
00085f1e | 472 | unsigned long attrs) |
b097186f | 473 | { |
6cf05463 | 474 | xen_unmap_single(hwdev, dev_addr, size, dir, attrs); |
b097186f KRW |
475 | } |
476 | EXPORT_SYMBOL_GPL(xen_swiotlb_unmap_page); | |
477 | ||
478 | /* | |
479 | * Make physical memory consistent for a single streaming mode DMA translation | |
480 | * after a transfer. | |
481 | * | |
482 | * If you perform a xen_swiotlb_map_page() but wish to interrogate the buffer | |
483 | * using the cpu, yet do not wish to teardown the dma mapping, you must | |
484 | * call this function before doing so. At the next point you give the dma | |
485 | * address back to the card, you must first perform a | |
486 | * xen_swiotlb_dma_sync_for_device, and then the device again owns the buffer | |
487 | */ | |
488 | static void | |
489 | xen_swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr, | |
490 | size_t size, enum dma_data_direction dir, | |
491 | enum dma_sync_target target) | |
492 | { | |
493 | phys_addr_t paddr = xen_bus_to_phys(dev_addr); | |
494 | ||
495 | BUG_ON(dir == DMA_NONE); | |
496 | ||
6cf05463 | 497 | if (target == SYNC_FOR_CPU) |
d6883e6f | 498 | xen_dma_sync_single_for_cpu(hwdev, dev_addr, size, dir); |
6cf05463 | 499 | |
b097186f | 500 | /* NOTE: We use dev_addr here, not paddr! */ |
6cf05463 | 501 | if (is_xen_swiotlb_buffer(dev_addr)) |
fbfda893 | 502 | swiotlb_tbl_sync_single(hwdev, paddr, size, dir, target); |
6cf05463 SS |
503 | |
504 | if (target == SYNC_FOR_DEVICE) | |
9490c6c6 | 505 | xen_dma_sync_single_for_device(hwdev, dev_addr, size, dir); |
b097186f KRW |
506 | |
507 | if (dir != DMA_FROM_DEVICE) | |
508 | return; | |
509 | ||
510 | dma_mark_clean(phys_to_virt(paddr), size); | |
511 | } | |
512 | ||
513 | void | |
514 | xen_swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr, | |
515 | size_t size, enum dma_data_direction dir) | |
516 | { | |
517 | xen_swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_CPU); | |
518 | } | |
519 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_single_for_cpu); | |
520 | ||
521 | void | |
522 | xen_swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr, | |
523 | size_t size, enum dma_data_direction dir) | |
524 | { | |
525 | xen_swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_DEVICE); | |
526 | } | |
527 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_single_for_device); | |
528 | ||
529 | /* | |
530 | * Map a set of buffers described by scatterlist in streaming mode for DMA. | |
531 | * This is the scatter-gather version of the above xen_swiotlb_map_page | |
532 | * interface. Here the scatter gather list elements are each tagged with the | |
533 | * appropriate dma address and length. They are obtained via | |
534 | * sg_dma_{address,length}(SG). | |
535 | * | |
536 | * NOTE: An implementation may be able to use a smaller number of | |
537 | * DMA address/length pairs than there are SG table elements. | |
538 | * (for example via virtual mapping capabilities) | |
539 | * The routine returns the number of addr/length pairs actually | |
540 | * used, at most nents. | |
541 | * | |
542 | * Device ownership issues as mentioned above for xen_swiotlb_map_page are the | |
543 | * same here. | |
544 | */ | |
545 | int | |
546 | xen_swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, | |
547 | int nelems, enum dma_data_direction dir, | |
00085f1e | 548 | unsigned long attrs) |
b097186f KRW |
549 | { |
550 | struct scatterlist *sg; | |
551 | int i; | |
552 | ||
553 | BUG_ON(dir == DMA_NONE); | |
554 | ||
555 | for_each_sg(sgl, sg, nelems, i) { | |
556 | phys_addr_t paddr = sg_phys(sg); | |
557 | dma_addr_t dev_addr = xen_phys_to_bus(paddr); | |
558 | ||
ae7871be | 559 | if (swiotlb_force == SWIOTLB_FORCE || |
291be10f | 560 | xen_arch_need_swiotlb(hwdev, paddr, dev_addr) || |
b097186f KRW |
561 | !dma_capable(hwdev, dev_addr, sg->length) || |
562 | range_straddles_page_boundary(paddr, sg->length)) { | |
e05ed4d1 AD |
563 | phys_addr_t map = swiotlb_tbl_map_single(hwdev, |
564 | start_dma_addr, | |
565 | sg_phys(sg), | |
566 | sg->length, | |
0443fa00 | 567 | dir, attrs); |
e05ed4d1 | 568 | if (map == SWIOTLB_MAP_ERROR) { |
783d0281 | 569 | dev_warn(hwdev, "swiotlb buffer is full\n"); |
b097186f KRW |
570 | /* Don't panic here, we expect map_sg users |
571 | to do proper error handling. */ | |
0443fa00 | 572 | attrs |= DMA_ATTR_SKIP_CPU_SYNC; |
b097186f KRW |
573 | xen_swiotlb_unmap_sg_attrs(hwdev, sgl, i, dir, |
574 | attrs); | |
781575cd | 575 | sg_dma_len(sgl) = 0; |
15177608 | 576 | return 0; |
b097186f | 577 | } |
f1225ee4 | 578 | dev_addr = xen_phys_to_bus(map); |
71bfae90 | 579 | xen_dma_map_page(hwdev, pfn_to_page(map >> PAGE_SHIFT), |
a0f2dee0 | 580 | dev_addr, |
71bfae90 SS |
581 | map & ~PAGE_MASK, |
582 | sg->length, | |
583 | dir, | |
584 | attrs); | |
f1225ee4 | 585 | sg->dma_address = dev_addr; |
6cf05463 SS |
586 | } else { |
587 | /* we are not interested in the dma_addr returned by | |
588 | * xen_dma_map_page, only in the potential cache flushes executed | |
589 | * by the function. */ | |
590 | xen_dma_map_page(hwdev, pfn_to_page(paddr >> PAGE_SHIFT), | |
a0f2dee0 | 591 | dev_addr, |
6cf05463 SS |
592 | paddr & ~PAGE_MASK, |
593 | sg->length, | |
594 | dir, | |
595 | attrs); | |
b097186f | 596 | sg->dma_address = dev_addr; |
6cf05463 | 597 | } |
781575cd | 598 | sg_dma_len(sg) = sg->length; |
b097186f KRW |
599 | } |
600 | return nelems; | |
601 | } | |
602 | EXPORT_SYMBOL_GPL(xen_swiotlb_map_sg_attrs); | |
603 | ||
b097186f KRW |
604 | /* |
605 | * Unmap a set of streaming mode DMA translations. Again, cpu read rules | |
606 | * concerning calls here are the same as for swiotlb_unmap_page() above. | |
607 | */ | |
608 | void | |
609 | xen_swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl, | |
610 | int nelems, enum dma_data_direction dir, | |
00085f1e | 611 | unsigned long attrs) |
b097186f KRW |
612 | { |
613 | struct scatterlist *sg; | |
614 | int i; | |
615 | ||
616 | BUG_ON(dir == DMA_NONE); | |
617 | ||
618 | for_each_sg(sgl, sg, nelems, i) | |
6cf05463 | 619 | xen_unmap_single(hwdev, sg->dma_address, sg_dma_len(sg), dir, attrs); |
b097186f KRW |
620 | |
621 | } | |
622 | EXPORT_SYMBOL_GPL(xen_swiotlb_unmap_sg_attrs); | |
623 | ||
b097186f KRW |
624 | /* |
625 | * Make physical memory consistent for a set of streaming mode DMA translations | |
626 | * after a transfer. | |
627 | * | |
628 | * The same as swiotlb_sync_single_* but for a scatter-gather list, same rules | |
629 | * and usage. | |
630 | */ | |
631 | static void | |
632 | xen_swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl, | |
633 | int nelems, enum dma_data_direction dir, | |
634 | enum dma_sync_target target) | |
635 | { | |
636 | struct scatterlist *sg; | |
637 | int i; | |
638 | ||
639 | for_each_sg(sgl, sg, nelems, i) | |
640 | xen_swiotlb_sync_single(hwdev, sg->dma_address, | |
781575cd | 641 | sg_dma_len(sg), dir, target); |
b097186f KRW |
642 | } |
643 | ||
644 | void | |
645 | xen_swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg, | |
646 | int nelems, enum dma_data_direction dir) | |
647 | { | |
648 | xen_swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_CPU); | |
649 | } | |
650 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_sg_for_cpu); | |
651 | ||
652 | void | |
653 | xen_swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, | |
654 | int nelems, enum dma_data_direction dir) | |
655 | { | |
656 | xen_swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_DEVICE); | |
657 | } | |
658 | EXPORT_SYMBOL_GPL(xen_swiotlb_sync_sg_for_device); | |
659 | ||
b097186f KRW |
660 | /* |
661 | * Return whether the given device DMA address mask can be supported | |
662 | * properly. For example, if your device can only drive the low 24-bits | |
663 | * during bus mastering, then you would pass 0x00ffffff as the mask to | |
664 | * this function. | |
665 | */ | |
666 | int | |
667 | xen_swiotlb_dma_supported(struct device *hwdev, u64 mask) | |
668 | { | |
669 | return xen_virt_to_bus(xen_io_tlb_end - 1) <= mask; | |
670 | } | |
671 | EXPORT_SYMBOL_GPL(xen_swiotlb_dma_supported); | |
eb1ddc00 SS |
672 | |
673 | int | |
674 | xen_swiotlb_set_dma_mask(struct device *dev, u64 dma_mask) | |
675 | { | |
676 | if (!dev->dma_mask || !xen_swiotlb_dma_supported(dev, dma_mask)) | |
677 | return -EIO; | |
678 | ||
679 | *dev->dma_mask = dma_mask; | |
680 | ||
681 | return 0; | |
682 | } | |
683 | EXPORT_SYMBOL_GPL(xen_swiotlb_set_dma_mask); |