]>
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> |
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 | ||
4d048dbc CH |
70 | #define XEN_SWIOTLB_ERROR_CODE (~(dma_addr_t)0x0) |
71 | ||
b097186f KRW |
72 | static char *xen_io_tlb_start, *xen_io_tlb_end; |
73 | static unsigned long xen_io_tlb_nslabs; | |
74 | /* | |
75 | * Quick lookup value of the bus address of the IOTLB. | |
76 | */ | |
77 | ||
b8b0f559 | 78 | static u64 start_dma_addr; |
b097186f | 79 | |
e17b2f11 | 80 | /* |
9435cce8 | 81 | * Both of these functions should avoid XEN_PFN_PHYS because phys_addr_t |
e17b2f11 IC |
82 | * can be 32bit when dma_addr_t is 64bit leading to a loss in |
83 | * information if the shift is done before casting to 64bit. | |
84 | */ | |
6b42a7ea | 85 | static inline dma_addr_t xen_phys_to_bus(phys_addr_t paddr) |
b097186f | 86 | { |
9435cce8 JG |
87 | unsigned long bfn = pfn_to_bfn(XEN_PFN_DOWN(paddr)); |
88 | dma_addr_t dma = (dma_addr_t)bfn << XEN_PAGE_SHIFT; | |
e17b2f11 | 89 | |
9435cce8 | 90 | dma |= paddr & ~XEN_PAGE_MASK; |
e17b2f11 IC |
91 | |
92 | return dma; | |
b097186f KRW |
93 | } |
94 | ||
6b42a7ea | 95 | static inline phys_addr_t xen_bus_to_phys(dma_addr_t baddr) |
b097186f | 96 | { |
9435cce8 JG |
97 | unsigned long xen_pfn = bfn_to_pfn(XEN_PFN_DOWN(baddr)); |
98 | dma_addr_t dma = (dma_addr_t)xen_pfn << XEN_PAGE_SHIFT; | |
e17b2f11 IC |
99 | phys_addr_t paddr = dma; |
100 | ||
9435cce8 | 101 | paddr |= baddr & ~XEN_PAGE_MASK; |
e17b2f11 IC |
102 | |
103 | return paddr; | |
b097186f KRW |
104 | } |
105 | ||
6b42a7ea | 106 | static inline dma_addr_t xen_virt_to_bus(void *address) |
b097186f KRW |
107 | { |
108 | return xen_phys_to_bus(virt_to_phys(address)); | |
109 | } | |
110 | ||
9435cce8 | 111 | static int check_pages_physically_contiguous(unsigned long xen_pfn, |
b097186f KRW |
112 | unsigned int offset, |
113 | size_t length) | |
114 | { | |
32e09870 | 115 | unsigned long next_bfn; |
b097186f KRW |
116 | int i; |
117 | int nr_pages; | |
118 | ||
9435cce8 JG |
119 | next_bfn = pfn_to_bfn(xen_pfn); |
120 | nr_pages = (offset + length + XEN_PAGE_SIZE-1) >> XEN_PAGE_SHIFT; | |
b097186f KRW |
121 | |
122 | for (i = 1; i < nr_pages; i++) { | |
9435cce8 | 123 | if (pfn_to_bfn(++xen_pfn) != ++next_bfn) |
b097186f KRW |
124 | return 0; |
125 | } | |
126 | return 1; | |
127 | } | |
128 | ||
6b42a7ea | 129 | static inline int range_straddles_page_boundary(phys_addr_t p, size_t size) |
b097186f | 130 | { |
9435cce8 JG |
131 | unsigned long xen_pfn = XEN_PFN_DOWN(p); |
132 | unsigned int offset = p & ~XEN_PAGE_MASK; | |
b097186f | 133 | |
9435cce8 | 134 | if (offset + size <= XEN_PAGE_SIZE) |
b097186f | 135 | return 0; |
9435cce8 | 136 | if (check_pages_physically_contiguous(xen_pfn, offset, size)) |
b097186f KRW |
137 | return 0; |
138 | return 1; | |
139 | } | |
140 | ||
141 | static int is_xen_swiotlb_buffer(dma_addr_t dma_addr) | |
142 | { | |
9435cce8 JG |
143 | unsigned long bfn = XEN_PFN_DOWN(dma_addr); |
144 | unsigned long xen_pfn = bfn_to_local_pfn(bfn); | |
145 | phys_addr_t paddr = XEN_PFN_PHYS(xen_pfn); | |
b097186f KRW |
146 | |
147 | /* If the address is outside our domain, it CAN | |
148 | * have the same virtual address as another address | |
149 | * in our domain. Therefore _only_ check address within our domain. | |
150 | */ | |
9435cce8 | 151 | if (pfn_valid(PFN_DOWN(paddr))) { |
b097186f KRW |
152 | return paddr >= virt_to_phys(xen_io_tlb_start) && |
153 | paddr < virt_to_phys(xen_io_tlb_end); | |
154 | } | |
155 | return 0; | |
156 | } | |
157 | ||
158 | static int max_dma_bits = 32; | |
159 | ||
160 | static int | |
161 | xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs) | |
162 | { | |
163 | int i, rc; | |
164 | int dma_bits; | |
69908907 | 165 | dma_addr_t dma_handle; |
1b65c4e5 | 166 | phys_addr_t p = virt_to_phys(buf); |
b097186f KRW |
167 | |
168 | dma_bits = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT) + PAGE_SHIFT; | |
169 | ||
170 | i = 0; | |
171 | do { | |
172 | int slabs = min(nslabs - i, (unsigned long)IO_TLB_SEGSIZE); | |
173 | ||
174 | do { | |
175 | rc = xen_create_contiguous_region( | |
1b65c4e5 | 176 | p + (i << IO_TLB_SHIFT), |
b097186f | 177 | get_order(slabs << IO_TLB_SHIFT), |
69908907 | 178 | dma_bits, &dma_handle); |
b097186f KRW |
179 | } while (rc && dma_bits++ < max_dma_bits); |
180 | if (rc) | |
181 | return rc; | |
182 | ||
183 | i += slabs; | |
184 | } while (i < nslabs); | |
185 | return 0; | |
186 | } | |
1cef36a5 KRW |
187 | static unsigned long xen_set_nslabs(unsigned long nr_tbl) |
188 | { | |
189 | if (!nr_tbl) { | |
190 | xen_io_tlb_nslabs = (64 * 1024 * 1024 >> IO_TLB_SHIFT); | |
191 | xen_io_tlb_nslabs = ALIGN(xen_io_tlb_nslabs, IO_TLB_SEGSIZE); | |
192 | } else | |
193 | xen_io_tlb_nslabs = nr_tbl; | |
b097186f | 194 | |
1cef36a5 KRW |
195 | return xen_io_tlb_nslabs << IO_TLB_SHIFT; |
196 | } | |
b097186f | 197 | |
5bab7864 KRW |
198 | enum xen_swiotlb_err { |
199 | XEN_SWIOTLB_UNKNOWN = 0, | |
200 | XEN_SWIOTLB_ENOMEM, | |
201 | XEN_SWIOTLB_EFIXUP | |
202 | }; | |
203 | ||
204 | static const char *xen_swiotlb_error(enum xen_swiotlb_err err) | |
205 | { | |
206 | switch (err) { | |
207 | case XEN_SWIOTLB_ENOMEM: | |
208 | return "Cannot allocate Xen-SWIOTLB buffer\n"; | |
209 | case XEN_SWIOTLB_EFIXUP: | |
210 | return "Failed to get contiguous memory for DMA from Xen!\n"\ | |
211 | "You either: don't have the permissions, do not have"\ | |
212 | " enough free memory under 4GB, or the hypervisor memory"\ | |
213 | " is too fragmented!"; | |
214 | default: | |
215 | break; | |
216 | } | |
217 | return ""; | |
218 | } | |
b8277600 | 219 | int __ref xen_swiotlb_init(int verbose, bool early) |
b097186f | 220 | { |
b8277600 | 221 | unsigned long bytes, order; |
f4b2f07b | 222 | int rc = -ENOMEM; |
5bab7864 | 223 | enum xen_swiotlb_err m_ret = XEN_SWIOTLB_UNKNOWN; |
f4b2f07b | 224 | unsigned int repeat = 3; |
5f98ecdb | 225 | |
1cef36a5 | 226 | xen_io_tlb_nslabs = swiotlb_nr_tbl(); |
f4b2f07b | 227 | retry: |
1cef36a5 | 228 | bytes = xen_set_nslabs(xen_io_tlb_nslabs); |
b8277600 | 229 | order = get_order(xen_io_tlb_nslabs << IO_TLB_SHIFT); |
b097186f KRW |
230 | /* |
231 | * Get IO TLB memory from any location. | |
232 | */ | |
b8277600 KRW |
233 | if (early) |
234 | xen_io_tlb_start = alloc_bootmem_pages(PAGE_ALIGN(bytes)); | |
235 | else { | |
236 | #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT)) | |
237 | #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT) | |
238 | while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) { | |
8746515d | 239 | xen_io_tlb_start = (void *)xen_get_swiotlb_free_pages(order); |
b8277600 KRW |
240 | if (xen_io_tlb_start) |
241 | break; | |
242 | order--; | |
243 | } | |
244 | if (order != get_order(bytes)) { | |
283c0972 JP |
245 | pr_warn("Warning: only able to allocate %ld MB for software IO TLB\n", |
246 | (PAGE_SIZE << order) >> 20); | |
b8277600 KRW |
247 | xen_io_tlb_nslabs = SLABS_PER_PAGE << order; |
248 | bytes = xen_io_tlb_nslabs << IO_TLB_SHIFT; | |
249 | } | |
250 | } | |
f4b2f07b | 251 | if (!xen_io_tlb_start) { |
5bab7864 | 252 | m_ret = XEN_SWIOTLB_ENOMEM; |
f4b2f07b KRW |
253 | goto error; |
254 | } | |
b097186f KRW |
255 | xen_io_tlb_end = xen_io_tlb_start + bytes; |
256 | /* | |
257 | * And replace that memory with pages under 4GB. | |
258 | */ | |
259 | rc = xen_swiotlb_fixup(xen_io_tlb_start, | |
260 | bytes, | |
261 | xen_io_tlb_nslabs); | |
f4b2f07b | 262 | if (rc) { |
b8277600 KRW |
263 | if (early) |
264 | free_bootmem(__pa(xen_io_tlb_start), PAGE_ALIGN(bytes)); | |
265 | else { | |
266 | free_pages((unsigned long)xen_io_tlb_start, order); | |
267 | xen_io_tlb_start = NULL; | |
268 | } | |
5bab7864 | 269 | m_ret = XEN_SWIOTLB_EFIXUP; |
b097186f | 270 | goto error; |
f4b2f07b | 271 | } |
b097186f | 272 | start_dma_addr = xen_virt_to_bus(xen_io_tlb_start); |
c468bdee | 273 | if (early) { |
ac2cbab2 YL |
274 | if (swiotlb_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs, |
275 | verbose)) | |
276 | panic("Cannot allocate SWIOTLB buffer"); | |
c468bdee KRW |
277 | rc = 0; |
278 | } else | |
b8277600 | 279 | rc = swiotlb_late_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs); |
7453c549 KRW |
280 | |
281 | if (!rc) | |
282 | swiotlb_set_max_segment(PAGE_SIZE); | |
283 | ||
b8277600 | 284 | return rc; |
b097186f | 285 | error: |
f4b2f07b KRW |
286 | if (repeat--) { |
287 | xen_io_tlb_nslabs = max(1024UL, /* Min is 2MB */ | |
288 | (xen_io_tlb_nslabs >> 1)); | |
283c0972 JP |
289 | pr_info("Lowering to %luMB\n", |
290 | (xen_io_tlb_nslabs << IO_TLB_SHIFT) >> 20); | |
f4b2f07b KRW |
291 | goto retry; |
292 | } | |
283c0972 | 293 | pr_err("%s (rc:%d)\n", xen_swiotlb_error(m_ret), rc); |
b8277600 KRW |
294 | if (early) |
295 | panic("%s (rc:%d)", xen_swiotlb_error(m_ret), rc); | |
296 | else | |
297 | free_pages((unsigned long)xen_io_tlb_start, order); | |
298 | return rc; | |
b097186f | 299 | } |
dceb1a68 CH |
300 | |
301 | static void * | |
b097186f | 302 | xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size, |
baa676fc | 303 | dma_addr_t *dma_handle, gfp_t flags, |
00085f1e | 304 | unsigned long attrs) |
b097186f KRW |
305 | { |
306 | void *ret; | |
307 | int order = get_order(size); | |
308 | u64 dma_mask = DMA_BIT_MASK(32); | |
6810df88 KRW |
309 | phys_addr_t phys; |
310 | dma_addr_t dev_addr; | |
b097186f KRW |
311 | |
312 | /* | |
313 | * Ignore region specifiers - the kernel's ideas of | |
314 | * pseudo-phys memory layout has nothing to do with the | |
315 | * machine physical layout. We can't allocate highmem | |
316 | * because we can't return a pointer to it. | |
317 | */ | |
318 | flags &= ~(__GFP_DMA | __GFP_HIGHMEM); | |
319 | ||
1b65c4e5 SS |
320 | /* On ARM this function returns an ioremap'ped virtual address for |
321 | * which virt_to_phys doesn't return the corresponding physical | |
322 | * address. In fact on ARM virt_to_phys only works for kernel direct | |
323 | * mapped RAM memory. Also see comment below. | |
324 | */ | |
325 | ret = xen_alloc_coherent_pages(hwdev, size, dma_handle, flags, attrs); | |
b097186f | 326 | |
6810df88 KRW |
327 | if (!ret) |
328 | return ret; | |
329 | ||
b097186f | 330 | if (hwdev && hwdev->coherent_dma_mask) |
b5031ed1 | 331 | dma_mask = dma_alloc_coherent_mask(hwdev, flags); |
b097186f | 332 | |
1b65c4e5 SS |
333 | /* At this point dma_handle is the physical address, next we are |
334 | * going to set it to the machine address. | |
335 | * Do not use virt_to_phys(ret) because on ARM it doesn't correspond | |
336 | * to *dma_handle. */ | |
337 | phys = *dma_handle; | |
6810df88 KRW |
338 | dev_addr = xen_phys_to_bus(phys); |
339 | if (((dev_addr + size - 1 <= dma_mask)) && | |
340 | !range_straddles_page_boundary(phys, size)) | |
341 | *dma_handle = dev_addr; | |
342 | else { | |
1b65c4e5 | 343 | if (xen_create_contiguous_region(phys, order, |
69908907 | 344 | fls64(dma_mask), dma_handle) != 0) { |
1b65c4e5 | 345 | xen_free_coherent_pages(hwdev, size, ret, (dma_addr_t)phys, attrs); |
b097186f KRW |
346 | return NULL; |
347 | } | |
b097186f | 348 | } |
6810df88 | 349 | memset(ret, 0, size); |
b097186f KRW |
350 | return ret; |
351 | } | |
b097186f | 352 | |
dceb1a68 | 353 | static void |
b097186f | 354 | xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr, |
00085f1e | 355 | dma_addr_t dev_addr, unsigned long attrs) |
b097186f KRW |
356 | { |
357 | int order = get_order(size); | |
6810df88 KRW |
358 | phys_addr_t phys; |
359 | u64 dma_mask = DMA_BIT_MASK(32); | |
b097186f | 360 | |
6810df88 KRW |
361 | if (hwdev && hwdev->coherent_dma_mask) |
362 | dma_mask = hwdev->coherent_dma_mask; | |
363 | ||
1b65c4e5 SS |
364 | /* do not use virt_to_phys because on ARM it doesn't return you the |
365 | * physical address */ | |
366 | phys = xen_bus_to_phys(dev_addr); | |
6810df88 KRW |
367 | |
368 | if (((dev_addr + size - 1 > dma_mask)) || | |
369 | range_straddles_page_boundary(phys, size)) | |
1b65c4e5 | 370 | xen_destroy_contiguous_region(phys, order); |
6810df88 | 371 | |
1b65c4e5 | 372 | xen_free_coherent_pages(hwdev, size, vaddr, (dma_addr_t)phys, attrs); |
b097186f | 373 | } |
b097186f KRW |
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 | */ | |
dceb1a68 | 382 | static dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page, |
b097186f KRW |
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) |
4d048dbc | 415 | return XEN_SWIOTLB_ERROR_CODE; |
b097186f | 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 | 429 | |
4d048dbc | 430 | return XEN_SWIOTLB_ERROR_CODE; |
b097186f | 431 | } |
b097186f KRW |
432 | |
433 | /* | |
434 | * Unmap a single streaming mode DMA translation. The dma_addr and size must | |
435 | * match what was provided for in a previous xen_swiotlb_map_page call. All | |
436 | * other usages are undefined. | |
437 | * | |
438 | * After this call, reads by the cpu to the buffer are guaranteed to see | |
439 | * whatever the device wrote there. | |
440 | */ | |
441 | static void xen_unmap_single(struct device *hwdev, dma_addr_t dev_addr, | |
6cf05463 | 442 | size_t size, enum dma_data_direction dir, |
00085f1e | 443 | unsigned long attrs) |
b097186f KRW |
444 | { |
445 | phys_addr_t paddr = xen_bus_to_phys(dev_addr); | |
446 | ||
447 | BUG_ON(dir == DMA_NONE); | |
448 | ||
d6883e6f | 449 | xen_dma_unmap_page(hwdev, dev_addr, size, dir, attrs); |
6cf05463 | 450 | |
b097186f KRW |
451 | /* NOTE: We use dev_addr here, not paddr! */ |
452 | if (is_xen_swiotlb_buffer(dev_addr)) { | |
0443fa00 | 453 | swiotlb_tbl_unmap_single(hwdev, paddr, size, dir, attrs); |
b097186f KRW |
454 | return; |
455 | } | |
456 | ||
457 | if (dir != DMA_FROM_DEVICE) | |
458 | return; | |
459 | ||
460 | /* | |
461 | * phys_to_virt doesn't work with hihgmem page but we could | |
462 | * call dma_mark_clean() with hihgmem page here. However, we | |
463 | * are fine since dma_mark_clean() is null on POWERPC. We can | |
464 | * make dma_mark_clean() take a physical address if necessary. | |
465 | */ | |
466 | dma_mark_clean(phys_to_virt(paddr), size); | |
467 | } | |
468 | ||
dceb1a68 | 469 | static void xen_swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr, |
b097186f | 470 | size_t size, enum dma_data_direction dir, |
00085f1e | 471 | unsigned long attrs) |
b097186f | 472 | { |
6cf05463 | 473 | xen_unmap_single(hwdev, dev_addr, size, dir, attrs); |
b097186f | 474 | } |
b097186f KRW |
475 | |
476 | /* | |
477 | * Make physical memory consistent for a single streaming mode DMA translation | |
478 | * after a transfer. | |
479 | * | |
480 | * If you perform a xen_swiotlb_map_page() but wish to interrogate the buffer | |
481 | * using the cpu, yet do not wish to teardown the dma mapping, you must | |
482 | * call this function before doing so. At the next point you give the dma | |
483 | * address back to the card, you must first perform a | |
484 | * xen_swiotlb_dma_sync_for_device, and then the device again owns the buffer | |
485 | */ | |
486 | static void | |
487 | xen_swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr, | |
488 | size_t size, enum dma_data_direction dir, | |
489 | enum dma_sync_target target) | |
490 | { | |
491 | phys_addr_t paddr = xen_bus_to_phys(dev_addr); | |
492 | ||
493 | BUG_ON(dir == DMA_NONE); | |
494 | ||
6cf05463 | 495 | if (target == SYNC_FOR_CPU) |
d6883e6f | 496 | xen_dma_sync_single_for_cpu(hwdev, dev_addr, size, dir); |
6cf05463 | 497 | |
b097186f | 498 | /* NOTE: We use dev_addr here, not paddr! */ |
6cf05463 | 499 | if (is_xen_swiotlb_buffer(dev_addr)) |
fbfda893 | 500 | swiotlb_tbl_sync_single(hwdev, paddr, size, dir, target); |
6cf05463 SS |
501 | |
502 | if (target == SYNC_FOR_DEVICE) | |
9490c6c6 | 503 | xen_dma_sync_single_for_device(hwdev, dev_addr, size, dir); |
b097186f KRW |
504 | |
505 | if (dir != DMA_FROM_DEVICE) | |
506 | return; | |
507 | ||
508 | dma_mark_clean(phys_to_virt(paddr), size); | |
509 | } | |
510 | ||
511 | void | |
512 | xen_swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr, | |
513 | size_t size, enum dma_data_direction dir) | |
514 | { | |
515 | xen_swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_CPU); | |
516 | } | |
b097186f KRW |
517 | |
518 | void | |
519 | xen_swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr, | |
520 | size_t size, enum dma_data_direction dir) | |
521 | { | |
522 | xen_swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_DEVICE); | |
523 | } | |
dceb1a68 CH |
524 | |
525 | /* | |
526 | * Unmap a set of streaming mode DMA translations. Again, cpu read rules | |
527 | * concerning calls here are the same as for swiotlb_unmap_page() above. | |
528 | */ | |
529 | static void | |
530 | xen_swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl, | |
531 | int nelems, enum dma_data_direction dir, | |
532 | unsigned long attrs) | |
533 | { | |
534 | struct scatterlist *sg; | |
535 | int i; | |
536 | ||
537 | BUG_ON(dir == DMA_NONE); | |
538 | ||
539 | for_each_sg(sgl, sg, nelems, i) | |
540 | xen_unmap_single(hwdev, sg->dma_address, sg_dma_len(sg), dir, attrs); | |
541 | ||
542 | } | |
b097186f KRW |
543 | |
544 | /* | |
545 | * Map a set of buffers described by scatterlist in streaming mode for DMA. | |
546 | * This is the scatter-gather version of the above xen_swiotlb_map_page | |
547 | * interface. Here the scatter gather list elements are each tagged with the | |
548 | * appropriate dma address and length. They are obtained via | |
549 | * sg_dma_{address,length}(SG). | |
550 | * | |
551 | * NOTE: An implementation may be able to use a smaller number of | |
552 | * DMA address/length pairs than there are SG table elements. | |
553 | * (for example via virtual mapping capabilities) | |
554 | * The routine returns the number of addr/length pairs actually | |
555 | * used, at most nents. | |
556 | * | |
557 | * Device ownership issues as mentioned above for xen_swiotlb_map_page are the | |
558 | * same here. | |
559 | */ | |
dceb1a68 | 560 | static int |
b097186f KRW |
561 | xen_swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, |
562 | int nelems, enum dma_data_direction dir, | |
00085f1e | 563 | unsigned long attrs) |
b097186f KRW |
564 | { |
565 | struct scatterlist *sg; | |
566 | int i; | |
567 | ||
568 | BUG_ON(dir == DMA_NONE); | |
569 | ||
570 | for_each_sg(sgl, sg, nelems, i) { | |
571 | phys_addr_t paddr = sg_phys(sg); | |
572 | dma_addr_t dev_addr = xen_phys_to_bus(paddr); | |
573 | ||
ae7871be | 574 | if (swiotlb_force == SWIOTLB_FORCE || |
291be10f | 575 | xen_arch_need_swiotlb(hwdev, paddr, dev_addr) || |
b097186f KRW |
576 | !dma_capable(hwdev, dev_addr, sg->length) || |
577 | range_straddles_page_boundary(paddr, sg->length)) { | |
e05ed4d1 AD |
578 | phys_addr_t map = swiotlb_tbl_map_single(hwdev, |
579 | start_dma_addr, | |
580 | sg_phys(sg), | |
581 | sg->length, | |
0443fa00 | 582 | dir, attrs); |
e05ed4d1 | 583 | if (map == SWIOTLB_MAP_ERROR) { |
783d0281 | 584 | dev_warn(hwdev, "swiotlb buffer is full\n"); |
b097186f KRW |
585 | /* Don't panic here, we expect map_sg users |
586 | to do proper error handling. */ | |
0443fa00 | 587 | attrs |= DMA_ATTR_SKIP_CPU_SYNC; |
b097186f KRW |
588 | xen_swiotlb_unmap_sg_attrs(hwdev, sgl, i, dir, |
589 | attrs); | |
781575cd | 590 | sg_dma_len(sgl) = 0; |
15177608 | 591 | return 0; |
b097186f | 592 | } |
f1225ee4 | 593 | dev_addr = xen_phys_to_bus(map); |
71bfae90 | 594 | xen_dma_map_page(hwdev, pfn_to_page(map >> PAGE_SHIFT), |
a0f2dee0 | 595 | dev_addr, |
71bfae90 SS |
596 | map & ~PAGE_MASK, |
597 | sg->length, | |
598 | dir, | |
599 | attrs); | |
f1225ee4 | 600 | sg->dma_address = dev_addr; |
6cf05463 SS |
601 | } else { |
602 | /* we are not interested in the dma_addr returned by | |
603 | * xen_dma_map_page, only in the potential cache flushes executed | |
604 | * by the function. */ | |
605 | xen_dma_map_page(hwdev, pfn_to_page(paddr >> PAGE_SHIFT), | |
a0f2dee0 | 606 | dev_addr, |
6cf05463 SS |
607 | paddr & ~PAGE_MASK, |
608 | sg->length, | |
609 | dir, | |
610 | attrs); | |
b097186f | 611 | sg->dma_address = dev_addr; |
6cf05463 | 612 | } |
781575cd | 613 | sg_dma_len(sg) = sg->length; |
b097186f KRW |
614 | } |
615 | return nelems; | |
616 | } | |
b097186f | 617 | |
b097186f KRW |
618 | /* |
619 | * Make physical memory consistent for a set of streaming mode DMA translations | |
620 | * after a transfer. | |
621 | * | |
622 | * The same as swiotlb_sync_single_* but for a scatter-gather list, same rules | |
623 | * and usage. | |
624 | */ | |
625 | static void | |
626 | xen_swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl, | |
627 | int nelems, enum dma_data_direction dir, | |
628 | enum dma_sync_target target) | |
629 | { | |
630 | struct scatterlist *sg; | |
631 | int i; | |
632 | ||
633 | for_each_sg(sgl, sg, nelems, i) | |
634 | xen_swiotlb_sync_single(hwdev, sg->dma_address, | |
781575cd | 635 | sg_dma_len(sg), dir, target); |
b097186f KRW |
636 | } |
637 | ||
dceb1a68 | 638 | static void |
b097186f KRW |
639 | xen_swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg, |
640 | int nelems, enum dma_data_direction dir) | |
641 | { | |
642 | xen_swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_CPU); | |
643 | } | |
b097186f | 644 | |
dceb1a68 | 645 | static void |
b097186f KRW |
646 | xen_swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, |
647 | int nelems, enum dma_data_direction dir) | |
648 | { | |
649 | xen_swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_DEVICE); | |
650 | } | |
b097186f | 651 | |
b097186f KRW |
652 | /* |
653 | * Return whether the given device DMA address mask can be supported | |
654 | * properly. For example, if your device can only drive the low 24-bits | |
655 | * during bus mastering, then you would pass 0x00ffffff as the mask to | |
656 | * this function. | |
657 | */ | |
dceb1a68 | 658 | static int |
b097186f KRW |
659 | xen_swiotlb_dma_supported(struct device *hwdev, u64 mask) |
660 | { | |
661 | return xen_virt_to_bus(xen_io_tlb_end - 1) <= mask; | |
662 | } | |
eb1ddc00 | 663 | |
7e91c7df SS |
664 | /* |
665 | * Create userspace mapping for the DMA-coherent memory. | |
666 | * This function should be called with the pages from the current domain only, | |
667 | * passing pages mapped from other domains would lead to memory corruption. | |
668 | */ | |
dceb1a68 | 669 | static int |
7e91c7df SS |
670 | xen_swiotlb_dma_mmap(struct device *dev, struct vm_area_struct *vma, |
671 | void *cpu_addr, dma_addr_t dma_addr, size_t size, | |
672 | unsigned long attrs) | |
673 | { | |
674 | #if defined(CONFIG_ARM) || defined(CONFIG_ARM64) | |
d5ff5061 SS |
675 | if (xen_get_dma_ops(dev)->mmap) |
676 | return xen_get_dma_ops(dev)->mmap(dev, vma, cpu_addr, | |
7e91c7df SS |
677 | dma_addr, size, attrs); |
678 | #endif | |
679 | return dma_common_mmap(dev, vma, cpu_addr, dma_addr, size); | |
680 | } | |
69369f52 AA |
681 | |
682 | /* | |
683 | * This function should be called with the pages from the current domain only, | |
684 | * passing pages mapped from other domains would lead to memory corruption. | |
685 | */ | |
dceb1a68 | 686 | static int |
69369f52 AA |
687 | xen_swiotlb_get_sgtable(struct device *dev, struct sg_table *sgt, |
688 | void *cpu_addr, dma_addr_t handle, size_t size, | |
689 | unsigned long attrs) | |
690 | { | |
691 | #if defined(CONFIG_ARM) || defined(CONFIG_ARM64) | |
d5ff5061 | 692 | if (xen_get_dma_ops(dev)->get_sgtable) { |
69369f52 AA |
693 | #if 0 |
694 | /* | |
695 | * This check verifies that the page belongs to the current domain and | |
696 | * is not one mapped from another domain. | |
697 | * This check is for debug only, and should not go to production build | |
698 | */ | |
699 | unsigned long bfn = PHYS_PFN(dma_to_phys(dev, handle)); | |
700 | BUG_ON (!page_is_ram(bfn)); | |
701 | #endif | |
d5ff5061 | 702 | return xen_get_dma_ops(dev)->get_sgtable(dev, sgt, cpu_addr, |
69369f52 AA |
703 | handle, size, attrs); |
704 | } | |
705 | #endif | |
706 | return dma_common_get_sgtable(dev, sgt, cpu_addr, handle, size); | |
707 | } | |
dceb1a68 | 708 | |
4d048dbc CH |
709 | static int xen_swiotlb_mapping_error(struct device *dev, dma_addr_t dma_addr) |
710 | { | |
711 | return dma_addr == XEN_SWIOTLB_ERROR_CODE; | |
712 | } | |
713 | ||
dceb1a68 CH |
714 | const struct dma_map_ops xen_swiotlb_dma_ops = { |
715 | .alloc = xen_swiotlb_alloc_coherent, | |
716 | .free = xen_swiotlb_free_coherent, | |
717 | .sync_single_for_cpu = xen_swiotlb_sync_single_for_cpu, | |
718 | .sync_single_for_device = xen_swiotlb_sync_single_for_device, | |
719 | .sync_sg_for_cpu = xen_swiotlb_sync_sg_for_cpu, | |
720 | .sync_sg_for_device = xen_swiotlb_sync_sg_for_device, | |
721 | .map_sg = xen_swiotlb_map_sg_attrs, | |
722 | .unmap_sg = xen_swiotlb_unmap_sg_attrs, | |
723 | .map_page = xen_swiotlb_map_page, | |
724 | .unmap_page = xen_swiotlb_unmap_page, | |
725 | .dma_supported = xen_swiotlb_dma_supported, | |
dceb1a68 CH |
726 | .mmap = xen_swiotlb_dma_mmap, |
727 | .get_sgtable = xen_swiotlb_get_sgtable, | |
4d048dbc | 728 | .mapping_error = xen_swiotlb_mapping_error, |
dceb1a68 | 729 | }; |