mfd: kempld-core: Constify variables that point to const structure
[linux-2.6-block.git] / drivers / xen / swiotlb-xen.c
CommitLineData
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 38#include <linux/bootmem.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
4d048dbc
CH
56#define XEN_SWIOTLB_ERROR_CODE (~(dma_addr_t)0x0)
57
b097186f
KRW
58static char *xen_io_tlb_start, *xen_io_tlb_end;
59static unsigned long xen_io_tlb_nslabs;
60/*
61 * Quick lookup value of the bus address of the IOTLB.
62 */
63
b8b0f559 64static u64 start_dma_addr;
b097186f 65
e17b2f11 66/*
9435cce8 67 * Both of these functions should avoid XEN_PFN_PHYS because phys_addr_t
e17b2f11
IC
68 * can be 32bit when dma_addr_t is 64bit leading to a loss in
69 * information if the shift is done before casting to 64bit.
70 */
6b42a7ea 71static inline dma_addr_t xen_phys_to_bus(phys_addr_t paddr)
b097186f 72{
9435cce8
JG
73 unsigned long bfn = pfn_to_bfn(XEN_PFN_DOWN(paddr));
74 dma_addr_t dma = (dma_addr_t)bfn << XEN_PAGE_SHIFT;
e17b2f11 75
9435cce8 76 dma |= paddr & ~XEN_PAGE_MASK;
e17b2f11
IC
77
78 return dma;
b097186f
KRW
79}
80
6b42a7ea 81static inline phys_addr_t xen_bus_to_phys(dma_addr_t baddr)
b097186f 82{
9435cce8
JG
83 unsigned long xen_pfn = bfn_to_pfn(XEN_PFN_DOWN(baddr));
84 dma_addr_t dma = (dma_addr_t)xen_pfn << XEN_PAGE_SHIFT;
e17b2f11
IC
85 phys_addr_t paddr = dma;
86
9435cce8 87 paddr |= baddr & ~XEN_PAGE_MASK;
e17b2f11
IC
88
89 return paddr;
b097186f
KRW
90}
91
6b42a7ea 92static inline dma_addr_t xen_virt_to_bus(void *address)
b097186f
KRW
93{
94 return xen_phys_to_bus(virt_to_phys(address));
95}
96
9435cce8 97static int check_pages_physically_contiguous(unsigned long xen_pfn,
b097186f
KRW
98 unsigned int offset,
99 size_t length)
100{
32e09870 101 unsigned long next_bfn;
b097186f
KRW
102 int i;
103 int nr_pages;
104
9435cce8
JG
105 next_bfn = pfn_to_bfn(xen_pfn);
106 nr_pages = (offset + length + XEN_PAGE_SIZE-1) >> XEN_PAGE_SHIFT;
b097186f
KRW
107
108 for (i = 1; i < nr_pages; i++) {
9435cce8 109 if (pfn_to_bfn(++xen_pfn) != ++next_bfn)
b097186f
KRW
110 return 0;
111 }
112 return 1;
113}
114
6b42a7ea 115static inline int range_straddles_page_boundary(phys_addr_t p, size_t size)
b097186f 116{
9435cce8
JG
117 unsigned long xen_pfn = XEN_PFN_DOWN(p);
118 unsigned int offset = p & ~XEN_PAGE_MASK;
b097186f 119
9435cce8 120 if (offset + size <= XEN_PAGE_SIZE)
b097186f 121 return 0;
9435cce8 122 if (check_pages_physically_contiguous(xen_pfn, offset, size))
b097186f
KRW
123 return 0;
124 return 1;
125}
126
127static int is_xen_swiotlb_buffer(dma_addr_t dma_addr)
128{
9435cce8
JG
129 unsigned long bfn = XEN_PFN_DOWN(dma_addr);
130 unsigned long xen_pfn = bfn_to_local_pfn(bfn);
131 phys_addr_t paddr = XEN_PFN_PHYS(xen_pfn);
b097186f
KRW
132
133 /* If the address is outside our domain, it CAN
134 * have the same virtual address as another address
135 * in our domain. Therefore _only_ check address within our domain.
136 */
9435cce8 137 if (pfn_valid(PFN_DOWN(paddr))) {
b097186f
KRW
138 return paddr >= virt_to_phys(xen_io_tlb_start) &&
139 paddr < virt_to_phys(xen_io_tlb_end);
140 }
141 return 0;
142}
143
144static int max_dma_bits = 32;
145
146static int
147xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs)
148{
149 int i, rc;
150 int dma_bits;
69908907 151 dma_addr_t dma_handle;
1b65c4e5 152 phys_addr_t p = virt_to_phys(buf);
b097186f
KRW
153
154 dma_bits = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT) + PAGE_SHIFT;
155
156 i = 0;
157 do {
158 int slabs = min(nslabs - i, (unsigned long)IO_TLB_SEGSIZE);
159
160 do {
161 rc = xen_create_contiguous_region(
1b65c4e5 162 p + (i << IO_TLB_SHIFT),
b097186f 163 get_order(slabs << IO_TLB_SHIFT),
69908907 164 dma_bits, &dma_handle);
b097186f
KRW
165 } while (rc && dma_bits++ < max_dma_bits);
166 if (rc)
167 return rc;
168
169 i += slabs;
170 } while (i < nslabs);
171 return 0;
172}
1cef36a5
KRW
173static unsigned long xen_set_nslabs(unsigned long nr_tbl)
174{
175 if (!nr_tbl) {
176 xen_io_tlb_nslabs = (64 * 1024 * 1024 >> IO_TLB_SHIFT);
177 xen_io_tlb_nslabs = ALIGN(xen_io_tlb_nslabs, IO_TLB_SEGSIZE);
178 } else
179 xen_io_tlb_nslabs = nr_tbl;
b097186f 180
1cef36a5
KRW
181 return xen_io_tlb_nslabs << IO_TLB_SHIFT;
182}
b097186f 183
5bab7864
KRW
184enum xen_swiotlb_err {
185 XEN_SWIOTLB_UNKNOWN = 0,
186 XEN_SWIOTLB_ENOMEM,
187 XEN_SWIOTLB_EFIXUP
188};
189
190static const char *xen_swiotlb_error(enum xen_swiotlb_err err)
191{
192 switch (err) {
193 case XEN_SWIOTLB_ENOMEM:
194 return "Cannot allocate Xen-SWIOTLB buffer\n";
195 case XEN_SWIOTLB_EFIXUP:
196 return "Failed to get contiguous memory for DMA from Xen!\n"\
197 "You either: don't have the permissions, do not have"\
198 " enough free memory under 4GB, or the hypervisor memory"\
199 " is too fragmented!";
200 default:
201 break;
202 }
203 return "";
204}
b8277600 205int __ref xen_swiotlb_init(int verbose, bool early)
b097186f 206{
b8277600 207 unsigned long bytes, order;
f4b2f07b 208 int rc = -ENOMEM;
5bab7864 209 enum xen_swiotlb_err m_ret = XEN_SWIOTLB_UNKNOWN;
f4b2f07b 210 unsigned int repeat = 3;
5f98ecdb 211
1cef36a5 212 xen_io_tlb_nslabs = swiotlb_nr_tbl();
f4b2f07b 213retry:
1cef36a5 214 bytes = xen_set_nslabs(xen_io_tlb_nslabs);
b8277600 215 order = get_order(xen_io_tlb_nslabs << IO_TLB_SHIFT);
b097186f
KRW
216 /*
217 * Get IO TLB memory from any location.
218 */
b8277600
KRW
219 if (early)
220 xen_io_tlb_start = alloc_bootmem_pages(PAGE_ALIGN(bytes));
221 else {
222#define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT))
223#define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT)
224 while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) {
8746515d 225 xen_io_tlb_start = (void *)xen_get_swiotlb_free_pages(order);
b8277600
KRW
226 if (xen_io_tlb_start)
227 break;
228 order--;
229 }
230 if (order != get_order(bytes)) {
283c0972
JP
231 pr_warn("Warning: only able to allocate %ld MB for software IO TLB\n",
232 (PAGE_SIZE << order) >> 20);
b8277600
KRW
233 xen_io_tlb_nslabs = SLABS_PER_PAGE << order;
234 bytes = xen_io_tlb_nslabs << IO_TLB_SHIFT;
235 }
236 }
f4b2f07b 237 if (!xen_io_tlb_start) {
5bab7864 238 m_ret = XEN_SWIOTLB_ENOMEM;
f4b2f07b
KRW
239 goto error;
240 }
b097186f
KRW
241 xen_io_tlb_end = xen_io_tlb_start + bytes;
242 /*
243 * And replace that memory with pages under 4GB.
244 */
245 rc = xen_swiotlb_fixup(xen_io_tlb_start,
246 bytes,
247 xen_io_tlb_nslabs);
f4b2f07b 248 if (rc) {
b8277600
KRW
249 if (early)
250 free_bootmem(__pa(xen_io_tlb_start), PAGE_ALIGN(bytes));
251 else {
252 free_pages((unsigned long)xen_io_tlb_start, order);
253 xen_io_tlb_start = NULL;
254 }
5bab7864 255 m_ret = XEN_SWIOTLB_EFIXUP;
b097186f 256 goto error;
f4b2f07b 257 }
b097186f 258 start_dma_addr = xen_virt_to_bus(xen_io_tlb_start);
c468bdee 259 if (early) {
ac2cbab2
YL
260 if (swiotlb_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs,
261 verbose))
262 panic("Cannot allocate SWIOTLB buffer");
c468bdee
KRW
263 rc = 0;
264 } else
b8277600 265 rc = swiotlb_late_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs);
7453c549
KRW
266
267 if (!rc)
268 swiotlb_set_max_segment(PAGE_SIZE);
269
b8277600 270 return rc;
b097186f 271error:
f4b2f07b
KRW
272 if (repeat--) {
273 xen_io_tlb_nslabs = max(1024UL, /* Min is 2MB */
274 (xen_io_tlb_nslabs >> 1));
283c0972
JP
275 pr_info("Lowering to %luMB\n",
276 (xen_io_tlb_nslabs << IO_TLB_SHIFT) >> 20);
f4b2f07b
KRW
277 goto retry;
278 }
283c0972 279 pr_err("%s (rc:%d)\n", xen_swiotlb_error(m_ret), rc);
b8277600
KRW
280 if (early)
281 panic("%s (rc:%d)", xen_swiotlb_error(m_ret), rc);
282 else
283 free_pages((unsigned long)xen_io_tlb_start, order);
284 return rc;
b097186f 285}
dceb1a68
CH
286
287static void *
b097186f 288xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size,
baa676fc 289 dma_addr_t *dma_handle, gfp_t flags,
00085f1e 290 unsigned long attrs)
b097186f
KRW
291{
292 void *ret;
293 int order = get_order(size);
294 u64 dma_mask = DMA_BIT_MASK(32);
6810df88
KRW
295 phys_addr_t phys;
296 dma_addr_t dev_addr;
b097186f
KRW
297
298 /*
299 * Ignore region specifiers - the kernel's ideas of
300 * pseudo-phys memory layout has nothing to do with the
301 * machine physical layout. We can't allocate highmem
302 * because we can't return a pointer to it.
303 */
304 flags &= ~(__GFP_DMA | __GFP_HIGHMEM);
305
1b65c4e5
SS
306 /* On ARM this function returns an ioremap'ped virtual address for
307 * which virt_to_phys doesn't return the corresponding physical
308 * address. In fact on ARM virt_to_phys only works for kernel direct
309 * mapped RAM memory. Also see comment below.
310 */
311 ret = xen_alloc_coherent_pages(hwdev, size, dma_handle, flags, attrs);
b097186f 312
6810df88
KRW
313 if (!ret)
314 return ret;
315
b097186f 316 if (hwdev && hwdev->coherent_dma_mask)
038d07a2 317 dma_mask = hwdev->coherent_dma_mask;
b097186f 318
1b65c4e5
SS
319 /* At this point dma_handle is the physical address, next we are
320 * going to set it to the machine address.
321 * Do not use virt_to_phys(ret) because on ARM it doesn't correspond
322 * to *dma_handle. */
323 phys = *dma_handle;
6810df88
KRW
324 dev_addr = xen_phys_to_bus(phys);
325 if (((dev_addr + size - 1 <= dma_mask)) &&
326 !range_straddles_page_boundary(phys, size))
327 *dma_handle = dev_addr;
328 else {
1b65c4e5 329 if (xen_create_contiguous_region(phys, order,
69908907 330 fls64(dma_mask), dma_handle) != 0) {
1b65c4e5 331 xen_free_coherent_pages(hwdev, size, ret, (dma_addr_t)phys, attrs);
b097186f
KRW
332 return NULL;
333 }
b097186f 334 }
6810df88 335 memset(ret, 0, size);
b097186f
KRW
336 return ret;
337}
b097186f 338
dceb1a68 339static void
b097186f 340xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr,
00085f1e 341 dma_addr_t dev_addr, unsigned long attrs)
b097186f
KRW
342{
343 int order = get_order(size);
6810df88
KRW
344 phys_addr_t phys;
345 u64 dma_mask = DMA_BIT_MASK(32);
b097186f 346
6810df88
KRW
347 if (hwdev && hwdev->coherent_dma_mask)
348 dma_mask = hwdev->coherent_dma_mask;
349
1b65c4e5
SS
350 /* do not use virt_to_phys because on ARM it doesn't return you the
351 * physical address */
352 phys = xen_bus_to_phys(dev_addr);
6810df88 353
4855c92d 354 if (((dev_addr + size - 1 <= dma_mask)) ||
6810df88 355 range_straddles_page_boundary(phys, size))
1b65c4e5 356 xen_destroy_contiguous_region(phys, order);
6810df88 357
1b65c4e5 358 xen_free_coherent_pages(hwdev, size, vaddr, (dma_addr_t)phys, attrs);
b097186f 359}
b097186f
KRW
360
361/*
362 * Map a single buffer of the indicated size for DMA in streaming mode. The
363 * physical address to use is returned.
364 *
365 * Once the device is given the dma address, the device owns this memory until
366 * either xen_swiotlb_unmap_page or xen_swiotlb_dma_sync_single is performed.
367 */
dceb1a68 368static dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page,
b097186f
KRW
369 unsigned long offset, size_t size,
370 enum dma_data_direction dir,
00085f1e 371 unsigned long attrs)
b097186f 372{
e05ed4d1 373 phys_addr_t map, phys = page_to_phys(page) + offset;
b097186f 374 dma_addr_t dev_addr = xen_phys_to_bus(phys);
b097186f
KRW
375
376 BUG_ON(dir == DMA_NONE);
377 /*
378 * If the address happens to be in the device's DMA window,
379 * we can safely return the device addr and not worry about bounce
380 * buffering it.
381 */
382 if (dma_capable(dev, dev_addr, size) &&
a4dba130 383 !range_straddles_page_boundary(phys, size) &&
291be10f 384 !xen_arch_need_swiotlb(dev, phys, dev_addr) &&
ae7871be 385 (swiotlb_force != SWIOTLB_FORCE)) {
6cf05463
SS
386 /* we are not interested in the dma_addr returned by
387 * xen_dma_map_page, only in the potential cache flushes executed
388 * by the function. */
a0f2dee0 389 xen_dma_map_page(dev, page, dev_addr, offset, size, dir, attrs);
b097186f 390 return dev_addr;
6cf05463 391 }
b097186f
KRW
392
393 /*
394 * Oh well, have to allocate and map a bounce buffer.
395 */
2b2b614d
ZK
396 trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force);
397
0443fa00
AD
398 map = swiotlb_tbl_map_single(dev, start_dma_addr, phys, size, dir,
399 attrs);
e05ed4d1 400 if (map == SWIOTLB_MAP_ERROR)
4d048dbc 401 return XEN_SWIOTLB_ERROR_CODE;
b097186f 402
f1225ee4 403 dev_addr = xen_phys_to_bus(map);
6cf05463 404 xen_dma_map_page(dev, pfn_to_page(map >> PAGE_SHIFT),
a0f2dee0 405 dev_addr, map & ~PAGE_MASK, size, dir, attrs);
b097186f
KRW
406
407 /*
408 * Ensure that the address returned is DMA'ble
409 */
76418421
AD
410 if (dma_capable(dev, dev_addr, size))
411 return dev_addr;
412
d29fa0cb
AD
413 attrs |= DMA_ATTR_SKIP_CPU_SYNC;
414 swiotlb_tbl_unmap_single(dev, map, size, dir, attrs);
76418421 415
4d048dbc 416 return XEN_SWIOTLB_ERROR_CODE;
b097186f 417}
b097186f
KRW
418
419/*
420 * Unmap a single streaming mode DMA translation. The dma_addr and size must
421 * match what was provided for in a previous xen_swiotlb_map_page call. All
422 * other usages are undefined.
423 *
424 * After this call, reads by the cpu to the buffer are guaranteed to see
425 * whatever the device wrote there.
426 */
427static void xen_unmap_single(struct device *hwdev, dma_addr_t dev_addr,
6cf05463 428 size_t size, enum dma_data_direction dir,
00085f1e 429 unsigned long attrs)
b097186f
KRW
430{
431 phys_addr_t paddr = xen_bus_to_phys(dev_addr);
432
433 BUG_ON(dir == DMA_NONE);
434
d6883e6f 435 xen_dma_unmap_page(hwdev, dev_addr, size, dir, attrs);
6cf05463 436
b097186f
KRW
437 /* NOTE: We use dev_addr here, not paddr! */
438 if (is_xen_swiotlb_buffer(dev_addr)) {
0443fa00 439 swiotlb_tbl_unmap_single(hwdev, paddr, size, dir, attrs);
b097186f
KRW
440 return;
441 }
442
443 if (dir != DMA_FROM_DEVICE)
444 return;
445
446 /*
447 * phys_to_virt doesn't work with hihgmem page but we could
448 * call dma_mark_clean() with hihgmem page here. However, we
449 * are fine since dma_mark_clean() is null on POWERPC. We can
450 * make dma_mark_clean() take a physical address if necessary.
451 */
452 dma_mark_clean(phys_to_virt(paddr), size);
453}
454
dceb1a68 455static void xen_swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr,
b097186f 456 size_t size, enum dma_data_direction dir,
00085f1e 457 unsigned long attrs)
b097186f 458{
6cf05463 459 xen_unmap_single(hwdev, dev_addr, size, dir, attrs);
b097186f 460}
b097186f
KRW
461
462/*
463 * Make physical memory consistent for a single streaming mode DMA translation
464 * after a transfer.
465 *
466 * If you perform a xen_swiotlb_map_page() but wish to interrogate the buffer
467 * using the cpu, yet do not wish to teardown the dma mapping, you must
468 * call this function before doing so. At the next point you give the dma
469 * address back to the card, you must first perform a
470 * xen_swiotlb_dma_sync_for_device, and then the device again owns the buffer
471 */
472static void
473xen_swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr,
474 size_t size, enum dma_data_direction dir,
475 enum dma_sync_target target)
476{
477 phys_addr_t paddr = xen_bus_to_phys(dev_addr);
478
479 BUG_ON(dir == DMA_NONE);
480
6cf05463 481 if (target == SYNC_FOR_CPU)
d6883e6f 482 xen_dma_sync_single_for_cpu(hwdev, dev_addr, size, dir);
6cf05463 483
b097186f 484 /* NOTE: We use dev_addr here, not paddr! */
6cf05463 485 if (is_xen_swiotlb_buffer(dev_addr))
fbfda893 486 swiotlb_tbl_sync_single(hwdev, paddr, size, dir, target);
6cf05463
SS
487
488 if (target == SYNC_FOR_DEVICE)
9490c6c6 489 xen_dma_sync_single_for_device(hwdev, dev_addr, size, dir);
b097186f
KRW
490
491 if (dir != DMA_FROM_DEVICE)
492 return;
493
494 dma_mark_clean(phys_to_virt(paddr), size);
495}
496
497void
498xen_swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr,
499 size_t size, enum dma_data_direction dir)
500{
501 xen_swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_CPU);
502}
b097186f
KRW
503
504void
505xen_swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr,
506 size_t size, enum dma_data_direction dir)
507{
508 xen_swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_DEVICE);
509}
dceb1a68
CH
510
511/*
512 * Unmap a set of streaming mode DMA translations. Again, cpu read rules
513 * concerning calls here are the same as for swiotlb_unmap_page() above.
514 */
515static void
516xen_swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl,
517 int nelems, enum dma_data_direction dir,
518 unsigned long attrs)
519{
520 struct scatterlist *sg;
521 int i;
522
523 BUG_ON(dir == DMA_NONE);
524
525 for_each_sg(sgl, sg, nelems, i)
526 xen_unmap_single(hwdev, sg->dma_address, sg_dma_len(sg), dir, attrs);
527
528}
b097186f
KRW
529
530/*
531 * Map a set of buffers described by scatterlist in streaming mode for DMA.
532 * This is the scatter-gather version of the above xen_swiotlb_map_page
533 * interface. Here the scatter gather list elements are each tagged with the
534 * appropriate dma address and length. They are obtained via
535 * sg_dma_{address,length}(SG).
536 *
537 * NOTE: An implementation may be able to use a smaller number of
538 * DMA address/length pairs than there are SG table elements.
539 * (for example via virtual mapping capabilities)
540 * The routine returns the number of addr/length pairs actually
541 * used, at most nents.
542 *
543 * Device ownership issues as mentioned above for xen_swiotlb_map_page are the
544 * same here.
545 */
dceb1a68 546static int
b097186f
KRW
547xen_swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl,
548 int nelems, enum dma_data_direction dir,
00085f1e 549 unsigned long attrs)
b097186f
KRW
550{
551 struct scatterlist *sg;
552 int i;
553
554 BUG_ON(dir == DMA_NONE);
555
556 for_each_sg(sgl, sg, nelems, i) {
557 phys_addr_t paddr = sg_phys(sg);
558 dma_addr_t dev_addr = xen_phys_to_bus(paddr);
559
ae7871be 560 if (swiotlb_force == SWIOTLB_FORCE ||
291be10f 561 xen_arch_need_swiotlb(hwdev, paddr, dev_addr) ||
b097186f
KRW
562 !dma_capable(hwdev, dev_addr, sg->length) ||
563 range_straddles_page_boundary(paddr, sg->length)) {
e05ed4d1
AD
564 phys_addr_t map = swiotlb_tbl_map_single(hwdev,
565 start_dma_addr,
566 sg_phys(sg),
567 sg->length,
0443fa00 568 dir, attrs);
e05ed4d1 569 if (map == SWIOTLB_MAP_ERROR) {
783d0281 570 dev_warn(hwdev, "swiotlb buffer is full\n");
b097186f
KRW
571 /* Don't panic here, we expect map_sg users
572 to do proper error handling. */
0443fa00 573 attrs |= DMA_ATTR_SKIP_CPU_SYNC;
b097186f
KRW
574 xen_swiotlb_unmap_sg_attrs(hwdev, sgl, i, dir,
575 attrs);
781575cd 576 sg_dma_len(sgl) = 0;
15177608 577 return 0;
b097186f 578 }
f1225ee4 579 dev_addr = xen_phys_to_bus(map);
71bfae90 580 xen_dma_map_page(hwdev, pfn_to_page(map >> PAGE_SHIFT),
a0f2dee0 581 dev_addr,
71bfae90
SS
582 map & ~PAGE_MASK,
583 sg->length,
584 dir,
585 attrs);
f1225ee4 586 sg->dma_address = dev_addr;
6cf05463
SS
587 } else {
588 /* we are not interested in the dma_addr returned by
589 * xen_dma_map_page, only in the potential cache flushes executed
590 * by the function. */
591 xen_dma_map_page(hwdev, pfn_to_page(paddr >> PAGE_SHIFT),
a0f2dee0 592 dev_addr,
6cf05463
SS
593 paddr & ~PAGE_MASK,
594 sg->length,
595 dir,
596 attrs);
b097186f 597 sg->dma_address = dev_addr;
6cf05463 598 }
781575cd 599 sg_dma_len(sg) = sg->length;
b097186f
KRW
600 }
601 return nelems;
602}
b097186f 603
b097186f
KRW
604/*
605 * Make physical memory consistent for a set of streaming mode DMA translations
606 * after a transfer.
607 *
608 * The same as swiotlb_sync_single_* but for a scatter-gather list, same rules
609 * and usage.
610 */
611static void
612xen_swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl,
613 int nelems, enum dma_data_direction dir,
614 enum dma_sync_target target)
615{
616 struct scatterlist *sg;
617 int i;
618
619 for_each_sg(sgl, sg, nelems, i)
620 xen_swiotlb_sync_single(hwdev, sg->dma_address,
781575cd 621 sg_dma_len(sg), dir, target);
b097186f
KRW
622}
623
dceb1a68 624static void
b097186f
KRW
625xen_swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg,
626 int nelems, enum dma_data_direction dir)
627{
628 xen_swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_CPU);
629}
b097186f 630
dceb1a68 631static void
b097186f
KRW
632xen_swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg,
633 int nelems, enum dma_data_direction dir)
634{
635 xen_swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_DEVICE);
636}
b097186f 637
b097186f
KRW
638/*
639 * Return whether the given device DMA address mask can be supported
640 * properly. For example, if your device can only drive the low 24-bits
641 * during bus mastering, then you would pass 0x00ffffff as the mask to
642 * this function.
643 */
dceb1a68 644static int
b097186f
KRW
645xen_swiotlb_dma_supported(struct device *hwdev, u64 mask)
646{
647 return xen_virt_to_bus(xen_io_tlb_end - 1) <= mask;
648}
eb1ddc00 649
7e91c7df
SS
650/*
651 * Create userspace mapping for the DMA-coherent memory.
652 * This function should be called with the pages from the current domain only,
653 * passing pages mapped from other domains would lead to memory corruption.
654 */
dceb1a68 655static int
7e91c7df
SS
656xen_swiotlb_dma_mmap(struct device *dev, struct vm_area_struct *vma,
657 void *cpu_addr, dma_addr_t dma_addr, size_t size,
658 unsigned long attrs)
659{
660#if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
d5ff5061
SS
661 if (xen_get_dma_ops(dev)->mmap)
662 return xen_get_dma_ops(dev)->mmap(dev, vma, cpu_addr,
7e91c7df
SS
663 dma_addr, size, attrs);
664#endif
665 return dma_common_mmap(dev, vma, cpu_addr, dma_addr, size);
666}
69369f52
AA
667
668/*
669 * This function should be called with the pages from the current domain only,
670 * passing pages mapped from other domains would lead to memory corruption.
671 */
dceb1a68 672static int
69369f52
AA
673xen_swiotlb_get_sgtable(struct device *dev, struct sg_table *sgt,
674 void *cpu_addr, dma_addr_t handle, size_t size,
675 unsigned long attrs)
676{
677#if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
d5ff5061 678 if (xen_get_dma_ops(dev)->get_sgtable) {
69369f52
AA
679#if 0
680 /*
681 * This check verifies that the page belongs to the current domain and
682 * is not one mapped from another domain.
683 * This check is for debug only, and should not go to production build
684 */
685 unsigned long bfn = PHYS_PFN(dma_to_phys(dev, handle));
686 BUG_ON (!page_is_ram(bfn));
687#endif
d5ff5061 688 return xen_get_dma_ops(dev)->get_sgtable(dev, sgt, cpu_addr,
69369f52
AA
689 handle, size, attrs);
690 }
691#endif
692 return dma_common_get_sgtable(dev, sgt, cpu_addr, handle, size);
693}
dceb1a68 694
4d048dbc
CH
695static int xen_swiotlb_mapping_error(struct device *dev, dma_addr_t dma_addr)
696{
697 return dma_addr == XEN_SWIOTLB_ERROR_CODE;
698}
699
dceb1a68
CH
700const struct dma_map_ops xen_swiotlb_dma_ops = {
701 .alloc = xen_swiotlb_alloc_coherent,
702 .free = xen_swiotlb_free_coherent,
703 .sync_single_for_cpu = xen_swiotlb_sync_single_for_cpu,
704 .sync_single_for_device = xen_swiotlb_sync_single_for_device,
705 .sync_sg_for_cpu = xen_swiotlb_sync_sg_for_cpu,
706 .sync_sg_for_device = xen_swiotlb_sync_sg_for_device,
707 .map_sg = xen_swiotlb_map_sg_attrs,
708 .unmap_sg = xen_swiotlb_unmap_sg_attrs,
709 .map_page = xen_swiotlb_map_page,
710 .unmap_page = xen_swiotlb_unmap_page,
711 .dma_supported = xen_swiotlb_dma_supported,
dceb1a68
CH
712 .mmap = xen_swiotlb_dma_mmap,
713 .get_sgtable = xen_swiotlb_get_sgtable,
4d048dbc 714 .mapping_error = xen_swiotlb_mapping_error,
dceb1a68 715};