2 * Dynamic DMA mapping support.
4 * This implementation is a fallback for platforms that do not support
5 * I/O TLBs (aka DMA address translation hardware).
6 * Copyright (C) 2000 Asit Mallick <Asit.K.Mallick@intel.com>
7 * Copyright (C) 2000 Goutham Rao <goutham.rao@intel.com>
8 * Copyright (C) 2000, 2003 Hewlett-Packard Co
9 * David Mosberger-Tang <davidm@hpl.hp.com>
11 * 03/05/07 davidm Switch from PCI-DMA to generic device DMA API.
12 * 00/12/13 davidm Rename to swiotlb.c and add mark_clean() to avoid
13 * unnecessary i-cache flushing.
14 * 04/07/.. ak Better overflow handling. Assorted fixes.
15 * 05/09/10 linville Add support for syncing ranges, support syncing for
16 * DMA_BIDIRECTIONAL mappings, miscellaneous cleanup.
17 * 08/12/11 beckyb Add highmem support
20 #include <linux/cache.h>
21 #include <linux/dma-mapping.h>
23 #include <linux/module.h>
24 #include <linux/spinlock.h>
25 #include <linux/string.h>
26 #include <linux/swiotlb.h>
27 #include <linux/pfn.h>
28 #include <linux/types.h>
29 #include <linux/ctype.h>
30 #include <linux/highmem.h>
34 #include <asm/scatterlist.h>
36 #include <linux/init.h>
37 #include <linux/bootmem.h>
38 #include <linux/iommu-helper.h>
40 #define OFFSET(val,align) ((unsigned long) \
41 ( (val) & ( (align) - 1)))
43 #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT))
46 * Minimum IO TLB size to bother booting with. Systems with mainly
47 * 64bit capable cards will only lightly use the swiotlb. If we can't
48 * allocate a contiguous 1MB, we're probably in trouble anyway.
50 #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT)
53 * Enumeration for sync targets
55 enum dma_sync_target
{
63 * Used to do a quick range check in unmap_single and
64 * sync_single_*, to see if the memory was in fact allocated by this
67 static char *io_tlb_start
, *io_tlb_end
;
70 * The number of IO TLB blocks (in groups of 64) betweeen io_tlb_start and
71 * io_tlb_end. This is command line adjustable via setup_io_tlb_npages.
73 static unsigned long io_tlb_nslabs
;
76 * When the IOMMU overflows we return a fallback buffer. This sets the size.
78 static unsigned long io_tlb_overflow
= 32*1024;
80 void *io_tlb_overflow_buffer
;
83 * This is a free list describing the number of free entries available from
86 static unsigned int *io_tlb_list
;
87 static unsigned int io_tlb_index
;
90 * We need to save away the original address corresponding to a mapped entry
91 * for the sync operations.
93 static phys_addr_t
*io_tlb_orig_addr
;
96 * Protect the above data structures in the map and unmap calls
98 static DEFINE_SPINLOCK(io_tlb_lock
);
101 setup_io_tlb_npages(char *str
)
104 io_tlb_nslabs
= simple_strtoul(str
, &str
, 0);
105 /* avoid tail segment of size < IO_TLB_SEGSIZE */
106 io_tlb_nslabs
= ALIGN(io_tlb_nslabs
, IO_TLB_SEGSIZE
);
110 if (!strcmp(str
, "force"))
114 __setup("swiotlb=", setup_io_tlb_npages
);
115 /* make io_tlb_overflow tunable too? */
117 dma_addr_t __weak
swiotlb_phys_to_bus(struct device
*hwdev
, phys_addr_t paddr
)
122 phys_addr_t __weak
swiotlb_bus_to_phys(struct device
*hwdev
, dma_addr_t baddr
)
127 static dma_addr_t
swiotlb_virt_to_bus(struct device
*hwdev
,
128 volatile void *address
)
130 return swiotlb_phys_to_bus(hwdev
, virt_to_phys(address
));
133 void * __weak
swiotlb_bus_to_virt(struct device
*hwdev
, dma_addr_t address
)
135 return phys_to_virt(swiotlb_bus_to_phys(hwdev
, address
));
138 int __weak
swiotlb_arch_address_needs_mapping(struct device
*hwdev
,
139 dma_addr_t addr
, size_t size
)
141 return !is_buffer_dma_capable(dma_get_mask(hwdev
), addr
, size
);
144 static void swiotlb_print_info(unsigned long bytes
)
146 phys_addr_t pstart
, pend
;
148 pstart
= virt_to_phys(io_tlb_start
);
149 pend
= virt_to_phys(io_tlb_end
);
151 printk(KERN_INFO
"Placing %luMB software IO TLB between %p - %p\n",
152 bytes
>> 20, io_tlb_start
, io_tlb_end
);
153 printk(KERN_INFO
"software IO TLB at phys %#llx - %#llx\n",
154 (unsigned long long)pstart
,
155 (unsigned long long)pend
);
159 * Statically reserve bounce buffer space and initialize bounce buffer data
160 * structures for the software IO TLB used to implement the DMA API.
163 swiotlb_init_with_default_size(size_t default_size
)
165 unsigned long i
, bytes
;
167 if (!io_tlb_nslabs
) {
168 io_tlb_nslabs
= (default_size
>> IO_TLB_SHIFT
);
169 io_tlb_nslabs
= ALIGN(io_tlb_nslabs
, IO_TLB_SEGSIZE
);
172 bytes
= io_tlb_nslabs
<< IO_TLB_SHIFT
;
175 * Get IO TLB memory from the low pages
177 io_tlb_start
= alloc_bootmem_low_pages(bytes
);
179 panic("Cannot allocate SWIOTLB buffer");
180 io_tlb_end
= io_tlb_start
+ bytes
;
183 * Allocate and initialize the free list array. This array is used
184 * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE
185 * between io_tlb_start and io_tlb_end.
187 io_tlb_list
= alloc_bootmem(io_tlb_nslabs
* sizeof(int));
188 for (i
= 0; i
< io_tlb_nslabs
; i
++)
189 io_tlb_list
[i
] = IO_TLB_SEGSIZE
- OFFSET(i
, IO_TLB_SEGSIZE
);
191 io_tlb_orig_addr
= alloc_bootmem(io_tlb_nslabs
* sizeof(phys_addr_t
));
194 * Get the overflow emergency buffer
196 io_tlb_overflow_buffer
= alloc_bootmem_low(io_tlb_overflow
);
197 if (!io_tlb_overflow_buffer
)
198 panic("Cannot allocate SWIOTLB overflow buffer!\n");
200 swiotlb_print_info(bytes
);
206 swiotlb_init_with_default_size(64 * (1<<20)); /* default to 64MB */
210 * Systems with larger DMA zones (those that don't support ISA) can
211 * initialize the swiotlb later using the slab allocator if needed.
212 * This should be just like above, but with some error catching.
215 swiotlb_late_init_with_default_size(size_t default_size
)
217 unsigned long i
, bytes
, req_nslabs
= io_tlb_nslabs
;
220 if (!io_tlb_nslabs
) {
221 io_tlb_nslabs
= (default_size
>> IO_TLB_SHIFT
);
222 io_tlb_nslabs
= ALIGN(io_tlb_nslabs
, IO_TLB_SEGSIZE
);
226 * Get IO TLB memory from the low pages
228 order
= get_order(io_tlb_nslabs
<< IO_TLB_SHIFT
);
229 io_tlb_nslabs
= SLABS_PER_PAGE
<< order
;
230 bytes
= io_tlb_nslabs
<< IO_TLB_SHIFT
;
232 while ((SLABS_PER_PAGE
<< order
) > IO_TLB_MIN_SLABS
) {
233 io_tlb_start
= (void *)__get_free_pages(GFP_DMA
| __GFP_NOWARN
,
243 if (order
!= get_order(bytes
)) {
244 printk(KERN_WARNING
"Warning: only able to allocate %ld MB "
245 "for software IO TLB\n", (PAGE_SIZE
<< order
) >> 20);
246 io_tlb_nslabs
= SLABS_PER_PAGE
<< order
;
247 bytes
= io_tlb_nslabs
<< IO_TLB_SHIFT
;
249 io_tlb_end
= io_tlb_start
+ bytes
;
250 memset(io_tlb_start
, 0, bytes
);
253 * Allocate and initialize the free list array. This array is used
254 * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE
255 * between io_tlb_start and io_tlb_end.
257 io_tlb_list
= (unsigned int *)__get_free_pages(GFP_KERNEL
,
258 get_order(io_tlb_nslabs
* sizeof(int)));
262 for (i
= 0; i
< io_tlb_nslabs
; i
++)
263 io_tlb_list
[i
] = IO_TLB_SEGSIZE
- OFFSET(i
, IO_TLB_SEGSIZE
);
266 io_tlb_orig_addr
= (phys_addr_t
*)
267 __get_free_pages(GFP_KERNEL
,
268 get_order(io_tlb_nslabs
*
269 sizeof(phys_addr_t
)));
270 if (!io_tlb_orig_addr
)
273 memset(io_tlb_orig_addr
, 0, io_tlb_nslabs
* sizeof(phys_addr_t
));
276 * Get the overflow emergency buffer
278 io_tlb_overflow_buffer
= (void *)__get_free_pages(GFP_DMA
,
279 get_order(io_tlb_overflow
));
280 if (!io_tlb_overflow_buffer
)
283 swiotlb_print_info(bytes
);
288 free_pages((unsigned long)io_tlb_orig_addr
,
289 get_order(io_tlb_nslabs
* sizeof(phys_addr_t
)));
290 io_tlb_orig_addr
= NULL
;
292 free_pages((unsigned long)io_tlb_list
, get_order(io_tlb_nslabs
*
297 free_pages((unsigned long)io_tlb_start
, order
);
300 io_tlb_nslabs
= req_nslabs
;
305 address_needs_mapping(struct device
*hwdev
, dma_addr_t addr
, size_t size
)
307 return swiotlb_arch_address_needs_mapping(hwdev
, addr
, size
);
310 static int is_swiotlb_buffer(char *addr
)
312 return addr
>= io_tlb_start
&& addr
< io_tlb_end
;
316 * Bounce: copy the swiotlb buffer back to the original dma location
318 static void swiotlb_bounce(phys_addr_t phys
, char *dma_addr
, size_t size
,
319 enum dma_data_direction dir
)
321 unsigned long pfn
= PFN_DOWN(phys
);
323 if (PageHighMem(pfn_to_page(pfn
))) {
324 /* The buffer does not have a mapping. Map it in and copy */
325 unsigned int offset
= phys
& ~PAGE_MASK
;
331 sz
= min_t(size_t, PAGE_SIZE
- offset
, size
);
333 local_irq_save(flags
);
334 buffer
= kmap_atomic(pfn_to_page(pfn
),
336 if (dir
== DMA_TO_DEVICE
)
337 memcpy(dma_addr
, buffer
+ offset
, sz
);
339 memcpy(buffer
+ offset
, dma_addr
, sz
);
340 kunmap_atomic(buffer
, KM_BOUNCE_READ
);
341 local_irq_restore(flags
);
349 if (dir
== DMA_TO_DEVICE
)
350 memcpy(dma_addr
, phys_to_virt(phys
), size
);
352 memcpy(phys_to_virt(phys
), dma_addr
, size
);
357 * Allocates bounce buffer and returns its kernel virtual address.
360 map_single(struct device
*hwdev
, phys_addr_t phys
, size_t size
, int dir
)
364 unsigned int nslots
, stride
, index
, wrap
;
366 unsigned long start_dma_addr
;
368 unsigned long offset_slots
;
369 unsigned long max_slots
;
371 mask
= dma_get_seg_boundary(hwdev
);
372 start_dma_addr
= swiotlb_virt_to_bus(hwdev
, io_tlb_start
) & mask
;
374 offset_slots
= ALIGN(start_dma_addr
, 1 << IO_TLB_SHIFT
) >> IO_TLB_SHIFT
;
377 * Carefully handle integer overflow which can occur when mask == ~0UL.
380 ? ALIGN(mask
+ 1, 1 << IO_TLB_SHIFT
) >> IO_TLB_SHIFT
381 : 1UL << (BITS_PER_LONG
- IO_TLB_SHIFT
);
384 * For mappings greater than a page, we limit the stride (and
385 * hence alignment) to a page size.
387 nslots
= ALIGN(size
, 1 << IO_TLB_SHIFT
) >> IO_TLB_SHIFT
;
388 if (size
> PAGE_SIZE
)
389 stride
= (1 << (PAGE_SHIFT
- IO_TLB_SHIFT
));
396 * Find suitable number of IO TLB entries size that will fit this
397 * request and allocate a buffer from that IO TLB pool.
399 spin_lock_irqsave(&io_tlb_lock
, flags
);
400 index
= ALIGN(io_tlb_index
, stride
);
401 if (index
>= io_tlb_nslabs
)
406 while (iommu_is_span_boundary(index
, nslots
, offset_slots
,
409 if (index
>= io_tlb_nslabs
)
416 * If we find a slot that indicates we have 'nslots' number of
417 * contiguous buffers, we allocate the buffers from that slot
418 * and mark the entries as '0' indicating unavailable.
420 if (io_tlb_list
[index
] >= nslots
) {
423 for (i
= index
; i
< (int) (index
+ nslots
); i
++)
425 for (i
= index
- 1; (OFFSET(i
, IO_TLB_SEGSIZE
) != IO_TLB_SEGSIZE
- 1) && io_tlb_list
[i
]; i
--)
426 io_tlb_list
[i
] = ++count
;
427 dma_addr
= io_tlb_start
+ (index
<< IO_TLB_SHIFT
);
430 * Update the indices to avoid searching in the next
433 io_tlb_index
= ((index
+ nslots
) < io_tlb_nslabs
434 ? (index
+ nslots
) : 0);
439 if (index
>= io_tlb_nslabs
)
441 } while (index
!= wrap
);
444 spin_unlock_irqrestore(&io_tlb_lock
, flags
);
447 spin_unlock_irqrestore(&io_tlb_lock
, flags
);
450 * Save away the mapping from the original address to the DMA address.
451 * This is needed when we sync the memory. Then we sync the buffer if
454 for (i
= 0; i
< nslots
; i
++)
455 io_tlb_orig_addr
[index
+i
] = phys
+ (i
<< IO_TLB_SHIFT
);
456 if (dir
== DMA_TO_DEVICE
|| dir
== DMA_BIDIRECTIONAL
)
457 swiotlb_bounce(phys
, dma_addr
, size
, DMA_TO_DEVICE
);
463 * dma_addr is the kernel virtual address of the bounce buffer to unmap.
466 do_unmap_single(struct device
*hwdev
, char *dma_addr
, size_t size
, int dir
)
469 int i
, count
, nslots
= ALIGN(size
, 1 << IO_TLB_SHIFT
) >> IO_TLB_SHIFT
;
470 int index
= (dma_addr
- io_tlb_start
) >> IO_TLB_SHIFT
;
471 phys_addr_t phys
= io_tlb_orig_addr
[index
];
474 * First, sync the memory before unmapping the entry
476 if (phys
&& ((dir
== DMA_FROM_DEVICE
) || (dir
== DMA_BIDIRECTIONAL
)))
477 swiotlb_bounce(phys
, dma_addr
, size
, DMA_FROM_DEVICE
);
480 * Return the buffer to the free list by setting the corresponding
481 * entries to indicate the number of contigous entries available.
482 * While returning the entries to the free list, we merge the entries
483 * with slots below and above the pool being returned.
485 spin_lock_irqsave(&io_tlb_lock
, flags
);
487 count
= ((index
+ nslots
) < ALIGN(index
+ 1, IO_TLB_SEGSIZE
) ?
488 io_tlb_list
[index
+ nslots
] : 0);
490 * Step 1: return the slots to the free list, merging the
491 * slots with superceeding slots
493 for (i
= index
+ nslots
- 1; i
>= index
; i
--)
494 io_tlb_list
[i
] = ++count
;
496 * Step 2: merge the returned slots with the preceding slots,
497 * if available (non zero)
499 for (i
= index
- 1; (OFFSET(i
, IO_TLB_SEGSIZE
) != IO_TLB_SEGSIZE
-1) && io_tlb_list
[i
]; i
--)
500 io_tlb_list
[i
] = ++count
;
502 spin_unlock_irqrestore(&io_tlb_lock
, flags
);
506 sync_single(struct device
*hwdev
, char *dma_addr
, size_t size
,
509 int index
= (dma_addr
- io_tlb_start
) >> IO_TLB_SHIFT
;
510 phys_addr_t phys
= io_tlb_orig_addr
[index
];
512 phys
+= ((unsigned long)dma_addr
& ((1 << IO_TLB_SHIFT
) - 1));
516 if (likely(dir
== DMA_FROM_DEVICE
|| dir
== DMA_BIDIRECTIONAL
))
517 swiotlb_bounce(phys
, dma_addr
, size
, DMA_FROM_DEVICE
);
519 BUG_ON(dir
!= DMA_TO_DEVICE
);
521 case SYNC_FOR_DEVICE
:
522 if (likely(dir
== DMA_TO_DEVICE
|| dir
== DMA_BIDIRECTIONAL
))
523 swiotlb_bounce(phys
, dma_addr
, size
, DMA_TO_DEVICE
);
525 BUG_ON(dir
!= DMA_FROM_DEVICE
);
533 swiotlb_alloc_coherent(struct device
*hwdev
, size_t size
,
534 dma_addr_t
*dma_handle
, gfp_t flags
)
538 int order
= get_order(size
);
539 u64 dma_mask
= DMA_BIT_MASK(32);
541 if (hwdev
&& hwdev
->coherent_dma_mask
)
542 dma_mask
= hwdev
->coherent_dma_mask
;
544 ret
= (void *)__get_free_pages(flags
, order
);
546 !is_buffer_dma_capable(dma_mask
, swiotlb_virt_to_bus(hwdev
, ret
),
549 * The allocated memory isn't reachable by the device.
551 free_pages((unsigned long) ret
, order
);
556 * We are either out of memory or the device can't DMA
557 * to GFP_DMA memory; fall back on map_single(), which
558 * will grab memory from the lowest available address range.
560 ret
= map_single(hwdev
, 0, size
, DMA_FROM_DEVICE
);
565 memset(ret
, 0, size
);
566 dev_addr
= swiotlb_virt_to_bus(hwdev
, ret
);
568 /* Confirm address can be DMA'd by device */
569 if (!is_buffer_dma_capable(dma_mask
, dev_addr
, size
)) {
570 printk("hwdev DMA mask = 0x%016Lx, dev_addr = 0x%016Lx\n",
571 (unsigned long long)dma_mask
,
572 (unsigned long long)dev_addr
);
574 /* DMA_TO_DEVICE to avoid memcpy in unmap_single */
575 do_unmap_single(hwdev
, ret
, size
, DMA_TO_DEVICE
);
578 *dma_handle
= dev_addr
;
581 EXPORT_SYMBOL(swiotlb_alloc_coherent
);
584 swiotlb_free_coherent(struct device
*hwdev
, size_t size
, void *vaddr
,
585 dma_addr_t dma_handle
)
587 WARN_ON(irqs_disabled());
588 if (!is_swiotlb_buffer(vaddr
))
589 free_pages((unsigned long) vaddr
, get_order(size
));
591 /* DMA_TO_DEVICE to avoid memcpy in unmap_single */
592 do_unmap_single(hwdev
, vaddr
, size
, DMA_TO_DEVICE
);
594 EXPORT_SYMBOL(swiotlb_free_coherent
);
597 swiotlb_full(struct device
*dev
, size_t size
, int dir
, int do_panic
)
600 * Ran out of IOMMU space for this operation. This is very bad.
601 * Unfortunately the drivers cannot handle this operation properly.
602 * unless they check for dma_mapping_error (most don't)
603 * When the mapping is small enough return a static buffer to limit
604 * the damage, or panic when the transfer is too big.
606 printk(KERN_ERR
"DMA: Out of SW-IOMMU space for %zu bytes at "
607 "device %s\n", size
, dev
? dev_name(dev
) : "?");
609 if (size
> io_tlb_overflow
&& do_panic
) {
610 if (dir
== DMA_FROM_DEVICE
|| dir
== DMA_BIDIRECTIONAL
)
611 panic("DMA: Memory would be corrupted\n");
612 if (dir
== DMA_TO_DEVICE
|| dir
== DMA_BIDIRECTIONAL
)
613 panic("DMA: Random memory would be DMAed\n");
618 * Map a single buffer of the indicated size for DMA in streaming mode. The
619 * physical address to use is returned.
621 * Once the device is given the dma address, the device owns this memory until
622 * either swiotlb_unmap_page or swiotlb_dma_sync_single is performed.
624 dma_addr_t
swiotlb_map_page(struct device
*dev
, struct page
*page
,
625 unsigned long offset
, size_t size
,
626 enum dma_data_direction dir
,
627 struct dma_attrs
*attrs
)
629 phys_addr_t phys
= page_to_phys(page
) + offset
;
630 dma_addr_t dev_addr
= swiotlb_phys_to_bus(dev
, phys
);
633 BUG_ON(dir
== DMA_NONE
);
635 * If the address happens to be in the device's DMA window,
636 * we can safely return the device addr and not worry about bounce
639 if (!address_needs_mapping(dev
, dev_addr
, size
) && !swiotlb_force
)
643 * Oh well, have to allocate and map a bounce buffer.
645 map
= map_single(dev
, phys
, size
, dir
);
647 swiotlb_full(dev
, size
, dir
, 1);
648 map
= io_tlb_overflow_buffer
;
651 dev_addr
= swiotlb_virt_to_bus(dev
, map
);
654 * Ensure that the address returned is DMA'ble
656 if (address_needs_mapping(dev
, dev_addr
, size
))
657 panic("map_single: bounce buffer is not DMA'ble");
661 EXPORT_SYMBOL_GPL(swiotlb_map_page
);
664 * Unmap a single streaming mode DMA translation. The dma_addr and size must
665 * match what was provided for in a previous swiotlb_map_page call. All
666 * other usages are undefined.
668 * After this call, reads by the cpu to the buffer are guaranteed to see
669 * whatever the device wrote there.
671 static void unmap_single(struct device
*hwdev
, dma_addr_t dev_addr
,
672 size_t size
, int dir
)
674 char *dma_addr
= swiotlb_bus_to_virt(hwdev
, dev_addr
);
676 BUG_ON(dir
== DMA_NONE
);
678 if (is_swiotlb_buffer(dma_addr
)) {
679 do_unmap_single(hwdev
, dma_addr
, size
, dir
);
683 if (dir
!= DMA_FROM_DEVICE
)
686 dma_mark_clean(dma_addr
, size
);
689 void swiotlb_unmap_page(struct device
*hwdev
, dma_addr_t dev_addr
,
690 size_t size
, enum dma_data_direction dir
,
691 struct dma_attrs
*attrs
)
693 unmap_single(hwdev
, dev_addr
, size
, dir
);
695 EXPORT_SYMBOL_GPL(swiotlb_unmap_page
);
698 * Make physical memory consistent for a single streaming mode DMA translation
701 * If you perform a swiotlb_map_page() but wish to interrogate the buffer
702 * using the cpu, yet do not wish to teardown the dma mapping, you must
703 * call this function before doing so. At the next point you give the dma
704 * address back to the card, you must first perform a
705 * swiotlb_dma_sync_for_device, and then the device again owns the buffer
708 swiotlb_sync_single(struct device
*hwdev
, dma_addr_t dev_addr
,
709 size_t size
, int dir
, int target
)
711 char *dma_addr
= swiotlb_bus_to_virt(hwdev
, dev_addr
);
713 BUG_ON(dir
== DMA_NONE
);
715 if (is_swiotlb_buffer(dma_addr
)) {
716 sync_single(hwdev
, dma_addr
, size
, dir
, target
);
720 if (dir
!= DMA_FROM_DEVICE
)
723 dma_mark_clean(dma_addr
, size
);
727 swiotlb_sync_single_for_cpu(struct device
*hwdev
, dma_addr_t dev_addr
,
728 size_t size
, enum dma_data_direction dir
)
730 swiotlb_sync_single(hwdev
, dev_addr
, size
, dir
, SYNC_FOR_CPU
);
732 EXPORT_SYMBOL(swiotlb_sync_single_for_cpu
);
735 swiotlb_sync_single_for_device(struct device
*hwdev
, dma_addr_t dev_addr
,
736 size_t size
, enum dma_data_direction dir
)
738 swiotlb_sync_single(hwdev
, dev_addr
, size
, dir
, SYNC_FOR_DEVICE
);
740 EXPORT_SYMBOL(swiotlb_sync_single_for_device
);
743 * Same as above, but for a sub-range of the mapping.
746 swiotlb_sync_single_range(struct device
*hwdev
, dma_addr_t dev_addr
,
747 unsigned long offset
, size_t size
,
750 swiotlb_sync_single(hwdev
, dev_addr
+ offset
, size
, dir
, target
);
754 swiotlb_sync_single_range_for_cpu(struct device
*hwdev
, dma_addr_t dev_addr
,
755 unsigned long offset
, size_t size
,
756 enum dma_data_direction dir
)
758 swiotlb_sync_single_range(hwdev
, dev_addr
, offset
, size
, dir
,
761 EXPORT_SYMBOL_GPL(swiotlb_sync_single_range_for_cpu
);
764 swiotlb_sync_single_range_for_device(struct device
*hwdev
, dma_addr_t dev_addr
,
765 unsigned long offset
, size_t size
,
766 enum dma_data_direction dir
)
768 swiotlb_sync_single_range(hwdev
, dev_addr
, offset
, size
, dir
,
771 EXPORT_SYMBOL_GPL(swiotlb_sync_single_range_for_device
);
774 * Map a set of buffers described by scatterlist in streaming mode for DMA.
775 * This is the scatter-gather version of the above swiotlb_map_page
776 * interface. Here the scatter gather list elements are each tagged with the
777 * appropriate dma address and length. They are obtained via
778 * sg_dma_{address,length}(SG).
780 * NOTE: An implementation may be able to use a smaller number of
781 * DMA address/length pairs than there are SG table elements.
782 * (for example via virtual mapping capabilities)
783 * The routine returns the number of addr/length pairs actually
784 * used, at most nents.
786 * Device ownership issues as mentioned above for swiotlb_map_page are the
790 swiotlb_map_sg_attrs(struct device
*hwdev
, struct scatterlist
*sgl
, int nelems
,
791 enum dma_data_direction dir
, struct dma_attrs
*attrs
)
793 struct scatterlist
*sg
;
796 BUG_ON(dir
== DMA_NONE
);
798 for_each_sg(sgl
, sg
, nelems
, i
) {
799 phys_addr_t paddr
= sg_phys(sg
);
800 dma_addr_t dev_addr
= swiotlb_phys_to_bus(hwdev
, paddr
);
803 address_needs_mapping(hwdev
, dev_addr
, sg
->length
)) {
804 void *map
= map_single(hwdev
, sg_phys(sg
),
807 /* Don't panic here, we expect map_sg users
808 to do proper error handling. */
809 swiotlb_full(hwdev
, sg
->length
, dir
, 0);
810 swiotlb_unmap_sg_attrs(hwdev
, sgl
, i
, dir
,
812 sgl
[0].dma_length
= 0;
815 sg
->dma_address
= swiotlb_virt_to_bus(hwdev
, map
);
817 sg
->dma_address
= dev_addr
;
818 sg
->dma_length
= sg
->length
;
822 EXPORT_SYMBOL(swiotlb_map_sg_attrs
);
825 swiotlb_map_sg(struct device
*hwdev
, struct scatterlist
*sgl
, int nelems
,
828 return swiotlb_map_sg_attrs(hwdev
, sgl
, nelems
, dir
, NULL
);
830 EXPORT_SYMBOL(swiotlb_map_sg
);
833 * Unmap a set of streaming mode DMA translations. Again, cpu read rules
834 * concerning calls here are the same as for swiotlb_unmap_page() above.
837 swiotlb_unmap_sg_attrs(struct device
*hwdev
, struct scatterlist
*sgl
,
838 int nelems
, enum dma_data_direction dir
, struct dma_attrs
*attrs
)
840 struct scatterlist
*sg
;
843 BUG_ON(dir
== DMA_NONE
);
845 for_each_sg(sgl
, sg
, nelems
, i
)
846 unmap_single(hwdev
, sg
->dma_address
, sg
->dma_length
, dir
);
849 EXPORT_SYMBOL(swiotlb_unmap_sg_attrs
);
852 swiotlb_unmap_sg(struct device
*hwdev
, struct scatterlist
*sgl
, int nelems
,
855 return swiotlb_unmap_sg_attrs(hwdev
, sgl
, nelems
, dir
, NULL
);
857 EXPORT_SYMBOL(swiotlb_unmap_sg
);
860 * Make physical memory consistent for a set of streaming mode DMA translations
863 * The same as swiotlb_sync_single_* but for a scatter-gather list, same rules
867 swiotlb_sync_sg(struct device
*hwdev
, struct scatterlist
*sgl
,
868 int nelems
, int dir
, int target
)
870 struct scatterlist
*sg
;
873 for_each_sg(sgl
, sg
, nelems
, i
)
874 swiotlb_sync_single(hwdev
, sg
->dma_address
,
875 sg
->dma_length
, dir
, target
);
879 swiotlb_sync_sg_for_cpu(struct device
*hwdev
, struct scatterlist
*sg
,
880 int nelems
, enum dma_data_direction dir
)
882 swiotlb_sync_sg(hwdev
, sg
, nelems
, dir
, SYNC_FOR_CPU
);
884 EXPORT_SYMBOL(swiotlb_sync_sg_for_cpu
);
887 swiotlb_sync_sg_for_device(struct device
*hwdev
, struct scatterlist
*sg
,
888 int nelems
, enum dma_data_direction dir
)
890 swiotlb_sync_sg(hwdev
, sg
, nelems
, dir
, SYNC_FOR_DEVICE
);
892 EXPORT_SYMBOL(swiotlb_sync_sg_for_device
);
895 swiotlb_dma_mapping_error(struct device
*hwdev
, dma_addr_t dma_addr
)
897 return (dma_addr
== swiotlb_virt_to_bus(hwdev
, io_tlb_overflow_buffer
));
899 EXPORT_SYMBOL(swiotlb_dma_mapping_error
);
902 * Return whether the given device DMA address mask can be supported
903 * properly. For example, if your device can only drive the low 24-bits
904 * during bus mastering, then you would pass 0x00ffffff as the mask to
908 swiotlb_dma_supported(struct device
*hwdev
, u64 mask
)
910 return swiotlb_virt_to_bus(hwdev
, io_tlb_end
- 1) <= mask
;
912 EXPORT_SYMBOL(swiotlb_dma_supported
);