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.
19 #include <linux/cache.h>
20 #include <linux/dma-mapping.h>
22 #include <linux/module.h>
23 #include <linux/spinlock.h>
24 #include <linux/swiotlb.h>
25 #include <linux/string.h>
26 #include <linux/swiotlb.h>
27 #include <linux/types.h>
28 #include <linux/ctype.h>
32 #include <asm/scatterlist.h>
34 #include <linux/init.h>
35 #include <linux/bootmem.h>
36 #include <linux/iommu-helper.h>
38 #define OFFSET(val,align) ((unsigned long) \
39 ( (val) & ( (align) - 1)))
41 #define SG_ENT_VIRT_ADDRESS(sg) (sg_virt((sg)))
42 #define SG_ENT_PHYS_ADDRESS(sg) virt_to_bus(SG_ENT_VIRT_ADDRESS(sg))
44 #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT))
47 * Minimum IO TLB size to bother booting with. Systems with mainly
48 * 64bit capable cards will only lightly use the swiotlb. If we can't
49 * allocate a contiguous 1MB, we're probably in trouble anyway.
51 #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT)
54 * Enumeration for sync targets
56 enum dma_sync_target {
64 * Used to do a quick range check in swiotlb_unmap_single and
65 * swiotlb_sync_single_*, to see if the memory was in fact allocated by this
68 static char *io_tlb_start, *io_tlb_end;
71 * The number of IO TLB blocks (in groups of 64) betweeen io_tlb_start and
72 * io_tlb_end. This is command line adjustable via setup_io_tlb_npages.
74 static unsigned long io_tlb_nslabs;
77 * When the IOMMU overflows we return a fallback buffer. This sets the size.
79 static unsigned long io_tlb_overflow = 32*1024;
81 void *io_tlb_overflow_buffer;
84 * This is a free list describing the number of free entries available from
87 static unsigned int *io_tlb_list;
88 static unsigned int io_tlb_index;
91 * We need to save away the original address corresponding to a mapped entry
92 * for the sync operations.
94 static unsigned char **io_tlb_orig_addr;
97 * Protect the above data structures in the map and unmap calls
99 static DEFINE_SPINLOCK(io_tlb_lock);
102 setup_io_tlb_npages(char *str)
105 io_tlb_nslabs = simple_strtoul(str, &str, 0);
106 /* avoid tail segment of size < IO_TLB_SEGSIZE */
107 io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
111 if (!strcmp(str, "force"))
115 __setup("swiotlb=", setup_io_tlb_npages);
116 /* make io_tlb_overflow tunable too? */
118 void * __weak swiotlb_alloc_boot(size_t size, unsigned long nslabs)
120 return alloc_bootmem_low_pages(size);
123 void * __weak swiotlb_alloc(unsigned order, unsigned long nslabs)
125 return (void *)__get_free_pages(GFP_DMA | __GFP_NOWARN, order);
128 dma_addr_t __weak swiotlb_phys_to_bus(phys_addr_t paddr)
133 phys_addr_t __weak swiotlb_bus_to_phys(dma_addr_t baddr)
138 static dma_addr_t swiotlb_virt_to_bus(volatile void *address)
140 return swiotlb_phys_to_bus(virt_to_phys(address));
143 static void *swiotlb_bus_to_virt(dma_addr_t address)
145 return phys_to_virt(swiotlb_bus_to_phys(address));
148 int __weak swiotlb_arch_range_needs_mapping(void *ptr, size_t size)
154 * Statically reserve bounce buffer space and initialize bounce buffer data
155 * structures for the software IO TLB used to implement the DMA API.
158 swiotlb_init_with_default_size(size_t default_size)
160 unsigned long i, bytes;
162 if (!io_tlb_nslabs) {
163 io_tlb_nslabs = (default_size >> IO_TLB_SHIFT);
164 io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
167 bytes = io_tlb_nslabs << IO_TLB_SHIFT;
170 * Get IO TLB memory from the low pages
172 io_tlb_start = swiotlb_alloc_boot(bytes, io_tlb_nslabs);
174 panic("Cannot allocate SWIOTLB buffer");
175 io_tlb_end = io_tlb_start + bytes;
178 * Allocate and initialize the free list array. This array is used
179 * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE
180 * between io_tlb_start and io_tlb_end.
182 io_tlb_list = alloc_bootmem(io_tlb_nslabs * sizeof(int));
183 for (i = 0; i < io_tlb_nslabs; i++)
184 io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE);
186 io_tlb_orig_addr = alloc_bootmem(io_tlb_nslabs * sizeof(char *));
189 * Get the overflow emergency buffer
191 io_tlb_overflow_buffer = alloc_bootmem_low(io_tlb_overflow);
192 if (!io_tlb_overflow_buffer)
193 panic("Cannot allocate SWIOTLB overflow buffer!\n");
195 printk(KERN_INFO "Placing software IO TLB between 0x%lx - 0x%lx\n",
196 swiotlb_virt_to_bus(io_tlb_start), swiotlb_virt_to_bus(io_tlb_end));
202 swiotlb_init_with_default_size(64 * (1<<20)); /* default to 64MB */
206 * Systems with larger DMA zones (those that don't support ISA) can
207 * initialize the swiotlb later using the slab allocator if needed.
208 * This should be just like above, but with some error catching.
211 swiotlb_late_init_with_default_size(size_t default_size)
213 unsigned long i, bytes, req_nslabs = io_tlb_nslabs;
216 if (!io_tlb_nslabs) {
217 io_tlb_nslabs = (default_size >> IO_TLB_SHIFT);
218 io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
222 * Get IO TLB memory from the low pages
224 order = get_order(io_tlb_nslabs << IO_TLB_SHIFT);
225 io_tlb_nslabs = SLABS_PER_PAGE << order;
226 bytes = io_tlb_nslabs << IO_TLB_SHIFT;
228 while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) {
229 io_tlb_start = swiotlb_alloc(order, io_tlb_nslabs);
238 if (order != get_order(bytes)) {
239 printk(KERN_WARNING "Warning: only able to allocate %ld MB "
240 "for software IO TLB\n", (PAGE_SIZE << order) >> 20);
241 io_tlb_nslabs = SLABS_PER_PAGE << order;
242 bytes = io_tlb_nslabs << IO_TLB_SHIFT;
244 io_tlb_end = io_tlb_start + bytes;
245 memset(io_tlb_start, 0, bytes);
248 * Allocate and initialize the free list array. This array is used
249 * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE
250 * between io_tlb_start and io_tlb_end.
252 io_tlb_list = (unsigned int *)__get_free_pages(GFP_KERNEL,
253 get_order(io_tlb_nslabs * sizeof(int)));
257 for (i = 0; i < io_tlb_nslabs; i++)
258 io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE);
261 io_tlb_orig_addr = (unsigned char **)__get_free_pages(GFP_KERNEL,
262 get_order(io_tlb_nslabs * sizeof(char *)));
263 if (!io_tlb_orig_addr)
266 memset(io_tlb_orig_addr, 0, io_tlb_nslabs * sizeof(char *));
269 * Get the overflow emergency buffer
271 io_tlb_overflow_buffer = (void *)__get_free_pages(GFP_DMA,
272 get_order(io_tlb_overflow));
273 if (!io_tlb_overflow_buffer)
276 printk(KERN_INFO "Placing %luMB software IO TLB between 0x%lx - "
277 "0x%lx\n", bytes >> 20,
278 swiotlb_virt_to_bus(io_tlb_start), swiotlb_virt_to_bus(io_tlb_end));
283 free_pages((unsigned long)io_tlb_orig_addr, get_order(io_tlb_nslabs *
285 io_tlb_orig_addr = NULL;
287 free_pages((unsigned long)io_tlb_list, get_order(io_tlb_nslabs *
292 free_pages((unsigned long)io_tlb_start, order);
295 io_tlb_nslabs = req_nslabs;
300 address_needs_mapping(struct device *hwdev, dma_addr_t addr, size_t size)
302 return !is_buffer_dma_capable(dma_get_mask(hwdev), addr, size);
305 static inline int range_needs_mapping(void *ptr, size_t size)
307 return swiotlb_force || swiotlb_arch_range_needs_mapping(ptr, size);
310 static int is_swiotlb_buffer(char *addr)
312 return addr >= io_tlb_start && addr < io_tlb_end;
316 * Allocates bounce buffer and returns its kernel virtual address.
319 map_single(struct device *hwdev, char *buffer, size_t size, int dir)
323 unsigned int nslots, stride, index, wrap;
325 unsigned long start_dma_addr;
327 unsigned long offset_slots;
328 unsigned long max_slots;
330 mask = dma_get_seg_boundary(hwdev);
331 start_dma_addr = swiotlb_virt_to_bus(io_tlb_start) & mask;
333 offset_slots = ALIGN(start_dma_addr, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT;
336 * Carefully handle integer overflow which can occur when mask == ~0UL.
339 ? ALIGN(mask + 1, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT
340 : 1UL << (BITS_PER_LONG - IO_TLB_SHIFT);
343 * For mappings greater than a page, we limit the stride (and
344 * hence alignment) to a page size.
346 nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT;
347 if (size > PAGE_SIZE)
348 stride = (1 << (PAGE_SHIFT - IO_TLB_SHIFT));
355 * Find suitable number of IO TLB entries size that will fit this
356 * request and allocate a buffer from that IO TLB pool.
358 spin_lock_irqsave(&io_tlb_lock, flags);
359 index = ALIGN(io_tlb_index, stride);
360 if (index >= io_tlb_nslabs)
365 while (iommu_is_span_boundary(index, nslots, offset_slots,
368 if (index >= io_tlb_nslabs)
375 * If we find a slot that indicates we have 'nslots' number of
376 * contiguous buffers, we allocate the buffers from that slot
377 * and mark the entries as '0' indicating unavailable.
379 if (io_tlb_list[index] >= nslots) {
382 for (i = index; i < (int) (index + nslots); i++)
384 for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE - 1) && io_tlb_list[i]; i--)
385 io_tlb_list[i] = ++count;
386 dma_addr = io_tlb_start + (index << IO_TLB_SHIFT);
389 * Update the indices to avoid searching in the next
392 io_tlb_index = ((index + nslots) < io_tlb_nslabs
393 ? (index + nslots) : 0);
398 if (index >= io_tlb_nslabs)
400 } while (index != wrap);
403 spin_unlock_irqrestore(&io_tlb_lock, flags);
406 spin_unlock_irqrestore(&io_tlb_lock, flags);
409 * Save away the mapping from the original address to the DMA address.
410 * This is needed when we sync the memory. Then we sync the buffer if
413 for (i = 0; i < nslots; i++)
414 io_tlb_orig_addr[index+i] = buffer + (i << IO_TLB_SHIFT);
415 if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL)
416 memcpy(dma_addr, buffer, size);
422 * dma_addr is the kernel virtual address of the bounce buffer to unmap.
425 unmap_single(struct device *hwdev, char *dma_addr, size_t size, int dir)
428 int i, count, nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT;
429 int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT;
430 char *buffer = io_tlb_orig_addr[index];
433 * First, sync the memory before unmapping the entry
435 if (buffer && ((dir == DMA_FROM_DEVICE) || (dir == DMA_BIDIRECTIONAL)))
437 * bounce... copy the data back into the original buffer * and
438 * delete the bounce buffer.
440 memcpy(buffer, dma_addr, size);
443 * Return the buffer to the free list by setting the corresponding
444 * entries to indicate the number of contigous entries available.
445 * While returning the entries to the free list, we merge the entries
446 * with slots below and above the pool being returned.
448 spin_lock_irqsave(&io_tlb_lock, flags);
450 count = ((index + nslots) < ALIGN(index + 1, IO_TLB_SEGSIZE) ?
451 io_tlb_list[index + nslots] : 0);
453 * Step 1: return the slots to the free list, merging the
454 * slots with superceeding slots
456 for (i = index + nslots - 1; i >= index; i--)
457 io_tlb_list[i] = ++count;
459 * Step 2: merge the returned slots with the preceding slots,
460 * if available (non zero)
462 for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE -1) && io_tlb_list[i]; i--)
463 io_tlb_list[i] = ++count;
465 spin_unlock_irqrestore(&io_tlb_lock, flags);
469 sync_single(struct device *hwdev, char *dma_addr, size_t size,
472 int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT;
473 char *buffer = io_tlb_orig_addr[index];
475 buffer += ((unsigned long)dma_addr & ((1 << IO_TLB_SHIFT) - 1));
479 if (likely(dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL))
480 memcpy(buffer, dma_addr, size);
482 BUG_ON(dir != DMA_TO_DEVICE);
484 case SYNC_FOR_DEVICE:
485 if (likely(dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL))
486 memcpy(dma_addr, buffer, size);
488 BUG_ON(dir != DMA_FROM_DEVICE);
496 swiotlb_alloc_coherent(struct device *hwdev, size_t size,
497 dma_addr_t *dma_handle, gfp_t flags)
501 int order = get_order(size);
502 u64 dma_mask = DMA_32BIT_MASK;
504 if (hwdev && hwdev->coherent_dma_mask)
505 dma_mask = hwdev->coherent_dma_mask;
507 ret = (void *)__get_free_pages(flags, order);
508 if (ret && !is_buffer_dma_capable(dma_mask, swiotlb_virt_to_bus(ret), size)) {
510 * The allocated memory isn't reachable by the device.
511 * Fall back on swiotlb_map_single().
513 free_pages((unsigned long) ret, order);
518 * We are either out of memory or the device can't DMA
519 * to GFP_DMA memory; fall back on
520 * swiotlb_map_single(), which will grab memory from
521 * the lowest available address range.
523 ret = map_single(hwdev, NULL, size, DMA_FROM_DEVICE);
528 memset(ret, 0, size);
529 dev_addr = swiotlb_virt_to_bus(ret);
531 /* Confirm address can be DMA'd by device */
532 if (!is_buffer_dma_capable(dma_mask, dev_addr, size)) {
533 printk("hwdev DMA mask = 0x%016Lx, dev_addr = 0x%016Lx\n",
534 (unsigned long long)dma_mask,
535 (unsigned long long)dev_addr);
537 /* DMA_TO_DEVICE to avoid memcpy in unmap_single */
538 unmap_single(hwdev, ret, size, DMA_TO_DEVICE);
541 *dma_handle = dev_addr;
546 swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr,
547 dma_addr_t dma_handle)
549 WARN_ON(irqs_disabled());
550 if (!is_swiotlb_buffer(vaddr))
551 free_pages((unsigned long) vaddr, get_order(size));
553 /* DMA_TO_DEVICE to avoid memcpy in unmap_single */
554 unmap_single(hwdev, vaddr, size, DMA_TO_DEVICE);
558 swiotlb_full(struct device *dev, size_t size, int dir, int do_panic)
561 * Ran out of IOMMU space for this operation. This is very bad.
562 * Unfortunately the drivers cannot handle this operation properly.
563 * unless they check for dma_mapping_error (most don't)
564 * When the mapping is small enough return a static buffer to limit
565 * the damage, or panic when the transfer is too big.
567 printk(KERN_ERR "DMA: Out of SW-IOMMU space for %zu bytes at "
568 "device %s\n", size, dev ? dev->bus_id : "?");
570 if (size > io_tlb_overflow && do_panic) {
571 if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
572 panic("DMA: Memory would be corrupted\n");
573 if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL)
574 panic("DMA: Random memory would be DMAed\n");
579 * Map a single buffer of the indicated size for DMA in streaming mode. The
580 * physical address to use is returned.
582 * Once the device is given the dma address, the device owns this memory until
583 * either swiotlb_unmap_single or swiotlb_dma_sync_single is performed.
586 swiotlb_map_single_attrs(struct device *hwdev, void *ptr, size_t size,
587 int dir, struct dma_attrs *attrs)
589 dma_addr_t dev_addr = swiotlb_virt_to_bus(ptr);
592 BUG_ON(dir == DMA_NONE);
594 * If the pointer passed in happens to be in the device's DMA window,
595 * we can safely return the device addr and not worry about bounce
598 if (!address_needs_mapping(hwdev, dev_addr, size) &&
599 !range_needs_mapping(ptr, size))
603 * Oh well, have to allocate and map a bounce buffer.
605 map = map_single(hwdev, ptr, size, dir);
607 swiotlb_full(hwdev, size, dir, 1);
608 map = io_tlb_overflow_buffer;
611 dev_addr = swiotlb_virt_to_bus(map);
614 * Ensure that the address returned is DMA'ble
616 if (address_needs_mapping(hwdev, dev_addr, size))
617 panic("map_single: bounce buffer is not DMA'ble");
621 EXPORT_SYMBOL(swiotlb_map_single_attrs);
624 swiotlb_map_single(struct device *hwdev, void *ptr, size_t size, int dir)
626 return swiotlb_map_single_attrs(hwdev, ptr, size, dir, NULL);
630 * Unmap a single streaming mode DMA translation. The dma_addr and size must
631 * match what was provided for in a previous swiotlb_map_single call. All
632 * other usages are undefined.
634 * After this call, reads by the cpu to the buffer are guaranteed to see
635 * whatever the device wrote there.
638 swiotlb_unmap_single_attrs(struct device *hwdev, dma_addr_t dev_addr,
639 size_t size, int dir, struct dma_attrs *attrs)
641 char *dma_addr = swiotlb_bus_to_virt(dev_addr);
643 BUG_ON(dir == DMA_NONE);
644 if (is_swiotlb_buffer(dma_addr))
645 unmap_single(hwdev, dma_addr, size, dir);
646 else if (dir == DMA_FROM_DEVICE)
647 dma_mark_clean(dma_addr, size);
649 EXPORT_SYMBOL(swiotlb_unmap_single_attrs);
652 swiotlb_unmap_single(struct device *hwdev, dma_addr_t dev_addr, size_t size,
655 return swiotlb_unmap_single_attrs(hwdev, dev_addr, size, dir, NULL);
658 * Make physical memory consistent for a single streaming mode DMA translation
661 * If you perform a swiotlb_map_single() but wish to interrogate the buffer
662 * using the cpu, yet do not wish to teardown the dma mapping, you must
663 * call this function before doing so. At the next point you give the dma
664 * address back to the card, you must first perform a
665 * swiotlb_dma_sync_for_device, and then the device again owns the buffer
668 swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr,
669 size_t size, int dir, int target)
671 char *dma_addr = swiotlb_bus_to_virt(dev_addr);
673 BUG_ON(dir == DMA_NONE);
674 if (is_swiotlb_buffer(dma_addr))
675 sync_single(hwdev, dma_addr, size, dir, target);
676 else if (dir == DMA_FROM_DEVICE)
677 dma_mark_clean(dma_addr, size);
681 swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr,
682 size_t size, int dir)
684 swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_CPU);
688 swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr,
689 size_t size, int dir)
691 swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_DEVICE);
695 * Same as above, but for a sub-range of the mapping.
698 swiotlb_sync_single_range(struct device *hwdev, dma_addr_t dev_addr,
699 unsigned long offset, size_t size,
702 char *dma_addr = swiotlb_bus_to_virt(dev_addr) + offset;
704 BUG_ON(dir == DMA_NONE);
705 if (is_swiotlb_buffer(dma_addr))
706 sync_single(hwdev, dma_addr, size, dir, target);
707 else if (dir == DMA_FROM_DEVICE)
708 dma_mark_clean(dma_addr, size);
712 swiotlb_sync_single_range_for_cpu(struct device *hwdev, dma_addr_t dev_addr,
713 unsigned long offset, size_t size, int dir)
715 swiotlb_sync_single_range(hwdev, dev_addr, offset, size, dir,
720 swiotlb_sync_single_range_for_device(struct device *hwdev, dma_addr_t dev_addr,
721 unsigned long offset, size_t size, int dir)
723 swiotlb_sync_single_range(hwdev, dev_addr, offset, size, dir,
727 void swiotlb_unmap_sg_attrs(struct device *, struct scatterlist *, int, int,
730 * Map a set of buffers described by scatterlist in streaming mode for DMA.
731 * This is the scatter-gather version of the above swiotlb_map_single
732 * interface. Here the scatter gather list elements are each tagged with the
733 * appropriate dma address and length. They are obtained via
734 * sg_dma_{address,length}(SG).
736 * NOTE: An implementation may be able to use a smaller number of
737 * DMA address/length pairs than there are SG table elements.
738 * (for example via virtual mapping capabilities)
739 * The routine returns the number of addr/length pairs actually
740 * used, at most nents.
742 * Device ownership issues as mentioned above for swiotlb_map_single are the
746 swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems,
747 int dir, struct dma_attrs *attrs)
749 struct scatterlist *sg;
754 BUG_ON(dir == DMA_NONE);
756 for_each_sg(sgl, sg, nelems, i) {
757 addr = SG_ENT_VIRT_ADDRESS(sg);
758 dev_addr = swiotlb_virt_to_bus(addr);
759 if (range_needs_mapping(sg_virt(sg), sg->length) ||
760 address_needs_mapping(hwdev, dev_addr, sg->length)) {
761 void *map = map_single(hwdev, addr, sg->length, dir);
763 /* Don't panic here, we expect map_sg users
764 to do proper error handling. */
765 swiotlb_full(hwdev, sg->length, dir, 0);
766 swiotlb_unmap_sg_attrs(hwdev, sgl, i, dir,
768 sgl[0].dma_length = 0;
771 sg->dma_address = swiotlb_virt_to_bus(map);
773 sg->dma_address = dev_addr;
774 sg->dma_length = sg->length;
778 EXPORT_SYMBOL(swiotlb_map_sg_attrs);
781 swiotlb_map_sg(struct device *hwdev, struct scatterlist *sgl, int nelems,
784 return swiotlb_map_sg_attrs(hwdev, sgl, nelems, dir, NULL);
788 * Unmap a set of streaming mode DMA translations. Again, cpu read rules
789 * concerning calls here are the same as for swiotlb_unmap_single() above.
792 swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl,
793 int nelems, int dir, struct dma_attrs *attrs)
795 struct scatterlist *sg;
798 BUG_ON(dir == DMA_NONE);
800 for_each_sg(sgl, sg, nelems, i) {
801 if (sg->dma_address != SG_ENT_PHYS_ADDRESS(sg))
802 unmap_single(hwdev, swiotlb_bus_to_virt(sg->dma_address),
803 sg->dma_length, dir);
804 else if (dir == DMA_FROM_DEVICE)
805 dma_mark_clean(SG_ENT_VIRT_ADDRESS(sg), sg->dma_length);
808 EXPORT_SYMBOL(swiotlb_unmap_sg_attrs);
811 swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sgl, int nelems,
814 return swiotlb_unmap_sg_attrs(hwdev, sgl, nelems, dir, NULL);
818 * Make physical memory consistent for a set of streaming mode DMA translations
821 * The same as swiotlb_sync_single_* but for a scatter-gather list, same rules
825 swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl,
826 int nelems, int dir, int target)
828 struct scatterlist *sg;
831 BUG_ON(dir == DMA_NONE);
833 for_each_sg(sgl, sg, nelems, i) {
834 if (sg->dma_address != SG_ENT_PHYS_ADDRESS(sg))
835 sync_single(hwdev, swiotlb_bus_to_virt(sg->dma_address),
836 sg->dma_length, dir, target);
837 else if (dir == DMA_FROM_DEVICE)
838 dma_mark_clean(SG_ENT_VIRT_ADDRESS(sg), sg->dma_length);
843 swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg,
846 swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_CPU);
850 swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg,
853 swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_DEVICE);
857 swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr)
859 return (dma_addr == swiotlb_virt_to_bus(io_tlb_overflow_buffer));
863 * Return whether the given device DMA address mask can be supported
864 * properly. For example, if your device can only drive the low 24-bits
865 * during bus mastering, then you would pass 0x00ffffff as the mask to
869 swiotlb_dma_supported(struct device *hwdev, u64 mask)
871 return swiotlb_virt_to_bus(io_tlb_end - 1) <= mask;
874 EXPORT_SYMBOL(swiotlb_map_single);
875 EXPORT_SYMBOL(swiotlb_unmap_single);
876 EXPORT_SYMBOL(swiotlb_map_sg);
877 EXPORT_SYMBOL(swiotlb_unmap_sg);
878 EXPORT_SYMBOL(swiotlb_sync_single_for_cpu);
879 EXPORT_SYMBOL(swiotlb_sync_single_for_device);
880 EXPORT_SYMBOL_GPL(swiotlb_sync_single_range_for_cpu);
881 EXPORT_SYMBOL_GPL(swiotlb_sync_single_range_for_device);
882 EXPORT_SYMBOL(swiotlb_sync_sg_for_cpu);
883 EXPORT_SYMBOL(swiotlb_sync_sg_for_device);
884 EXPORT_SYMBOL(swiotlb_dma_mapping_error);
885 EXPORT_SYMBOL(swiotlb_alloc_coherent);
886 EXPORT_SYMBOL(swiotlb_free_coherent);
887 EXPORT_SYMBOL(swiotlb_dma_supported);