1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * copy offload engine support
4 *
5 * Copyright © 2006, Intel Corporation.
6 *
7 * Dan Williams <dan.j.williams@intel.com>
8 *
9 * with architecture considerations by:
10 * Neil Brown <neilb@suse.de>
11 * Jeff Garzik <jeff@garzik.org>
12 */
13 #include <linux/kernel.h>
14 #include <linux/highmem.h>
15 #include <linux/module.h>
16 #include <linux/mm.h>
17 #include <linux/dma-mapping.h>
18 #include <linux/async_tx.h>
19
20 /**
21 * async_memcpy - attempt to copy memory with a dma engine.
22 * @dest: destination page
23 * @src: src page
24 * @dest_offset: offset into 'dest' to start transaction
25 * @src_offset: offset into 'src' to start transaction
26 * @len: length in bytes
27 * @submit: submission / completion modifiers
28 *
29 * honored flags: ASYNC_TX_ACK
30 */
31 struct dma_async_tx_descriptor *
async_memcpy(struct page * dest,struct page * src,unsigned int dest_offset,unsigned int src_offset,size_t len,struct async_submit_ctl * submit)32 async_memcpy(struct page *dest, struct page *src, unsigned int dest_offset,
33 unsigned int src_offset, size_t len,
34 struct async_submit_ctl *submit)
35 {
36 struct dma_chan *chan = async_tx_find_channel(submit, DMA_MEMCPY,
37 &dest, 1, &src, 1, len);
38 struct dma_device *device = chan ? chan->device : NULL;
39 struct dma_async_tx_descriptor *tx = NULL;
40 struct dmaengine_unmap_data *unmap = NULL;
41
42 if (device)
43 unmap = dmaengine_get_unmap_data(device->dev, 2, GFP_NOWAIT);
44
45 if (unmap && is_dma_copy_aligned(device, src_offset, dest_offset, len)) {
46 unsigned long dma_prep_flags = 0;
47
48 if (submit->cb_fn)
49 dma_prep_flags |= DMA_PREP_INTERRUPT;
50 if (submit->flags & ASYNC_TX_FENCE)
51 dma_prep_flags |= DMA_PREP_FENCE;
52
53 unmap->to_cnt = 1;
54 unmap->addr[0] = dma_map_page(device->dev, src, src_offset, len,
55 DMA_TO_DEVICE);
56 unmap->from_cnt = 1;
57 unmap->addr[1] = dma_map_page(device->dev, dest, dest_offset, len,
58 DMA_FROM_DEVICE);
59 unmap->len = len;
60
61 tx = device->device_prep_dma_memcpy(chan, unmap->addr[1],
62 unmap->addr[0], len,
63 dma_prep_flags);
64 }
65
66 if (tx) {
67 pr_debug("%s: (async) len: %zu\n", __func__, len);
68
69 dma_set_unmap(tx, unmap);
70 async_tx_submit(chan, tx, submit);
71 } else {
72 void *dest_buf, *src_buf;
73 pr_debug("%s: (sync) len: %zu\n", __func__, len);
74
75 /* wait for any prerequisite operations */
76 async_tx_quiesce(&submit->depend_tx);
77
78 dest_buf = kmap_atomic(dest) + dest_offset;
79 src_buf = kmap_atomic(src) + src_offset;
80
81 memcpy(dest_buf, src_buf, len);
82
83 kunmap_atomic(src_buf);
84 kunmap_atomic(dest_buf);
85
86 async_tx_sync_epilog(submit);
87 }
88
89 dmaengine_unmap_put(unmap);
90
91 return tx;
92 }
93 EXPORT_SYMBOL_GPL(async_memcpy);
94
95 MODULE_AUTHOR("Intel Corporation");
96 MODULE_DESCRIPTION("asynchronous memcpy api");
97 MODULE_LICENSE("GPL");
98