1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License.  See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Copyright (C) 2000  Ani Joshi <ajoshi@unixbox.com>
7 * Copyright (C) 2000, 2001  Ralf Baechle <ralf@gnu.org>
8 * Copyright (C) 2005 Ilya A. Volynets-Evenbakh <ilya@total-knowledge.com>
9 * swiped from i386, and cloned for MIPS by Geert, polished by Ralf.
10 * IP32 changes by Ilya.
11 * Copyright (C) 2010 Cavium Networks, Inc.
12 */
13#include <linux/dma-mapping.h>
14#include <linux/scatterlist.h>
15#include <linux/bootmem.h>
16#include <linux/export.h>
17#include <linux/swiotlb.h>
18#include <linux/types.h>
19#include <linux/init.h>
20#include <linux/mm.h>
21
22#include <asm/bootinfo.h>
23
24#include <asm/octeon/octeon.h>
25
26#ifdef CONFIG_PCI
27#include <asm/octeon/pci-octeon.h>
28#include <asm/octeon/cvmx-npi-defs.h>
29#include <asm/octeon/cvmx-pci-defs.h>
30
31static dma_addr_t octeon_hole_phys_to_dma(phys_addr_t paddr)
32{
33	if (paddr >= CVMX_PCIE_BAR1_PHYS_BASE && paddr < (CVMX_PCIE_BAR1_PHYS_BASE + CVMX_PCIE_BAR1_PHYS_SIZE))
34		return paddr - CVMX_PCIE_BAR1_PHYS_BASE + CVMX_PCIE_BAR1_RC_BASE;
35	else
36		return paddr;
37}
38
39static phys_addr_t octeon_hole_dma_to_phys(dma_addr_t daddr)
40{
41	if (daddr >= CVMX_PCIE_BAR1_RC_BASE)
42		return daddr + CVMX_PCIE_BAR1_PHYS_BASE - CVMX_PCIE_BAR1_RC_BASE;
43	else
44		return daddr;
45}
46
47static dma_addr_t octeon_gen1_phys_to_dma(struct device *dev, phys_addr_t paddr)
48{
49	if (paddr >= 0x410000000ull && paddr < 0x420000000ull)
50		paddr -= 0x400000000ull;
51	return octeon_hole_phys_to_dma(paddr);
52}
53
54static phys_addr_t octeon_gen1_dma_to_phys(struct device *dev, dma_addr_t daddr)
55{
56	daddr = octeon_hole_dma_to_phys(daddr);
57
58	if (daddr >= 0x10000000ull && daddr < 0x20000000ull)
59		daddr += 0x400000000ull;
60
61	return daddr;
62}
63
64static dma_addr_t octeon_gen2_phys_to_dma(struct device *dev, phys_addr_t paddr)
65{
66	return octeon_hole_phys_to_dma(paddr);
67}
68
69static phys_addr_t octeon_gen2_dma_to_phys(struct device *dev, dma_addr_t daddr)
70{
71	return octeon_hole_dma_to_phys(daddr);
72}
73
74static dma_addr_t octeon_big_phys_to_dma(struct device *dev, phys_addr_t paddr)
75{
76	if (paddr >= 0x410000000ull && paddr < 0x420000000ull)
77		paddr -= 0x400000000ull;
78
79	/* Anything in the BAR1 hole or above goes via BAR2 */
80	if (paddr >= 0xf0000000ull)
81		paddr = OCTEON_BAR2_PCI_ADDRESS + paddr;
82
83	return paddr;
84}
85
86static phys_addr_t octeon_big_dma_to_phys(struct device *dev, dma_addr_t daddr)
87{
88	if (daddr >= OCTEON_BAR2_PCI_ADDRESS)
89		daddr -= OCTEON_BAR2_PCI_ADDRESS;
90
91	if (daddr >= 0x10000000ull && daddr < 0x20000000ull)
92		daddr += 0x400000000ull;
93	return daddr;
94}
95
96static dma_addr_t octeon_small_phys_to_dma(struct device *dev,
97					   phys_addr_t paddr)
98{
99	if (paddr >= 0x410000000ull && paddr < 0x420000000ull)
100		paddr -= 0x400000000ull;
101
102	/* Anything not in the BAR1 range goes via BAR2 */
103	if (paddr >= octeon_bar1_pci_phys && paddr < octeon_bar1_pci_phys + 0x8000000ull)
104		paddr = paddr - octeon_bar1_pci_phys;
105	else
106		paddr = OCTEON_BAR2_PCI_ADDRESS + paddr;
107
108	return paddr;
109}
110
111static phys_addr_t octeon_small_dma_to_phys(struct device *dev,
112					    dma_addr_t daddr)
113{
114	if (daddr >= OCTEON_BAR2_PCI_ADDRESS)
115		daddr -= OCTEON_BAR2_PCI_ADDRESS;
116	else
117		daddr += octeon_bar1_pci_phys;
118
119	if (daddr >= 0x10000000ull && daddr < 0x20000000ull)
120		daddr += 0x400000000ull;
121	return daddr;
122}
123
124#endif /* CONFIG_PCI */
125
126static dma_addr_t octeon_dma_map_page(struct device *dev, struct page *page,
127	unsigned long offset, size_t size, enum dma_data_direction direction,
128	struct dma_attrs *attrs)
129{
130	dma_addr_t daddr = swiotlb_map_page(dev, page, offset, size,
131					    direction, attrs);
132	mb();
133
134	return daddr;
135}
136
137static int octeon_dma_map_sg(struct device *dev, struct scatterlist *sg,
138	int nents, enum dma_data_direction direction, struct dma_attrs *attrs)
139{
140	int r = swiotlb_map_sg_attrs(dev, sg, nents, direction, attrs);
141	mb();
142	return r;
143}
144
145static void octeon_dma_sync_single_for_device(struct device *dev,
146	dma_addr_t dma_handle, size_t size, enum dma_data_direction direction)
147{
148	swiotlb_sync_single_for_device(dev, dma_handle, size, direction);
149	mb();
150}
151
152static void octeon_dma_sync_sg_for_device(struct device *dev,
153	struct scatterlist *sg, int nelems, enum dma_data_direction direction)
154{
155	swiotlb_sync_sg_for_device(dev, sg, nelems, direction);
156	mb();
157}
158
159static void *octeon_dma_alloc_coherent(struct device *dev, size_t size,
160	dma_addr_t *dma_handle, gfp_t gfp, struct dma_attrs *attrs)
161{
162	void *ret;
163
164	if (dma_alloc_from_coherent(dev, size, dma_handle, &ret))
165		return ret;
166
167	/* ignore region specifiers */
168	gfp &= ~(__GFP_DMA | __GFP_DMA32 | __GFP_HIGHMEM);
169
170#ifdef CONFIG_ZONE_DMA
171	if (dev == NULL)
172		gfp |= __GFP_DMA;
173	else if (dev->coherent_dma_mask <= DMA_BIT_MASK(24))
174		gfp |= __GFP_DMA;
175	else
176#endif
177#ifdef CONFIG_ZONE_DMA32
178	     if (dev->coherent_dma_mask <= DMA_BIT_MASK(32))
179		gfp |= __GFP_DMA32;
180	else
181#endif
182		;
183
184	/* Don't invoke OOM killer */
185	gfp |= __GFP_NORETRY;
186
187	ret = swiotlb_alloc_coherent(dev, size, dma_handle, gfp);
188
189	mb();
190
191	return ret;
192}
193
194static void octeon_dma_free_coherent(struct device *dev, size_t size,
195	void *vaddr, dma_addr_t dma_handle, struct dma_attrs *attrs)
196{
197	int order = get_order(size);
198
199	if (dma_release_from_coherent(dev, order, vaddr))
200		return;
201
202	swiotlb_free_coherent(dev, size, vaddr, dma_handle);
203}
204
205static dma_addr_t octeon_unity_phys_to_dma(struct device *dev, phys_addr_t paddr)
206{
207	return paddr;
208}
209
210static phys_addr_t octeon_unity_dma_to_phys(struct device *dev, dma_addr_t daddr)
211{
212	return daddr;
213}
214
215struct octeon_dma_map_ops {
216	struct dma_map_ops dma_map_ops;
217	dma_addr_t (*phys_to_dma)(struct device *dev, phys_addr_t paddr);
218	phys_addr_t (*dma_to_phys)(struct device *dev, dma_addr_t daddr);
219};
220
221dma_addr_t phys_to_dma(struct device *dev, phys_addr_t paddr)
222{
223	struct octeon_dma_map_ops *ops = container_of(get_dma_ops(dev),
224						      struct octeon_dma_map_ops,
225						      dma_map_ops);
226
227	return ops->phys_to_dma(dev, paddr);
228}
229EXPORT_SYMBOL(phys_to_dma);
230
231phys_addr_t dma_to_phys(struct device *dev, dma_addr_t daddr)
232{
233	struct octeon_dma_map_ops *ops = container_of(get_dma_ops(dev),
234						      struct octeon_dma_map_ops,
235						      dma_map_ops);
236
237	return ops->dma_to_phys(dev, daddr);
238}
239EXPORT_SYMBOL(dma_to_phys);
240
241static struct octeon_dma_map_ops octeon_linear_dma_map_ops = {
242	.dma_map_ops = {
243		.alloc = octeon_dma_alloc_coherent,
244		.free = octeon_dma_free_coherent,
245		.map_page = octeon_dma_map_page,
246		.unmap_page = swiotlb_unmap_page,
247		.map_sg = octeon_dma_map_sg,
248		.unmap_sg = swiotlb_unmap_sg_attrs,
249		.sync_single_for_cpu = swiotlb_sync_single_for_cpu,
250		.sync_single_for_device = octeon_dma_sync_single_for_device,
251		.sync_sg_for_cpu = swiotlb_sync_sg_for_cpu,
252		.sync_sg_for_device = octeon_dma_sync_sg_for_device,
253		.mapping_error = swiotlb_dma_mapping_error,
254		.dma_supported = swiotlb_dma_supported
255	},
256	.phys_to_dma = octeon_unity_phys_to_dma,
257	.dma_to_phys = octeon_unity_dma_to_phys
258};
259
260char *octeon_swiotlb;
261
262void __init plat_swiotlb_setup(void)
263{
264	int i;
265	phys_t max_addr;
266	phys_t addr_size;
267	size_t swiotlbsize;
268	unsigned long swiotlb_nslabs;
269
270	max_addr = 0;
271	addr_size = 0;
272
273	for (i = 0 ; i < boot_mem_map.nr_map; i++) {
274		struct boot_mem_map_entry *e = &boot_mem_map.map[i];
275		if (e->type != BOOT_MEM_RAM && e->type != BOOT_MEM_INIT_RAM)
276			continue;
277
278		/* These addresses map low for PCI. */
279		if (e->addr > 0x410000000ull && !OCTEON_IS_MODEL(OCTEON_CN6XXX))
280			continue;
281
282		addr_size += e->size;
283
284		if (max_addr < e->addr + e->size)
285			max_addr = e->addr + e->size;
286
287	}
288
289	swiotlbsize = PAGE_SIZE;
290
291#ifdef CONFIG_PCI
292	/*
293	 * For OCTEON_DMA_BAR_TYPE_SMALL, size the iotlb at 1/4 memory
294	 * size to a maximum of 64MB
295	 */
296	if (OCTEON_IS_MODEL(OCTEON_CN31XX)
297	    || OCTEON_IS_MODEL(OCTEON_CN38XX_PASS2)) {
298		swiotlbsize = addr_size / 4;
299		if (swiotlbsize > 64 * (1<<20))
300			swiotlbsize = 64 * (1<<20);
301	} else if (max_addr > 0xf0000000ul) {
302		/*
303		 * Otherwise only allocate a big iotlb if there is
304		 * memory past the BAR1 hole.
305		 */
306		swiotlbsize = 64 * (1<<20);
307	}
308#endif
309#ifdef CONFIG_USB_OCTEON_OHCI
310	/* OCTEON II ohci is only 32-bit. */
311	if (OCTEON_IS_MODEL(OCTEON_CN6XXX) && max_addr >= 0x100000000ul)
312		swiotlbsize = 64 * (1<<20);
313#endif
314	swiotlb_nslabs = swiotlbsize >> IO_TLB_SHIFT;
315	swiotlb_nslabs = ALIGN(swiotlb_nslabs, IO_TLB_SEGSIZE);
316	swiotlbsize = swiotlb_nslabs << IO_TLB_SHIFT;
317
318	octeon_swiotlb = alloc_bootmem_low_pages(swiotlbsize);
319
320	if (swiotlb_init_with_tbl(octeon_swiotlb, swiotlb_nslabs, 1) == -ENOMEM)
321		panic("Cannot allocate SWIOTLB buffer");
322
323	mips_dma_map_ops = &octeon_linear_dma_map_ops.dma_map_ops;
324}
325
326#ifdef CONFIG_PCI
327static struct octeon_dma_map_ops _octeon_pci_dma_map_ops = {
328	.dma_map_ops = {
329		.alloc = octeon_dma_alloc_coherent,
330		.free = octeon_dma_free_coherent,
331		.map_page = octeon_dma_map_page,
332		.unmap_page = swiotlb_unmap_page,
333		.map_sg = octeon_dma_map_sg,
334		.unmap_sg = swiotlb_unmap_sg_attrs,
335		.sync_single_for_cpu = swiotlb_sync_single_for_cpu,
336		.sync_single_for_device = octeon_dma_sync_single_for_device,
337		.sync_sg_for_cpu = swiotlb_sync_sg_for_cpu,
338		.sync_sg_for_device = octeon_dma_sync_sg_for_device,
339		.mapping_error = swiotlb_dma_mapping_error,
340		.dma_supported = swiotlb_dma_supported
341	},
342};
343
344struct dma_map_ops *octeon_pci_dma_map_ops;
345
346void __init octeon_pci_dma_init(void)
347{
348	switch (octeon_dma_bar_type) {
349	case OCTEON_DMA_BAR_TYPE_PCIE2:
350		_octeon_pci_dma_map_ops.phys_to_dma = octeon_gen2_phys_to_dma;
351		_octeon_pci_dma_map_ops.dma_to_phys = octeon_gen2_dma_to_phys;
352		break;
353	case OCTEON_DMA_BAR_TYPE_PCIE:
354		_octeon_pci_dma_map_ops.phys_to_dma = octeon_gen1_phys_to_dma;
355		_octeon_pci_dma_map_ops.dma_to_phys = octeon_gen1_dma_to_phys;
356		break;
357	case OCTEON_DMA_BAR_TYPE_BIG:
358		_octeon_pci_dma_map_ops.phys_to_dma = octeon_big_phys_to_dma;
359		_octeon_pci_dma_map_ops.dma_to_phys = octeon_big_dma_to_phys;
360		break;
361	case OCTEON_DMA_BAR_TYPE_SMALL:
362		_octeon_pci_dma_map_ops.phys_to_dma = octeon_small_phys_to_dma;
363		_octeon_pci_dma_map_ops.dma_to_phys = octeon_small_dma_to_phys;
364		break;
365	default:
366		BUG();
367	}
368	octeon_pci_dma_map_ops = &_octeon_pci_dma_map_ops.dma_map_ops;
369}
370#endif /* CONFIG_PCI */
371