xref: /openbmc/linux/drivers/dma/qcom/bam_dma.c (revision 2d972b6a)
1 /*
2  * Copyright (c) 2013-2014, The Linux Foundation. All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 and
6  * only version 2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  */
14 /*
15  * QCOM BAM DMA engine driver
16  *
17  * QCOM BAM DMA blocks are distributed amongst a number of the on-chip
18  * peripherals on the MSM 8x74.  The configuration of the channels are dependent
19  * on the way they are hard wired to that specific peripheral.  The peripheral
20  * device tree entries specify the configuration of each channel.
21  *
22  * The DMA controller requires the use of external memory for storage of the
23  * hardware descriptors for each channel.  The descriptor FIFO is accessed as a
24  * circular buffer and operations are managed according to the offset within the
25  * FIFO.  After pipe/channel reset, all of the pipe registers and internal state
26  * are back to defaults.
27  *
28  * During DMA operations, we write descriptors to the FIFO, being careful to
29  * handle wrapping and then write the last FIFO offset to that channel's
30  * P_EVNT_REG register to kick off the transaction.  The P_SW_OFSTS register
31  * indicates the current FIFO offset that is being processed, so there is some
32  * indication of where the hardware is currently working.
33  */
34 
35 #include <linux/kernel.h>
36 #include <linux/io.h>
37 #include <linux/init.h>
38 #include <linux/slab.h>
39 #include <linux/module.h>
40 #include <linux/interrupt.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/scatterlist.h>
43 #include <linux/device.h>
44 #include <linux/platform_device.h>
45 #include <linux/of.h>
46 #include <linux/of_address.h>
47 #include <linux/of_irq.h>
48 #include <linux/of_dma.h>
49 #include <linux/circ_buf.h>
50 #include <linux/clk.h>
51 #include <linux/dmaengine.h>
52 #include <linux/pm_runtime.h>
53 
54 #include "../dmaengine.h"
55 #include "../virt-dma.h"
56 
57 struct bam_desc_hw {
58 	__le32 addr;		/* Buffer physical address */
59 	__le16 size;		/* Buffer size in bytes */
60 	__le16 flags;
61 };
62 
63 #define BAM_DMA_AUTOSUSPEND_DELAY 100
64 
65 #define DESC_FLAG_INT BIT(15)
66 #define DESC_FLAG_EOT BIT(14)
67 #define DESC_FLAG_EOB BIT(13)
68 #define DESC_FLAG_NWD BIT(12)
69 #define DESC_FLAG_CMD BIT(11)
70 
71 struct bam_async_desc {
72 	struct virt_dma_desc vd;
73 
74 	u32 num_desc;
75 	u32 xfer_len;
76 
77 	/* transaction flags, EOT|EOB|NWD */
78 	u16 flags;
79 
80 	struct bam_desc_hw *curr_desc;
81 
82 	/* list node for the desc in the bam_chan list of descriptors */
83 	struct list_head desc_node;
84 	enum dma_transfer_direction dir;
85 	size_t length;
86 	struct bam_desc_hw desc[0];
87 };
88 
89 enum bam_reg {
90 	BAM_CTRL,
91 	BAM_REVISION,
92 	BAM_NUM_PIPES,
93 	BAM_DESC_CNT_TRSHLD,
94 	BAM_IRQ_SRCS,
95 	BAM_IRQ_SRCS_MSK,
96 	BAM_IRQ_SRCS_UNMASKED,
97 	BAM_IRQ_STTS,
98 	BAM_IRQ_CLR,
99 	BAM_IRQ_EN,
100 	BAM_CNFG_BITS,
101 	BAM_IRQ_SRCS_EE,
102 	BAM_IRQ_SRCS_MSK_EE,
103 	BAM_P_CTRL,
104 	BAM_P_RST,
105 	BAM_P_HALT,
106 	BAM_P_IRQ_STTS,
107 	BAM_P_IRQ_CLR,
108 	BAM_P_IRQ_EN,
109 	BAM_P_EVNT_DEST_ADDR,
110 	BAM_P_EVNT_REG,
111 	BAM_P_SW_OFSTS,
112 	BAM_P_DATA_FIFO_ADDR,
113 	BAM_P_DESC_FIFO_ADDR,
114 	BAM_P_EVNT_GEN_TRSHLD,
115 	BAM_P_FIFO_SIZES,
116 };
117 
118 struct reg_offset_data {
119 	u32 base_offset;
120 	unsigned int pipe_mult, evnt_mult, ee_mult;
121 };
122 
123 static const struct reg_offset_data bam_v1_3_reg_info[] = {
124 	[BAM_CTRL]		= { 0x0F80, 0x00, 0x00, 0x00 },
125 	[BAM_REVISION]		= { 0x0F84, 0x00, 0x00, 0x00 },
126 	[BAM_NUM_PIPES]		= { 0x0FBC, 0x00, 0x00, 0x00 },
127 	[BAM_DESC_CNT_TRSHLD]	= { 0x0F88, 0x00, 0x00, 0x00 },
128 	[BAM_IRQ_SRCS]		= { 0x0F8C, 0x00, 0x00, 0x00 },
129 	[BAM_IRQ_SRCS_MSK]	= { 0x0F90, 0x00, 0x00, 0x00 },
130 	[BAM_IRQ_SRCS_UNMASKED]	= { 0x0FB0, 0x00, 0x00, 0x00 },
131 	[BAM_IRQ_STTS]		= { 0x0F94, 0x00, 0x00, 0x00 },
132 	[BAM_IRQ_CLR]		= { 0x0F98, 0x00, 0x00, 0x00 },
133 	[BAM_IRQ_EN]		= { 0x0F9C, 0x00, 0x00, 0x00 },
134 	[BAM_CNFG_BITS]		= { 0x0FFC, 0x00, 0x00, 0x00 },
135 	[BAM_IRQ_SRCS_EE]	= { 0x1800, 0x00, 0x00, 0x80 },
136 	[BAM_IRQ_SRCS_MSK_EE]	= { 0x1804, 0x00, 0x00, 0x80 },
137 	[BAM_P_CTRL]		= { 0x0000, 0x80, 0x00, 0x00 },
138 	[BAM_P_RST]		= { 0x0004, 0x80, 0x00, 0x00 },
139 	[BAM_P_HALT]		= { 0x0008, 0x80, 0x00, 0x00 },
140 	[BAM_P_IRQ_STTS]	= { 0x0010, 0x80, 0x00, 0x00 },
141 	[BAM_P_IRQ_CLR]		= { 0x0014, 0x80, 0x00, 0x00 },
142 	[BAM_P_IRQ_EN]		= { 0x0018, 0x80, 0x00, 0x00 },
143 	[BAM_P_EVNT_DEST_ADDR]	= { 0x102C, 0x00, 0x40, 0x00 },
144 	[BAM_P_EVNT_REG]	= { 0x1018, 0x00, 0x40, 0x00 },
145 	[BAM_P_SW_OFSTS]	= { 0x1000, 0x00, 0x40, 0x00 },
146 	[BAM_P_DATA_FIFO_ADDR]	= { 0x1024, 0x00, 0x40, 0x00 },
147 	[BAM_P_DESC_FIFO_ADDR]	= { 0x101C, 0x00, 0x40, 0x00 },
148 	[BAM_P_EVNT_GEN_TRSHLD]	= { 0x1028, 0x00, 0x40, 0x00 },
149 	[BAM_P_FIFO_SIZES]	= { 0x1020, 0x00, 0x40, 0x00 },
150 };
151 
152 static const struct reg_offset_data bam_v1_4_reg_info[] = {
153 	[BAM_CTRL]		= { 0x0000, 0x00, 0x00, 0x00 },
154 	[BAM_REVISION]		= { 0x0004, 0x00, 0x00, 0x00 },
155 	[BAM_NUM_PIPES]		= { 0x003C, 0x00, 0x00, 0x00 },
156 	[BAM_DESC_CNT_TRSHLD]	= { 0x0008, 0x00, 0x00, 0x00 },
157 	[BAM_IRQ_SRCS]		= { 0x000C, 0x00, 0x00, 0x00 },
158 	[BAM_IRQ_SRCS_MSK]	= { 0x0010, 0x00, 0x00, 0x00 },
159 	[BAM_IRQ_SRCS_UNMASKED]	= { 0x0030, 0x00, 0x00, 0x00 },
160 	[BAM_IRQ_STTS]		= { 0x0014, 0x00, 0x00, 0x00 },
161 	[BAM_IRQ_CLR]		= { 0x0018, 0x00, 0x00, 0x00 },
162 	[BAM_IRQ_EN]		= { 0x001C, 0x00, 0x00, 0x00 },
163 	[BAM_CNFG_BITS]		= { 0x007C, 0x00, 0x00, 0x00 },
164 	[BAM_IRQ_SRCS_EE]	= { 0x0800, 0x00, 0x00, 0x80 },
165 	[BAM_IRQ_SRCS_MSK_EE]	= { 0x0804, 0x00, 0x00, 0x80 },
166 	[BAM_P_CTRL]		= { 0x1000, 0x1000, 0x00, 0x00 },
167 	[BAM_P_RST]		= { 0x1004, 0x1000, 0x00, 0x00 },
168 	[BAM_P_HALT]		= { 0x1008, 0x1000, 0x00, 0x00 },
169 	[BAM_P_IRQ_STTS]	= { 0x1010, 0x1000, 0x00, 0x00 },
170 	[BAM_P_IRQ_CLR]		= { 0x1014, 0x1000, 0x00, 0x00 },
171 	[BAM_P_IRQ_EN]		= { 0x1018, 0x1000, 0x00, 0x00 },
172 	[BAM_P_EVNT_DEST_ADDR]	= { 0x182C, 0x00, 0x1000, 0x00 },
173 	[BAM_P_EVNT_REG]	= { 0x1818, 0x00, 0x1000, 0x00 },
174 	[BAM_P_SW_OFSTS]	= { 0x1800, 0x00, 0x1000, 0x00 },
175 	[BAM_P_DATA_FIFO_ADDR]	= { 0x1824, 0x00, 0x1000, 0x00 },
176 	[BAM_P_DESC_FIFO_ADDR]	= { 0x181C, 0x00, 0x1000, 0x00 },
177 	[BAM_P_EVNT_GEN_TRSHLD]	= { 0x1828, 0x00, 0x1000, 0x00 },
178 	[BAM_P_FIFO_SIZES]	= { 0x1820, 0x00, 0x1000, 0x00 },
179 };
180 
181 static const struct reg_offset_data bam_v1_7_reg_info[] = {
182 	[BAM_CTRL]		= { 0x00000, 0x00, 0x00, 0x00 },
183 	[BAM_REVISION]		= { 0x01000, 0x00, 0x00, 0x00 },
184 	[BAM_NUM_PIPES]		= { 0x01008, 0x00, 0x00, 0x00 },
185 	[BAM_DESC_CNT_TRSHLD]	= { 0x00008, 0x00, 0x00, 0x00 },
186 	[BAM_IRQ_SRCS]		= { 0x03010, 0x00, 0x00, 0x00 },
187 	[BAM_IRQ_SRCS_MSK]	= { 0x03014, 0x00, 0x00, 0x00 },
188 	[BAM_IRQ_SRCS_UNMASKED]	= { 0x03018, 0x00, 0x00, 0x00 },
189 	[BAM_IRQ_STTS]		= { 0x00014, 0x00, 0x00, 0x00 },
190 	[BAM_IRQ_CLR]		= { 0x00018, 0x00, 0x00, 0x00 },
191 	[BAM_IRQ_EN]		= { 0x0001C, 0x00, 0x00, 0x00 },
192 	[BAM_CNFG_BITS]		= { 0x0007C, 0x00, 0x00, 0x00 },
193 	[BAM_IRQ_SRCS_EE]	= { 0x03000, 0x00, 0x00, 0x1000 },
194 	[BAM_IRQ_SRCS_MSK_EE]	= { 0x03004, 0x00, 0x00, 0x1000 },
195 	[BAM_P_CTRL]		= { 0x13000, 0x1000, 0x00, 0x00 },
196 	[BAM_P_RST]		= { 0x13004, 0x1000, 0x00, 0x00 },
197 	[BAM_P_HALT]		= { 0x13008, 0x1000, 0x00, 0x00 },
198 	[BAM_P_IRQ_STTS]	= { 0x13010, 0x1000, 0x00, 0x00 },
199 	[BAM_P_IRQ_CLR]		= { 0x13014, 0x1000, 0x00, 0x00 },
200 	[BAM_P_IRQ_EN]		= { 0x13018, 0x1000, 0x00, 0x00 },
201 	[BAM_P_EVNT_DEST_ADDR]	= { 0x1382C, 0x00, 0x1000, 0x00 },
202 	[BAM_P_EVNT_REG]	= { 0x13818, 0x00, 0x1000, 0x00 },
203 	[BAM_P_SW_OFSTS]	= { 0x13800, 0x00, 0x1000, 0x00 },
204 	[BAM_P_DATA_FIFO_ADDR]	= { 0x13824, 0x00, 0x1000, 0x00 },
205 	[BAM_P_DESC_FIFO_ADDR]	= { 0x1381C, 0x00, 0x1000, 0x00 },
206 	[BAM_P_EVNT_GEN_TRSHLD]	= { 0x13828, 0x00, 0x1000, 0x00 },
207 	[BAM_P_FIFO_SIZES]	= { 0x13820, 0x00, 0x1000, 0x00 },
208 };
209 
210 /* BAM CTRL */
211 #define BAM_SW_RST			BIT(0)
212 #define BAM_EN				BIT(1)
213 #define BAM_EN_ACCUM			BIT(4)
214 #define BAM_TESTBUS_SEL_SHIFT		5
215 #define BAM_TESTBUS_SEL_MASK		0x3F
216 #define BAM_DESC_CACHE_SEL_SHIFT	13
217 #define BAM_DESC_CACHE_SEL_MASK		0x3
218 #define BAM_CACHED_DESC_STORE		BIT(15)
219 #define IBC_DISABLE			BIT(16)
220 
221 /* BAM REVISION */
222 #define REVISION_SHIFT		0
223 #define REVISION_MASK		0xFF
224 #define NUM_EES_SHIFT		8
225 #define NUM_EES_MASK		0xF
226 #define CE_BUFFER_SIZE		BIT(13)
227 #define AXI_ACTIVE		BIT(14)
228 #define USE_VMIDMT		BIT(15)
229 #define SECURED			BIT(16)
230 #define BAM_HAS_NO_BYPASS	BIT(17)
231 #define HIGH_FREQUENCY_BAM	BIT(18)
232 #define INACTIV_TMRS_EXST	BIT(19)
233 #define NUM_INACTIV_TMRS	BIT(20)
234 #define DESC_CACHE_DEPTH_SHIFT	21
235 #define DESC_CACHE_DEPTH_1	(0 << DESC_CACHE_DEPTH_SHIFT)
236 #define DESC_CACHE_DEPTH_2	(1 << DESC_CACHE_DEPTH_SHIFT)
237 #define DESC_CACHE_DEPTH_3	(2 << DESC_CACHE_DEPTH_SHIFT)
238 #define DESC_CACHE_DEPTH_4	(3 << DESC_CACHE_DEPTH_SHIFT)
239 #define CMD_DESC_EN		BIT(23)
240 #define INACTIV_TMR_BASE_SHIFT	24
241 #define INACTIV_TMR_BASE_MASK	0xFF
242 
243 /* BAM NUM PIPES */
244 #define BAM_NUM_PIPES_SHIFT		0
245 #define BAM_NUM_PIPES_MASK		0xFF
246 #define PERIPH_NON_PIPE_GRP_SHIFT	16
247 #define PERIPH_NON_PIP_GRP_MASK		0xFF
248 #define BAM_NON_PIPE_GRP_SHIFT		24
249 #define BAM_NON_PIPE_GRP_MASK		0xFF
250 
251 /* BAM CNFG BITS */
252 #define BAM_PIPE_CNFG		BIT(2)
253 #define BAM_FULL_PIPE		BIT(11)
254 #define BAM_NO_EXT_P_RST	BIT(12)
255 #define BAM_IBC_DISABLE		BIT(13)
256 #define BAM_SB_CLK_REQ		BIT(14)
257 #define BAM_PSM_CSW_REQ		BIT(15)
258 #define BAM_PSM_P_RES		BIT(16)
259 #define BAM_AU_P_RES		BIT(17)
260 #define BAM_SI_P_RES		BIT(18)
261 #define BAM_WB_P_RES		BIT(19)
262 #define BAM_WB_BLK_CSW		BIT(20)
263 #define BAM_WB_CSW_ACK_IDL	BIT(21)
264 #define BAM_WB_RETR_SVPNT	BIT(22)
265 #define BAM_WB_DSC_AVL_P_RST	BIT(23)
266 #define BAM_REG_P_EN		BIT(24)
267 #define BAM_PSM_P_HD_DATA	BIT(25)
268 #define BAM_AU_ACCUMED		BIT(26)
269 #define BAM_CMD_ENABLE		BIT(27)
270 
271 #define BAM_CNFG_BITS_DEFAULT	(BAM_PIPE_CNFG |	\
272 				 BAM_NO_EXT_P_RST |	\
273 				 BAM_IBC_DISABLE |	\
274 				 BAM_SB_CLK_REQ |	\
275 				 BAM_PSM_CSW_REQ |	\
276 				 BAM_PSM_P_RES |	\
277 				 BAM_AU_P_RES |		\
278 				 BAM_SI_P_RES |		\
279 				 BAM_WB_P_RES |		\
280 				 BAM_WB_BLK_CSW |	\
281 				 BAM_WB_CSW_ACK_IDL |	\
282 				 BAM_WB_RETR_SVPNT |	\
283 				 BAM_WB_DSC_AVL_P_RST |	\
284 				 BAM_REG_P_EN |		\
285 				 BAM_PSM_P_HD_DATA |	\
286 				 BAM_AU_ACCUMED |	\
287 				 BAM_CMD_ENABLE)
288 
289 /* PIPE CTRL */
290 #define P_EN			BIT(1)
291 #define P_DIRECTION		BIT(3)
292 #define P_SYS_STRM		BIT(4)
293 #define P_SYS_MODE		BIT(5)
294 #define P_AUTO_EOB		BIT(6)
295 #define P_AUTO_EOB_SEL_SHIFT	7
296 #define P_AUTO_EOB_SEL_512	(0 << P_AUTO_EOB_SEL_SHIFT)
297 #define P_AUTO_EOB_SEL_256	(1 << P_AUTO_EOB_SEL_SHIFT)
298 #define P_AUTO_EOB_SEL_128	(2 << P_AUTO_EOB_SEL_SHIFT)
299 #define P_AUTO_EOB_SEL_64	(3 << P_AUTO_EOB_SEL_SHIFT)
300 #define P_PREFETCH_LIMIT_SHIFT	9
301 #define P_PREFETCH_LIMIT_32	(0 << P_PREFETCH_LIMIT_SHIFT)
302 #define P_PREFETCH_LIMIT_16	(1 << P_PREFETCH_LIMIT_SHIFT)
303 #define P_PREFETCH_LIMIT_4	(2 << P_PREFETCH_LIMIT_SHIFT)
304 #define P_WRITE_NWD		BIT(11)
305 #define P_LOCK_GROUP_SHIFT	16
306 #define P_LOCK_GROUP_MASK	0x1F
307 
308 /* BAM_DESC_CNT_TRSHLD */
309 #define CNT_TRSHLD		0xffff
310 #define DEFAULT_CNT_THRSHLD	0x4
311 
312 /* BAM_IRQ_SRCS */
313 #define BAM_IRQ			BIT(31)
314 #define P_IRQ			0x7fffffff
315 
316 /* BAM_IRQ_SRCS_MSK */
317 #define BAM_IRQ_MSK		BAM_IRQ
318 #define P_IRQ_MSK		P_IRQ
319 
320 /* BAM_IRQ_STTS */
321 #define BAM_TIMER_IRQ		BIT(4)
322 #define BAM_EMPTY_IRQ		BIT(3)
323 #define BAM_ERROR_IRQ		BIT(2)
324 #define BAM_HRESP_ERR_IRQ	BIT(1)
325 
326 /* BAM_IRQ_CLR */
327 #define BAM_TIMER_CLR		BIT(4)
328 #define BAM_EMPTY_CLR		BIT(3)
329 #define BAM_ERROR_CLR		BIT(2)
330 #define BAM_HRESP_ERR_CLR	BIT(1)
331 
332 /* BAM_IRQ_EN */
333 #define BAM_TIMER_EN		BIT(4)
334 #define BAM_EMPTY_EN		BIT(3)
335 #define BAM_ERROR_EN		BIT(2)
336 #define BAM_HRESP_ERR_EN	BIT(1)
337 
338 /* BAM_P_IRQ_EN */
339 #define P_PRCSD_DESC_EN		BIT(0)
340 #define P_TIMER_EN		BIT(1)
341 #define P_WAKE_EN		BIT(2)
342 #define P_OUT_OF_DESC_EN	BIT(3)
343 #define P_ERR_EN		BIT(4)
344 #define P_TRNSFR_END_EN		BIT(5)
345 #define P_DEFAULT_IRQS_EN	(P_PRCSD_DESC_EN | P_ERR_EN | P_TRNSFR_END_EN)
346 
347 /* BAM_P_SW_OFSTS */
348 #define P_SW_OFSTS_MASK		0xffff
349 
350 #define BAM_DESC_FIFO_SIZE	SZ_32K
351 #define MAX_DESCRIPTORS (BAM_DESC_FIFO_SIZE / sizeof(struct bam_desc_hw) - 1)
352 #define BAM_FIFO_SIZE	(SZ_32K - 8)
353 #define IS_BUSY(chan)	(CIRC_SPACE(bchan->tail, bchan->head,\
354 			 MAX_DESCRIPTORS + 1) == 0)
355 
356 struct bam_chan {
357 	struct virt_dma_chan vc;
358 
359 	struct bam_device *bdev;
360 
361 	/* configuration from device tree */
362 	u32 id;
363 
364 	/* runtime configuration */
365 	struct dma_slave_config slave;
366 
367 	/* fifo storage */
368 	struct bam_desc_hw *fifo_virt;
369 	dma_addr_t fifo_phys;
370 
371 	/* fifo markers */
372 	unsigned short head;		/* start of active descriptor entries */
373 	unsigned short tail;		/* end of active descriptor entries */
374 
375 	unsigned int initialized;	/* is the channel hw initialized? */
376 	unsigned int paused;		/* is the channel paused? */
377 	unsigned int reconfigure;	/* new slave config? */
378 	/* list of descriptors currently processed */
379 	struct list_head desc_list;
380 
381 	struct list_head node;
382 };
383 
384 static inline struct bam_chan *to_bam_chan(struct dma_chan *common)
385 {
386 	return container_of(common, struct bam_chan, vc.chan);
387 }
388 
389 struct bam_device {
390 	void __iomem *regs;
391 	struct device *dev;
392 	struct dma_device common;
393 	struct device_dma_parameters dma_parms;
394 	struct bam_chan *channels;
395 	u32 num_channels;
396 	u32 num_ees;
397 
398 	/* execution environment ID, from DT */
399 	u32 ee;
400 	bool controlled_remotely;
401 
402 	const struct reg_offset_data *layout;
403 
404 	struct clk *bamclk;
405 	int irq;
406 
407 	/* dma start transaction tasklet */
408 	struct tasklet_struct task;
409 };
410 
411 /**
412  * bam_addr - returns BAM register address
413  * @bdev: bam device
414  * @pipe: pipe instance (ignored when register doesn't have multiple instances)
415  * @reg:  register enum
416  */
417 static inline void __iomem *bam_addr(struct bam_device *bdev, u32 pipe,
418 		enum bam_reg reg)
419 {
420 	const struct reg_offset_data r = bdev->layout[reg];
421 
422 	return bdev->regs + r.base_offset +
423 		r.pipe_mult * pipe +
424 		r.evnt_mult * pipe +
425 		r.ee_mult * bdev->ee;
426 }
427 
428 /**
429  * bam_reset_channel - Reset individual BAM DMA channel
430  * @bchan: bam channel
431  *
432  * This function resets a specific BAM channel
433  */
434 static void bam_reset_channel(struct bam_chan *bchan)
435 {
436 	struct bam_device *bdev = bchan->bdev;
437 
438 	lockdep_assert_held(&bchan->vc.lock);
439 
440 	/* reset channel */
441 	writel_relaxed(1, bam_addr(bdev, bchan->id, BAM_P_RST));
442 	writel_relaxed(0, bam_addr(bdev, bchan->id, BAM_P_RST));
443 
444 	/* don't allow cpu to reorder BAM register accesses done after this */
445 	wmb();
446 
447 	/* make sure hw is initialized when channel is used the first time  */
448 	bchan->initialized = 0;
449 }
450 
451 /**
452  * bam_chan_init_hw - Initialize channel hardware
453  * @bchan: bam channel
454  *
455  * This function resets and initializes the BAM channel
456  */
457 static void bam_chan_init_hw(struct bam_chan *bchan,
458 	enum dma_transfer_direction dir)
459 {
460 	struct bam_device *bdev = bchan->bdev;
461 	u32 val;
462 
463 	/* Reset the channel to clear internal state of the FIFO */
464 	bam_reset_channel(bchan);
465 
466 	/*
467 	 * write out 8 byte aligned address.  We have enough space for this
468 	 * because we allocated 1 more descriptor (8 bytes) than we can use
469 	 */
470 	writel_relaxed(ALIGN(bchan->fifo_phys, sizeof(struct bam_desc_hw)),
471 			bam_addr(bdev, bchan->id, BAM_P_DESC_FIFO_ADDR));
472 	writel_relaxed(BAM_FIFO_SIZE,
473 			bam_addr(bdev, bchan->id, BAM_P_FIFO_SIZES));
474 
475 	/* enable the per pipe interrupts, enable EOT, ERR, and INT irqs */
476 	writel_relaxed(P_DEFAULT_IRQS_EN,
477 			bam_addr(bdev, bchan->id, BAM_P_IRQ_EN));
478 
479 	/* unmask the specific pipe and EE combo */
480 	val = readl_relaxed(bam_addr(bdev, 0, BAM_IRQ_SRCS_MSK_EE));
481 	val |= BIT(bchan->id);
482 	writel_relaxed(val, bam_addr(bdev, 0, BAM_IRQ_SRCS_MSK_EE));
483 
484 	/* don't allow cpu to reorder the channel enable done below */
485 	wmb();
486 
487 	/* set fixed direction and mode, then enable channel */
488 	val = P_EN | P_SYS_MODE;
489 	if (dir == DMA_DEV_TO_MEM)
490 		val |= P_DIRECTION;
491 
492 	writel_relaxed(val, bam_addr(bdev, bchan->id, BAM_P_CTRL));
493 
494 	bchan->initialized = 1;
495 
496 	/* init FIFO pointers */
497 	bchan->head = 0;
498 	bchan->tail = 0;
499 }
500 
501 /**
502  * bam_alloc_chan - Allocate channel resources for DMA channel.
503  * @chan: specified channel
504  *
505  * This function allocates the FIFO descriptor memory
506  */
507 static int bam_alloc_chan(struct dma_chan *chan)
508 {
509 	struct bam_chan *bchan = to_bam_chan(chan);
510 	struct bam_device *bdev = bchan->bdev;
511 
512 	if (bchan->fifo_virt)
513 		return 0;
514 
515 	/* allocate FIFO descriptor space, but only if necessary */
516 	bchan->fifo_virt = dma_alloc_wc(bdev->dev, BAM_DESC_FIFO_SIZE,
517 					&bchan->fifo_phys, GFP_KERNEL);
518 
519 	if (!bchan->fifo_virt) {
520 		dev_err(bdev->dev, "Failed to allocate desc fifo\n");
521 		return -ENOMEM;
522 	}
523 
524 	return 0;
525 }
526 
527 /**
528  * bam_free_chan - Frees dma resources associated with specific channel
529  * @chan: specified channel
530  *
531  * Free the allocated fifo descriptor memory and channel resources
532  *
533  */
534 static void bam_free_chan(struct dma_chan *chan)
535 {
536 	struct bam_chan *bchan = to_bam_chan(chan);
537 	struct bam_device *bdev = bchan->bdev;
538 	u32 val;
539 	unsigned long flags;
540 	int ret;
541 
542 	ret = pm_runtime_get_sync(bdev->dev);
543 	if (ret < 0)
544 		return;
545 
546 	vchan_free_chan_resources(to_virt_chan(chan));
547 
548 	if (!list_empty(&bchan->desc_list)) {
549 		dev_err(bchan->bdev->dev, "Cannot free busy channel\n");
550 		goto err;
551 	}
552 
553 	spin_lock_irqsave(&bchan->vc.lock, flags);
554 	bam_reset_channel(bchan);
555 	spin_unlock_irqrestore(&bchan->vc.lock, flags);
556 
557 	dma_free_wc(bdev->dev, BAM_DESC_FIFO_SIZE, bchan->fifo_virt,
558 		    bchan->fifo_phys);
559 	bchan->fifo_virt = NULL;
560 
561 	/* mask irq for pipe/channel */
562 	val = readl_relaxed(bam_addr(bdev, 0, BAM_IRQ_SRCS_MSK_EE));
563 	val &= ~BIT(bchan->id);
564 	writel_relaxed(val, bam_addr(bdev, 0, BAM_IRQ_SRCS_MSK_EE));
565 
566 	/* disable irq */
567 	writel_relaxed(0, bam_addr(bdev, bchan->id, BAM_P_IRQ_EN));
568 
569 err:
570 	pm_runtime_mark_last_busy(bdev->dev);
571 	pm_runtime_put_autosuspend(bdev->dev);
572 }
573 
574 /**
575  * bam_slave_config - set slave configuration for channel
576  * @chan: dma channel
577  * @cfg: slave configuration
578  *
579  * Sets slave configuration for channel
580  *
581  */
582 static int bam_slave_config(struct dma_chan *chan,
583 			    struct dma_slave_config *cfg)
584 {
585 	struct bam_chan *bchan = to_bam_chan(chan);
586 	unsigned long flag;
587 
588 	spin_lock_irqsave(&bchan->vc.lock, flag);
589 	memcpy(&bchan->slave, cfg, sizeof(*cfg));
590 	bchan->reconfigure = 1;
591 	spin_unlock_irqrestore(&bchan->vc.lock, flag);
592 
593 	return 0;
594 }
595 
596 /**
597  * bam_prep_slave_sg - Prep slave sg transaction
598  *
599  * @chan: dma channel
600  * @sgl: scatter gather list
601  * @sg_len: length of sg
602  * @direction: DMA transfer direction
603  * @flags: DMA flags
604  * @context: transfer context (unused)
605  */
606 static struct dma_async_tx_descriptor *bam_prep_slave_sg(struct dma_chan *chan,
607 	struct scatterlist *sgl, unsigned int sg_len,
608 	enum dma_transfer_direction direction, unsigned long flags,
609 	void *context)
610 {
611 	struct bam_chan *bchan = to_bam_chan(chan);
612 	struct bam_device *bdev = bchan->bdev;
613 	struct bam_async_desc *async_desc;
614 	struct scatterlist *sg;
615 	u32 i;
616 	struct bam_desc_hw *desc;
617 	unsigned int num_alloc = 0;
618 
619 
620 	if (!is_slave_direction(direction)) {
621 		dev_err(bdev->dev, "invalid dma direction\n");
622 		return NULL;
623 	}
624 
625 	/* calculate number of required entries */
626 	for_each_sg(sgl, sg, sg_len, i)
627 		num_alloc += DIV_ROUND_UP(sg_dma_len(sg), BAM_FIFO_SIZE);
628 
629 	/* allocate enough room to accomodate the number of entries */
630 	async_desc = kzalloc(sizeof(*async_desc) +
631 			(num_alloc * sizeof(struct bam_desc_hw)), GFP_NOWAIT);
632 
633 	if (!async_desc)
634 		goto err_out;
635 
636 	if (flags & DMA_PREP_FENCE)
637 		async_desc->flags |= DESC_FLAG_NWD;
638 
639 	if (flags & DMA_PREP_INTERRUPT)
640 		async_desc->flags |= DESC_FLAG_EOT;
641 
642 	async_desc->num_desc = num_alloc;
643 	async_desc->curr_desc = async_desc->desc;
644 	async_desc->dir = direction;
645 
646 	/* fill in temporary descriptors */
647 	desc = async_desc->desc;
648 	for_each_sg(sgl, sg, sg_len, i) {
649 		unsigned int remainder = sg_dma_len(sg);
650 		unsigned int curr_offset = 0;
651 
652 		do {
653 			if (flags & DMA_PREP_CMD)
654 				desc->flags |= cpu_to_le16(DESC_FLAG_CMD);
655 
656 			desc->addr = cpu_to_le32(sg_dma_address(sg) +
657 						 curr_offset);
658 
659 			if (remainder > BAM_FIFO_SIZE) {
660 				desc->size = cpu_to_le16(BAM_FIFO_SIZE);
661 				remainder -= BAM_FIFO_SIZE;
662 				curr_offset += BAM_FIFO_SIZE;
663 			} else {
664 				desc->size = cpu_to_le16(remainder);
665 				remainder = 0;
666 			}
667 
668 			async_desc->length += desc->size;
669 			desc++;
670 		} while (remainder > 0);
671 	}
672 
673 	return vchan_tx_prep(&bchan->vc, &async_desc->vd, flags);
674 
675 err_out:
676 	kfree(async_desc);
677 	return NULL;
678 }
679 
680 /**
681  * bam_dma_terminate_all - terminate all transactions on a channel
682  * @bchan: bam dma channel
683  *
684  * Dequeues and frees all transactions
685  * No callbacks are done
686  *
687  */
688 static int bam_dma_terminate_all(struct dma_chan *chan)
689 {
690 	struct bam_chan *bchan = to_bam_chan(chan);
691 	struct bam_async_desc *async_desc, *tmp;
692 	unsigned long flag;
693 	LIST_HEAD(head);
694 
695 	/* remove all transactions, including active transaction */
696 	spin_lock_irqsave(&bchan->vc.lock, flag);
697 	list_for_each_entry_safe(async_desc, tmp,
698 				 &bchan->desc_list, desc_node) {
699 		list_add(&async_desc->vd.node, &bchan->vc.desc_issued);
700 		list_del(&async_desc->desc_node);
701 	}
702 
703 	vchan_get_all_descriptors(&bchan->vc, &head);
704 	spin_unlock_irqrestore(&bchan->vc.lock, flag);
705 
706 	vchan_dma_desc_free_list(&bchan->vc, &head);
707 
708 	return 0;
709 }
710 
711 /**
712  * bam_pause - Pause DMA channel
713  * @chan: dma channel
714  *
715  */
716 static int bam_pause(struct dma_chan *chan)
717 {
718 	struct bam_chan *bchan = to_bam_chan(chan);
719 	struct bam_device *bdev = bchan->bdev;
720 	unsigned long flag;
721 	int ret;
722 
723 	ret = pm_runtime_get_sync(bdev->dev);
724 	if (ret < 0)
725 		return ret;
726 
727 	spin_lock_irqsave(&bchan->vc.lock, flag);
728 	writel_relaxed(1, bam_addr(bdev, bchan->id, BAM_P_HALT));
729 	bchan->paused = 1;
730 	spin_unlock_irqrestore(&bchan->vc.lock, flag);
731 	pm_runtime_mark_last_busy(bdev->dev);
732 	pm_runtime_put_autosuspend(bdev->dev);
733 
734 	return 0;
735 }
736 
737 /**
738  * bam_resume - Resume DMA channel operations
739  * @chan: dma channel
740  *
741  */
742 static int bam_resume(struct dma_chan *chan)
743 {
744 	struct bam_chan *bchan = to_bam_chan(chan);
745 	struct bam_device *bdev = bchan->bdev;
746 	unsigned long flag;
747 	int ret;
748 
749 	ret = pm_runtime_get_sync(bdev->dev);
750 	if (ret < 0)
751 		return ret;
752 
753 	spin_lock_irqsave(&bchan->vc.lock, flag);
754 	writel_relaxed(0, bam_addr(bdev, bchan->id, BAM_P_HALT));
755 	bchan->paused = 0;
756 	spin_unlock_irqrestore(&bchan->vc.lock, flag);
757 	pm_runtime_mark_last_busy(bdev->dev);
758 	pm_runtime_put_autosuspend(bdev->dev);
759 
760 	return 0;
761 }
762 
763 /**
764  * process_channel_irqs - processes the channel interrupts
765  * @bdev: bam controller
766  *
767  * This function processes the channel interrupts
768  *
769  */
770 static u32 process_channel_irqs(struct bam_device *bdev)
771 {
772 	u32 i, srcs, pipe_stts, offset, avail;
773 	unsigned long flags;
774 	struct bam_async_desc *async_desc, *tmp;
775 
776 	srcs = readl_relaxed(bam_addr(bdev, 0, BAM_IRQ_SRCS_EE));
777 
778 	/* return early if no pipe/channel interrupts are present */
779 	if (!(srcs & P_IRQ))
780 		return srcs;
781 
782 	for (i = 0; i < bdev->num_channels; i++) {
783 		struct bam_chan *bchan = &bdev->channels[i];
784 
785 		if (!(srcs & BIT(i)))
786 			continue;
787 
788 		/* clear pipe irq */
789 		pipe_stts = readl_relaxed(bam_addr(bdev, i, BAM_P_IRQ_STTS));
790 
791 		writel_relaxed(pipe_stts, bam_addr(bdev, i, BAM_P_IRQ_CLR));
792 
793 		spin_lock_irqsave(&bchan->vc.lock, flags);
794 
795 		offset = readl_relaxed(bam_addr(bdev, i, BAM_P_SW_OFSTS)) &
796 				       P_SW_OFSTS_MASK;
797 		offset /= sizeof(struct bam_desc_hw);
798 
799 		/* Number of bytes available to read */
800 		avail = CIRC_CNT(offset, bchan->head, MAX_DESCRIPTORS + 1);
801 
802 		list_for_each_entry_safe(async_desc, tmp,
803 					 &bchan->desc_list, desc_node) {
804 			/* Not enough data to read */
805 			if (avail < async_desc->xfer_len)
806 				break;
807 
808 			/* manage FIFO */
809 			bchan->head += async_desc->xfer_len;
810 			bchan->head %= MAX_DESCRIPTORS;
811 
812 			async_desc->num_desc -= async_desc->xfer_len;
813 			async_desc->curr_desc += async_desc->xfer_len;
814 			avail -= async_desc->xfer_len;
815 
816 			/*
817 			 * if complete, process cookie. Otherwise
818 			 * push back to front of desc_issued so that
819 			 * it gets restarted by the tasklet
820 			 */
821 			if (!async_desc->num_desc) {
822 				vchan_cookie_complete(&async_desc->vd);
823 			} else {
824 				list_add(&async_desc->vd.node,
825 					 &bchan->vc.desc_issued);
826 			}
827 			list_del(&async_desc->desc_node);
828 		}
829 
830 		spin_unlock_irqrestore(&bchan->vc.lock, flags);
831 	}
832 
833 	return srcs;
834 }
835 
836 /**
837  * bam_dma_irq - irq handler for bam controller
838  * @irq: IRQ of interrupt
839  * @data: callback data
840  *
841  * IRQ handler for the bam controller
842  */
843 static irqreturn_t bam_dma_irq(int irq, void *data)
844 {
845 	struct bam_device *bdev = data;
846 	u32 clr_mask = 0, srcs = 0;
847 	int ret;
848 
849 	srcs |= process_channel_irqs(bdev);
850 
851 	/* kick off tasklet to start next dma transfer */
852 	if (srcs & P_IRQ)
853 		tasklet_schedule(&bdev->task);
854 
855 	ret = pm_runtime_get_sync(bdev->dev);
856 	if (ret < 0)
857 		return ret;
858 
859 	if (srcs & BAM_IRQ) {
860 		clr_mask = readl_relaxed(bam_addr(bdev, 0, BAM_IRQ_STTS));
861 
862 		/*
863 		 * don't allow reorder of the various accesses to the BAM
864 		 * registers
865 		 */
866 		mb();
867 
868 		writel_relaxed(clr_mask, bam_addr(bdev, 0, BAM_IRQ_CLR));
869 	}
870 
871 	pm_runtime_mark_last_busy(bdev->dev);
872 	pm_runtime_put_autosuspend(bdev->dev);
873 
874 	return IRQ_HANDLED;
875 }
876 
877 /**
878  * bam_tx_status - returns status of transaction
879  * @chan: dma channel
880  * @cookie: transaction cookie
881  * @txstate: DMA transaction state
882  *
883  * Return status of dma transaction
884  */
885 static enum dma_status bam_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
886 		struct dma_tx_state *txstate)
887 {
888 	struct bam_chan *bchan = to_bam_chan(chan);
889 	struct bam_async_desc *async_desc;
890 	struct virt_dma_desc *vd;
891 	int ret;
892 	size_t residue = 0;
893 	unsigned int i;
894 	unsigned long flags;
895 
896 	ret = dma_cookie_status(chan, cookie, txstate);
897 	if (ret == DMA_COMPLETE)
898 		return ret;
899 
900 	if (!txstate)
901 		return bchan->paused ? DMA_PAUSED : ret;
902 
903 	spin_lock_irqsave(&bchan->vc.lock, flags);
904 	vd = vchan_find_desc(&bchan->vc, cookie);
905 	if (vd) {
906 		residue = container_of(vd, struct bam_async_desc, vd)->length;
907 	} else {
908 		list_for_each_entry(async_desc, &bchan->desc_list, desc_node) {
909 			if (async_desc->vd.tx.cookie != cookie)
910 				continue;
911 
912 			for (i = 0; i < async_desc->num_desc; i++)
913 				residue += async_desc->curr_desc[i].size;
914 		}
915 	}
916 
917 	spin_unlock_irqrestore(&bchan->vc.lock, flags);
918 
919 	dma_set_residue(txstate, residue);
920 
921 	if (ret == DMA_IN_PROGRESS && bchan->paused)
922 		ret = DMA_PAUSED;
923 
924 	return ret;
925 }
926 
927 /**
928  * bam_apply_new_config
929  * @bchan: bam dma channel
930  * @dir: DMA direction
931  */
932 static void bam_apply_new_config(struct bam_chan *bchan,
933 	enum dma_transfer_direction dir)
934 {
935 	struct bam_device *bdev = bchan->bdev;
936 	u32 maxburst;
937 
938 	if (!bdev->controlled_remotely) {
939 		if (dir == DMA_DEV_TO_MEM)
940 			maxburst = bchan->slave.src_maxburst;
941 		else
942 			maxburst = bchan->slave.dst_maxburst;
943 
944 		writel_relaxed(maxburst,
945 			       bam_addr(bdev, 0, BAM_DESC_CNT_TRSHLD));
946 	}
947 
948 	bchan->reconfigure = 0;
949 }
950 
951 /**
952  * bam_start_dma - start next transaction
953  * @bchan - bam dma channel
954  */
955 static void bam_start_dma(struct bam_chan *bchan)
956 {
957 	struct virt_dma_desc *vd = vchan_next_desc(&bchan->vc);
958 	struct bam_device *bdev = bchan->bdev;
959 	struct bam_async_desc *async_desc = NULL;
960 	struct bam_desc_hw *desc;
961 	struct bam_desc_hw *fifo = PTR_ALIGN(bchan->fifo_virt,
962 					sizeof(struct bam_desc_hw));
963 	int ret;
964 	unsigned int avail;
965 	struct dmaengine_desc_callback cb;
966 
967 	lockdep_assert_held(&bchan->vc.lock);
968 
969 	if (!vd)
970 		return;
971 
972 	ret = pm_runtime_get_sync(bdev->dev);
973 	if (ret < 0)
974 		return;
975 
976 	while (vd && !IS_BUSY(bchan)) {
977 		list_del(&vd->node);
978 
979 		async_desc = container_of(vd, struct bam_async_desc, vd);
980 
981 		/* on first use, initialize the channel hardware */
982 		if (!bchan->initialized)
983 			bam_chan_init_hw(bchan, async_desc->dir);
984 
985 		/* apply new slave config changes, if necessary */
986 		if (bchan->reconfigure)
987 			bam_apply_new_config(bchan, async_desc->dir);
988 
989 		desc = async_desc->curr_desc;
990 		avail = CIRC_SPACE(bchan->tail, bchan->head,
991 				   MAX_DESCRIPTORS + 1);
992 
993 		if (async_desc->num_desc > avail)
994 			async_desc->xfer_len = avail;
995 		else
996 			async_desc->xfer_len = async_desc->num_desc;
997 
998 		/* set any special flags on the last descriptor */
999 		if (async_desc->num_desc == async_desc->xfer_len)
1000 			desc[async_desc->xfer_len - 1].flags |=
1001 						cpu_to_le16(async_desc->flags);
1002 
1003 		vd = vchan_next_desc(&bchan->vc);
1004 
1005 		dmaengine_desc_get_callback(&async_desc->vd.tx, &cb);
1006 
1007 		/*
1008 		 * An interrupt is generated at this desc, if
1009 		 *  - FIFO is FULL.
1010 		 *  - No more descriptors to add.
1011 		 *  - If a callback completion was requested for this DESC,
1012 		 *     In this case, BAM will deliver the completion callback
1013 		 *     for this desc and continue processing the next desc.
1014 		 */
1015 		if (((avail <= async_desc->xfer_len) || !vd ||
1016 		     dmaengine_desc_callback_valid(&cb)) &&
1017 		    !(async_desc->flags & DESC_FLAG_EOT))
1018 			desc[async_desc->xfer_len - 1].flags |=
1019 				cpu_to_le16(DESC_FLAG_INT);
1020 
1021 		if (bchan->tail + async_desc->xfer_len > MAX_DESCRIPTORS) {
1022 			u32 partial = MAX_DESCRIPTORS - bchan->tail;
1023 
1024 			memcpy(&fifo[bchan->tail], desc,
1025 			       partial * sizeof(struct bam_desc_hw));
1026 			memcpy(fifo, &desc[partial],
1027 			       (async_desc->xfer_len - partial) *
1028 				sizeof(struct bam_desc_hw));
1029 		} else {
1030 			memcpy(&fifo[bchan->tail], desc,
1031 			       async_desc->xfer_len *
1032 			       sizeof(struct bam_desc_hw));
1033 		}
1034 
1035 		bchan->tail += async_desc->xfer_len;
1036 		bchan->tail %= MAX_DESCRIPTORS;
1037 		list_add_tail(&async_desc->desc_node, &bchan->desc_list);
1038 	}
1039 
1040 	/* ensure descriptor writes and dma start not reordered */
1041 	wmb();
1042 	writel_relaxed(bchan->tail * sizeof(struct bam_desc_hw),
1043 			bam_addr(bdev, bchan->id, BAM_P_EVNT_REG));
1044 
1045 	pm_runtime_mark_last_busy(bdev->dev);
1046 	pm_runtime_put_autosuspend(bdev->dev);
1047 }
1048 
1049 /**
1050  * dma_tasklet - DMA IRQ tasklet
1051  * @data: tasklet argument (bam controller structure)
1052  *
1053  * Sets up next DMA operation and then processes all completed transactions
1054  */
1055 static void dma_tasklet(unsigned long data)
1056 {
1057 	struct bam_device *bdev = (struct bam_device *)data;
1058 	struct bam_chan *bchan;
1059 	unsigned long flags;
1060 	unsigned int i;
1061 
1062 	/* go through the channels and kick off transactions */
1063 	for (i = 0; i < bdev->num_channels; i++) {
1064 		bchan = &bdev->channels[i];
1065 		spin_lock_irqsave(&bchan->vc.lock, flags);
1066 
1067 		if (!list_empty(&bchan->vc.desc_issued) && !IS_BUSY(bchan))
1068 			bam_start_dma(bchan);
1069 		spin_unlock_irqrestore(&bchan->vc.lock, flags);
1070 	}
1071 
1072 }
1073 
1074 /**
1075  * bam_issue_pending - starts pending transactions
1076  * @chan: dma channel
1077  *
1078  * Calls tasklet directly which in turn starts any pending transactions
1079  */
1080 static void bam_issue_pending(struct dma_chan *chan)
1081 {
1082 	struct bam_chan *bchan = to_bam_chan(chan);
1083 	unsigned long flags;
1084 
1085 	spin_lock_irqsave(&bchan->vc.lock, flags);
1086 
1087 	/* if work pending and idle, start a transaction */
1088 	if (vchan_issue_pending(&bchan->vc) && !IS_BUSY(bchan))
1089 		bam_start_dma(bchan);
1090 
1091 	spin_unlock_irqrestore(&bchan->vc.lock, flags);
1092 }
1093 
1094 /**
1095  * bam_dma_free_desc - free descriptor memory
1096  * @vd: virtual descriptor
1097  *
1098  */
1099 static void bam_dma_free_desc(struct virt_dma_desc *vd)
1100 {
1101 	struct bam_async_desc *async_desc = container_of(vd,
1102 			struct bam_async_desc, vd);
1103 
1104 	kfree(async_desc);
1105 }
1106 
1107 static struct dma_chan *bam_dma_xlate(struct of_phandle_args *dma_spec,
1108 		struct of_dma *of)
1109 {
1110 	struct bam_device *bdev = container_of(of->of_dma_data,
1111 					struct bam_device, common);
1112 	unsigned int request;
1113 
1114 	if (dma_spec->args_count != 1)
1115 		return NULL;
1116 
1117 	request = dma_spec->args[0];
1118 	if (request >= bdev->num_channels)
1119 		return NULL;
1120 
1121 	return dma_get_slave_channel(&(bdev->channels[request].vc.chan));
1122 }
1123 
1124 /**
1125  * bam_init
1126  * @bdev: bam device
1127  *
1128  * Initialization helper for global bam registers
1129  */
1130 static int bam_init(struct bam_device *bdev)
1131 {
1132 	u32 val;
1133 
1134 	/* read revision and configuration information */
1135 	if (!bdev->num_ees) {
1136 		val = readl_relaxed(bam_addr(bdev, 0, BAM_REVISION));
1137 		bdev->num_ees = (val >> NUM_EES_SHIFT) & NUM_EES_MASK;
1138 	}
1139 
1140 	/* check that configured EE is within range */
1141 	if (bdev->ee >= bdev->num_ees)
1142 		return -EINVAL;
1143 
1144 	if (!bdev->num_channels) {
1145 		val = readl_relaxed(bam_addr(bdev, 0, BAM_NUM_PIPES));
1146 		bdev->num_channels = val & BAM_NUM_PIPES_MASK;
1147 	}
1148 
1149 	if (bdev->controlled_remotely)
1150 		return 0;
1151 
1152 	/* s/w reset bam */
1153 	/* after reset all pipes are disabled and idle */
1154 	val = readl_relaxed(bam_addr(bdev, 0, BAM_CTRL));
1155 	val |= BAM_SW_RST;
1156 	writel_relaxed(val, bam_addr(bdev, 0, BAM_CTRL));
1157 	val &= ~BAM_SW_RST;
1158 	writel_relaxed(val, bam_addr(bdev, 0, BAM_CTRL));
1159 
1160 	/* make sure previous stores are visible before enabling BAM */
1161 	wmb();
1162 
1163 	/* enable bam */
1164 	val |= BAM_EN;
1165 	writel_relaxed(val, bam_addr(bdev, 0, BAM_CTRL));
1166 
1167 	/* set descriptor threshhold, start with 4 bytes */
1168 	writel_relaxed(DEFAULT_CNT_THRSHLD,
1169 			bam_addr(bdev, 0, BAM_DESC_CNT_TRSHLD));
1170 
1171 	/* Enable default set of h/w workarounds, ie all except BAM_FULL_PIPE */
1172 	writel_relaxed(BAM_CNFG_BITS_DEFAULT, bam_addr(bdev, 0, BAM_CNFG_BITS));
1173 
1174 	/* enable irqs for errors */
1175 	writel_relaxed(BAM_ERROR_EN | BAM_HRESP_ERR_EN,
1176 			bam_addr(bdev, 0, BAM_IRQ_EN));
1177 
1178 	/* unmask global bam interrupt */
1179 	writel_relaxed(BAM_IRQ_MSK, bam_addr(bdev, 0, BAM_IRQ_SRCS_MSK_EE));
1180 
1181 	return 0;
1182 }
1183 
1184 static void bam_channel_init(struct bam_device *bdev, struct bam_chan *bchan,
1185 	u32 index)
1186 {
1187 	bchan->id = index;
1188 	bchan->bdev = bdev;
1189 
1190 	vchan_init(&bchan->vc, &bdev->common);
1191 	bchan->vc.desc_free = bam_dma_free_desc;
1192 	INIT_LIST_HEAD(&bchan->desc_list);
1193 }
1194 
1195 static const struct of_device_id bam_of_match[] = {
1196 	{ .compatible = "qcom,bam-v1.3.0", .data = &bam_v1_3_reg_info },
1197 	{ .compatible = "qcom,bam-v1.4.0", .data = &bam_v1_4_reg_info },
1198 	{ .compatible = "qcom,bam-v1.7.0", .data = &bam_v1_7_reg_info },
1199 	{}
1200 };
1201 
1202 MODULE_DEVICE_TABLE(of, bam_of_match);
1203 
1204 static int bam_dma_probe(struct platform_device *pdev)
1205 {
1206 	struct bam_device *bdev;
1207 	const struct of_device_id *match;
1208 	struct resource *iores;
1209 	int ret, i;
1210 
1211 	bdev = devm_kzalloc(&pdev->dev, sizeof(*bdev), GFP_KERNEL);
1212 	if (!bdev)
1213 		return -ENOMEM;
1214 
1215 	bdev->dev = &pdev->dev;
1216 
1217 	match = of_match_node(bam_of_match, pdev->dev.of_node);
1218 	if (!match) {
1219 		dev_err(&pdev->dev, "Unsupported BAM module\n");
1220 		return -ENODEV;
1221 	}
1222 
1223 	bdev->layout = match->data;
1224 
1225 	iores = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1226 	bdev->regs = devm_ioremap_resource(&pdev->dev, iores);
1227 	if (IS_ERR(bdev->regs))
1228 		return PTR_ERR(bdev->regs);
1229 
1230 	bdev->irq = platform_get_irq(pdev, 0);
1231 	if (bdev->irq < 0)
1232 		return bdev->irq;
1233 
1234 	ret = of_property_read_u32(pdev->dev.of_node, "qcom,ee", &bdev->ee);
1235 	if (ret) {
1236 		dev_err(bdev->dev, "Execution environment unspecified\n");
1237 		return ret;
1238 	}
1239 
1240 	bdev->controlled_remotely = of_property_read_bool(pdev->dev.of_node,
1241 						"qcom,controlled-remotely");
1242 
1243 	if (bdev->controlled_remotely) {
1244 		ret = of_property_read_u32(pdev->dev.of_node, "num-channels",
1245 					   &bdev->num_channels);
1246 		if (ret)
1247 			dev_err(bdev->dev, "num-channels unspecified in dt\n");
1248 
1249 		ret = of_property_read_u32(pdev->dev.of_node, "qcom,num-ees",
1250 					   &bdev->num_ees);
1251 		if (ret)
1252 			dev_err(bdev->dev, "num-ees unspecified in dt\n");
1253 	}
1254 
1255 	bdev->bamclk = devm_clk_get(bdev->dev, "bam_clk");
1256 	if (IS_ERR(bdev->bamclk)) {
1257 		if (!bdev->controlled_remotely)
1258 			return PTR_ERR(bdev->bamclk);
1259 
1260 		bdev->bamclk = NULL;
1261 	}
1262 
1263 	ret = clk_prepare_enable(bdev->bamclk);
1264 	if (ret) {
1265 		dev_err(bdev->dev, "failed to prepare/enable clock\n");
1266 		return ret;
1267 	}
1268 
1269 	ret = bam_init(bdev);
1270 	if (ret)
1271 		goto err_disable_clk;
1272 
1273 	tasklet_init(&bdev->task, dma_tasklet, (unsigned long)bdev);
1274 
1275 	bdev->channels = devm_kcalloc(bdev->dev, bdev->num_channels,
1276 				sizeof(*bdev->channels), GFP_KERNEL);
1277 
1278 	if (!bdev->channels) {
1279 		ret = -ENOMEM;
1280 		goto err_tasklet_kill;
1281 	}
1282 
1283 	/* allocate and initialize channels */
1284 	INIT_LIST_HEAD(&bdev->common.channels);
1285 
1286 	for (i = 0; i < bdev->num_channels; i++)
1287 		bam_channel_init(bdev, &bdev->channels[i], i);
1288 
1289 	ret = devm_request_irq(bdev->dev, bdev->irq, bam_dma_irq,
1290 			IRQF_TRIGGER_HIGH, "bam_dma", bdev);
1291 	if (ret)
1292 		goto err_bam_channel_exit;
1293 
1294 	/* set max dma segment size */
1295 	bdev->common.dev = bdev->dev;
1296 	bdev->common.dev->dma_parms = &bdev->dma_parms;
1297 	ret = dma_set_max_seg_size(bdev->common.dev, BAM_FIFO_SIZE);
1298 	if (ret) {
1299 		dev_err(bdev->dev, "cannot set maximum segment size\n");
1300 		goto err_bam_channel_exit;
1301 	}
1302 
1303 	platform_set_drvdata(pdev, bdev);
1304 
1305 	/* set capabilities */
1306 	dma_cap_zero(bdev->common.cap_mask);
1307 	dma_cap_set(DMA_SLAVE, bdev->common.cap_mask);
1308 
1309 	/* initialize dmaengine apis */
1310 	bdev->common.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
1311 	bdev->common.residue_granularity = DMA_RESIDUE_GRANULARITY_SEGMENT;
1312 	bdev->common.src_addr_widths = DMA_SLAVE_BUSWIDTH_4_BYTES;
1313 	bdev->common.dst_addr_widths = DMA_SLAVE_BUSWIDTH_4_BYTES;
1314 	bdev->common.device_alloc_chan_resources = bam_alloc_chan;
1315 	bdev->common.device_free_chan_resources = bam_free_chan;
1316 	bdev->common.device_prep_slave_sg = bam_prep_slave_sg;
1317 	bdev->common.device_config = bam_slave_config;
1318 	bdev->common.device_pause = bam_pause;
1319 	bdev->common.device_resume = bam_resume;
1320 	bdev->common.device_terminate_all = bam_dma_terminate_all;
1321 	bdev->common.device_issue_pending = bam_issue_pending;
1322 	bdev->common.device_tx_status = bam_tx_status;
1323 	bdev->common.dev = bdev->dev;
1324 
1325 	ret = dma_async_device_register(&bdev->common);
1326 	if (ret) {
1327 		dev_err(bdev->dev, "failed to register dma async device\n");
1328 		goto err_bam_channel_exit;
1329 	}
1330 
1331 	ret = of_dma_controller_register(pdev->dev.of_node, bam_dma_xlate,
1332 					&bdev->common);
1333 	if (ret)
1334 		goto err_unregister_dma;
1335 
1336 	if (bdev->controlled_remotely) {
1337 		pm_runtime_disable(&pdev->dev);
1338 		return 0;
1339 	}
1340 
1341 	pm_runtime_irq_safe(&pdev->dev);
1342 	pm_runtime_set_autosuspend_delay(&pdev->dev, BAM_DMA_AUTOSUSPEND_DELAY);
1343 	pm_runtime_use_autosuspend(&pdev->dev);
1344 	pm_runtime_mark_last_busy(&pdev->dev);
1345 	pm_runtime_set_active(&pdev->dev);
1346 	pm_runtime_enable(&pdev->dev);
1347 
1348 	return 0;
1349 
1350 err_unregister_dma:
1351 	dma_async_device_unregister(&bdev->common);
1352 err_bam_channel_exit:
1353 	for (i = 0; i < bdev->num_channels; i++)
1354 		tasklet_kill(&bdev->channels[i].vc.task);
1355 err_tasklet_kill:
1356 	tasklet_kill(&bdev->task);
1357 err_disable_clk:
1358 	clk_disable_unprepare(bdev->bamclk);
1359 
1360 	return ret;
1361 }
1362 
1363 static int bam_dma_remove(struct platform_device *pdev)
1364 {
1365 	struct bam_device *bdev = platform_get_drvdata(pdev);
1366 	u32 i;
1367 
1368 	pm_runtime_force_suspend(&pdev->dev);
1369 
1370 	of_dma_controller_free(pdev->dev.of_node);
1371 	dma_async_device_unregister(&bdev->common);
1372 
1373 	/* mask all interrupts for this execution environment */
1374 	writel_relaxed(0, bam_addr(bdev, 0,  BAM_IRQ_SRCS_MSK_EE));
1375 
1376 	devm_free_irq(bdev->dev, bdev->irq, bdev);
1377 
1378 	for (i = 0; i < bdev->num_channels; i++) {
1379 		bam_dma_terminate_all(&bdev->channels[i].vc.chan);
1380 		tasklet_kill(&bdev->channels[i].vc.task);
1381 
1382 		if (!bdev->channels[i].fifo_virt)
1383 			continue;
1384 
1385 		dma_free_wc(bdev->dev, BAM_DESC_FIFO_SIZE,
1386 			    bdev->channels[i].fifo_virt,
1387 			    bdev->channels[i].fifo_phys);
1388 	}
1389 
1390 	tasklet_kill(&bdev->task);
1391 
1392 	clk_disable_unprepare(bdev->bamclk);
1393 
1394 	return 0;
1395 }
1396 
1397 static int __maybe_unused bam_dma_runtime_suspend(struct device *dev)
1398 {
1399 	struct bam_device *bdev = dev_get_drvdata(dev);
1400 
1401 	clk_disable(bdev->bamclk);
1402 
1403 	return 0;
1404 }
1405 
1406 static int __maybe_unused bam_dma_runtime_resume(struct device *dev)
1407 {
1408 	struct bam_device *bdev = dev_get_drvdata(dev);
1409 	int ret;
1410 
1411 	ret = clk_enable(bdev->bamclk);
1412 	if (ret < 0) {
1413 		dev_err(dev, "clk_enable failed: %d\n", ret);
1414 		return ret;
1415 	}
1416 
1417 	return 0;
1418 }
1419 
1420 static int __maybe_unused bam_dma_suspend(struct device *dev)
1421 {
1422 	struct bam_device *bdev = dev_get_drvdata(dev);
1423 
1424 	if (!bdev->controlled_remotely)
1425 		pm_runtime_force_suspend(dev);
1426 
1427 	clk_unprepare(bdev->bamclk);
1428 
1429 	return 0;
1430 }
1431 
1432 static int __maybe_unused bam_dma_resume(struct device *dev)
1433 {
1434 	struct bam_device *bdev = dev_get_drvdata(dev);
1435 	int ret;
1436 
1437 	ret = clk_prepare(bdev->bamclk);
1438 	if (ret)
1439 		return ret;
1440 
1441 	if (!bdev->controlled_remotely)
1442 		pm_runtime_force_resume(dev);
1443 
1444 	return 0;
1445 }
1446 
1447 static const struct dev_pm_ops bam_dma_pm_ops = {
1448 	SET_LATE_SYSTEM_SLEEP_PM_OPS(bam_dma_suspend, bam_dma_resume)
1449 	SET_RUNTIME_PM_OPS(bam_dma_runtime_suspend, bam_dma_runtime_resume,
1450 				NULL)
1451 };
1452 
1453 static struct platform_driver bam_dma_driver = {
1454 	.probe = bam_dma_probe,
1455 	.remove = bam_dma_remove,
1456 	.driver = {
1457 		.name = "bam-dma-engine",
1458 		.pm = &bam_dma_pm_ops,
1459 		.of_match_table = bam_of_match,
1460 	},
1461 };
1462 
1463 module_platform_driver(bam_dma_driver);
1464 
1465 MODULE_AUTHOR("Andy Gross <agross@codeaurora.org>");
1466 MODULE_DESCRIPTION("QCOM BAM DMA engine driver");
1467 MODULE_LICENSE("GPL v2");
1468