xref: /linux/drivers/media/pci/intel/ipu3/ipu3-cio2.c (revision 57b54d74dd5c559bd35f2affaf11d8828aaf5733)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2017 Intel Corporation
4  *
5  * Based partially on Intel IPU4 driver written by
6  *  Sakari Ailus <sakari.ailus@linux.intel.com>
7  *  Samu Onkalo <samu.onkalo@intel.com>
8  *  Jouni Högander <jouni.hogander@intel.com>
9  *  Jouni Ukkonen <jouni.ukkonen@intel.com>
10  *  Antti Laakso <antti.laakso@intel.com>
11  * et al.
12  *
13  */
14 
15 #include <linux/delay.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/pci.h>
19 #include <linux/pm_runtime.h>
20 #include <linux/property.h>
21 #include <linux/vmalloc.h>
22 #include <media/v4l2-ctrls.h>
23 #include <media/v4l2-device.h>
24 #include <media/v4l2-event.h>
25 #include <media/v4l2-fwnode.h>
26 #include <media/v4l2-ioctl.h>
27 #include <media/videobuf2-dma-sg.h>
28 
29 #include "ipu3-cio2.h"
30 
31 struct ipu3_cio2_fmt {
32 	u32 mbus_code;
33 	u32 fourcc;
34 	u8 mipicode;
35 };
36 
37 /*
38  * These are raw formats used in Intel's third generation of
39  * Image Processing Unit known as IPU3.
40  * 10bit raw bayer packed, 32 bytes for every 25 pixels,
41  * last LSB 6 bits unused.
42  */
43 static const struct ipu3_cio2_fmt formats[] = {
44 	{	/* put default entry at beginning */
45 		.mbus_code	= MEDIA_BUS_FMT_SGRBG10_1X10,
46 		.fourcc		= V4L2_PIX_FMT_IPU3_SGRBG10,
47 		.mipicode	= 0x2b,
48 	}, {
49 		.mbus_code	= MEDIA_BUS_FMT_SGBRG10_1X10,
50 		.fourcc		= V4L2_PIX_FMT_IPU3_SGBRG10,
51 		.mipicode	= 0x2b,
52 	}, {
53 		.mbus_code	= MEDIA_BUS_FMT_SBGGR10_1X10,
54 		.fourcc		= V4L2_PIX_FMT_IPU3_SBGGR10,
55 		.mipicode	= 0x2b,
56 	}, {
57 		.mbus_code	= MEDIA_BUS_FMT_SRGGB10_1X10,
58 		.fourcc		= V4L2_PIX_FMT_IPU3_SRGGB10,
59 		.mipicode	= 0x2b,
60 	},
61 };
62 
63 /*
64  * cio2_find_format - lookup color format by fourcc or/and media bus code
65  * @pixelformat: fourcc to match, ignored if null
66  * @mbus_code: media bus code to match, ignored if null
67  */
68 static const struct ipu3_cio2_fmt *cio2_find_format(const u32 *pixelformat,
69 						    const u32 *mbus_code)
70 {
71 	unsigned int i;
72 
73 	for (i = 0; i < ARRAY_SIZE(formats); i++) {
74 		if (pixelformat && *pixelformat != formats[i].fourcc)
75 			continue;
76 		if (mbus_code && *mbus_code != formats[i].mbus_code)
77 			continue;
78 
79 		return &formats[i];
80 	}
81 
82 	return NULL;
83 }
84 
85 static inline u32 cio2_bytesperline(const unsigned int width)
86 {
87 	/*
88 	 * 64 bytes for every 50 pixels, the line length
89 	 * in bytes is multiple of 64 (line end alignment).
90 	 */
91 	return DIV_ROUND_UP(width, 50) * 64;
92 }
93 
94 /**************** FBPT operations ****************/
95 
96 static void cio2_fbpt_exit_dummy(struct cio2_device *cio2)
97 {
98 	if (cio2->dummy_lop) {
99 		dma_free_coherent(&cio2->pci_dev->dev, CIO2_PAGE_SIZE,
100 				  cio2->dummy_lop, cio2->dummy_lop_bus_addr);
101 		cio2->dummy_lop = NULL;
102 	}
103 	if (cio2->dummy_page) {
104 		dma_free_coherent(&cio2->pci_dev->dev, CIO2_PAGE_SIZE,
105 				  cio2->dummy_page, cio2->dummy_page_bus_addr);
106 		cio2->dummy_page = NULL;
107 	}
108 }
109 
110 static int cio2_fbpt_init_dummy(struct cio2_device *cio2)
111 {
112 	unsigned int i;
113 
114 	cio2->dummy_page = dma_alloc_coherent(&cio2->pci_dev->dev,
115 					      CIO2_PAGE_SIZE,
116 					      &cio2->dummy_page_bus_addr,
117 					      GFP_KERNEL);
118 	cio2->dummy_lop = dma_alloc_coherent(&cio2->pci_dev->dev,
119 					     CIO2_PAGE_SIZE,
120 					     &cio2->dummy_lop_bus_addr,
121 					     GFP_KERNEL);
122 	if (!cio2->dummy_page || !cio2->dummy_lop) {
123 		cio2_fbpt_exit_dummy(cio2);
124 		return -ENOMEM;
125 	}
126 	/*
127 	 * List of Pointers(LOP) contains 1024x32b pointers to 4KB page each
128 	 * Initialize each entry to dummy_page bus base address.
129 	 */
130 	for (i = 0; i < CIO2_PAGE_SIZE / sizeof(*cio2->dummy_lop); i++)
131 		cio2->dummy_lop[i] = cio2->dummy_page_bus_addr >> PAGE_SHIFT;
132 
133 	return 0;
134 }
135 
136 static void cio2_fbpt_entry_enable(struct cio2_device *cio2,
137 				   struct cio2_fbpt_entry entry[CIO2_MAX_LOPS])
138 {
139 	/*
140 	 * The CPU first initializes some fields in fbpt, then sets
141 	 * the VALID bit, this barrier is to ensure that the DMA(device)
142 	 * does not see the VALID bit enabled before other fields are
143 	 * initialized; otherwise it could lead to havoc.
144 	 */
145 	dma_wmb();
146 
147 	/*
148 	 * Request interrupts for start and completion
149 	 * Valid bit is applicable only to 1st entry
150 	 */
151 	entry[0].first_entry.ctrl = CIO2_FBPT_CTRL_VALID |
152 		CIO2_FBPT_CTRL_IOC | CIO2_FBPT_CTRL_IOS;
153 }
154 
155 /* Initialize fpbt entries to point to dummy frame */
156 static void cio2_fbpt_entry_init_dummy(struct cio2_device *cio2,
157 				       struct cio2_fbpt_entry
158 				       entry[CIO2_MAX_LOPS])
159 {
160 	unsigned int i;
161 
162 	entry[0].first_entry.first_page_offset = 0;
163 	entry[1].second_entry.num_of_pages =
164 		CIO2_PAGE_SIZE / sizeof(u32) * CIO2_MAX_LOPS;
165 	entry[1].second_entry.last_page_available_bytes = CIO2_PAGE_SIZE - 1;
166 
167 	for (i = 0; i < CIO2_MAX_LOPS; i++)
168 		entry[i].lop_page_addr = cio2->dummy_lop_bus_addr >> PAGE_SHIFT;
169 
170 	cio2_fbpt_entry_enable(cio2, entry);
171 }
172 
173 /* Initialize fpbt entries to point to a given buffer */
174 static void cio2_fbpt_entry_init_buf(struct cio2_device *cio2,
175 				     struct cio2_buffer *b,
176 				     struct cio2_fbpt_entry
177 				     entry[CIO2_MAX_LOPS])
178 {
179 	struct vb2_buffer *vb = &b->vbb.vb2_buf;
180 	unsigned int length = vb->planes[0].length;
181 	int remaining, i;
182 
183 	entry[0].first_entry.first_page_offset = b->offset;
184 	remaining = length + entry[0].first_entry.first_page_offset;
185 	entry[1].second_entry.num_of_pages =
186 		DIV_ROUND_UP(remaining, CIO2_PAGE_SIZE);
187 	/*
188 	 * last_page_available_bytes has the offset of the last byte in the
189 	 * last page which is still accessible by DMA. DMA cannot access
190 	 * beyond this point. Valid range for this is from 0 to 4095.
191 	 * 0 indicates 1st byte in the page is DMA accessible.
192 	 * 4095 (CIO2_PAGE_SIZE - 1) means every single byte in the last page
193 	 * is available for DMA transfer.
194 	 */
195 	entry[1].second_entry.last_page_available_bytes =
196 			(remaining & ~PAGE_MASK) ?
197 				(remaining & ~PAGE_MASK) - 1 :
198 				CIO2_PAGE_SIZE - 1;
199 	/* Fill FBPT */
200 	remaining = length;
201 	i = 0;
202 	while (remaining > 0) {
203 		entry->lop_page_addr = b->lop_bus_addr[i] >> PAGE_SHIFT;
204 		remaining -= CIO2_PAGE_SIZE / sizeof(u32) * CIO2_PAGE_SIZE;
205 		entry++;
206 		i++;
207 	}
208 
209 	/*
210 	 * The first not meaningful FBPT entry should point to a valid LOP
211 	 */
212 	entry->lop_page_addr = cio2->dummy_lop_bus_addr >> PAGE_SHIFT;
213 
214 	cio2_fbpt_entry_enable(cio2, entry);
215 }
216 
217 static int cio2_fbpt_init(struct cio2_device *cio2, struct cio2_queue *q)
218 {
219 	struct device *dev = &cio2->pci_dev->dev;
220 
221 	q->fbpt = dma_alloc_coherent(dev, CIO2_FBPT_SIZE, &q->fbpt_bus_addr,
222 				     GFP_KERNEL);
223 	if (!q->fbpt)
224 		return -ENOMEM;
225 
226 	memset(q->fbpt, 0, CIO2_FBPT_SIZE);
227 
228 	return 0;
229 }
230 
231 static void cio2_fbpt_exit(struct cio2_queue *q, struct device *dev)
232 {
233 	dma_free_coherent(dev, CIO2_FBPT_SIZE, q->fbpt, q->fbpt_bus_addr);
234 }
235 
236 /**************** CSI2 hardware setup ****************/
237 
238 /*
239  * The CSI2 receiver has several parameters affecting
240  * the receiver timings. These depend on the MIPI bus frequency
241  * F in Hz (sensor transmitter rate) as follows:
242  *     register value = (A/1e9 + B * UI) / COUNT_ACC
243  * where
244  *      UI = 1 / (2 * F) in seconds
245  *      COUNT_ACC = counter accuracy in seconds
246  *      For IPU3 COUNT_ACC = 0.0625
247  *
248  * A and B are coefficients from the table below,
249  * depending whether the register minimum or maximum value is
250  * calculated.
251  *                                     Minimum     Maximum
252  * Clock lane                          A     B     A     B
253  * reg_rx_csi_dly_cnt_termen_clane     0     0    38     0
254  * reg_rx_csi_dly_cnt_settle_clane    95    -8   300   -16
255  * Data lanes
256  * reg_rx_csi_dly_cnt_termen_dlane0    0     0    35     4
257  * reg_rx_csi_dly_cnt_settle_dlane0   85    -2   145    -6
258  * reg_rx_csi_dly_cnt_termen_dlane1    0     0    35     4
259  * reg_rx_csi_dly_cnt_settle_dlane1   85    -2   145    -6
260  * reg_rx_csi_dly_cnt_termen_dlane2    0     0    35     4
261  * reg_rx_csi_dly_cnt_settle_dlane2   85    -2   145    -6
262  * reg_rx_csi_dly_cnt_termen_dlane3    0     0    35     4
263  * reg_rx_csi_dly_cnt_settle_dlane3   85    -2   145    -6
264  *
265  * We use the minimum values of both A and B.
266  */
267 
268 /*
269  * shift for keeping value range suitable for 32-bit integer arithmetics
270  */
271 #define LIMIT_SHIFT	8
272 
273 static s32 cio2_rx_timing(s32 a, s32 b, s64 freq, int def)
274 {
275 	const u32 accinv = 16; /* invert of counter resolution */
276 	const u32 uiinv = 500000000; /* 1e9 / 2 */
277 	s32 r;
278 
279 	freq >>= LIMIT_SHIFT;
280 
281 	if (WARN_ON(freq <= 0 || freq > S32_MAX))
282 		return def;
283 	/*
284 	 * b could be 0, -2 or -8, so |accinv * b| is always
285 	 * less than (1 << ds) and thus |r| < 500000000.
286 	 */
287 	r = accinv * b * (uiinv >> LIMIT_SHIFT);
288 	r = r / (s32)freq;
289 	/* max value of a is 95 */
290 	r += accinv * a;
291 
292 	return r;
293 };
294 
295 /* Calculate the the delay value for termination enable of clock lane HS Rx */
296 static int cio2_csi2_calc_timing(struct cio2_device *cio2, struct cio2_queue *q,
297 				 struct cio2_csi2_timing *timing)
298 {
299 	struct device *dev = &cio2->pci_dev->dev;
300 	struct v4l2_querymenu qm = {.id = V4L2_CID_LINK_FREQ, };
301 	struct v4l2_ctrl *link_freq;
302 	s64 freq;
303 	int r;
304 
305 	if (!q->sensor)
306 		return -ENODEV;
307 
308 	link_freq = v4l2_ctrl_find(q->sensor->ctrl_handler, V4L2_CID_LINK_FREQ);
309 	if (!link_freq) {
310 		dev_err(dev, "failed to find LINK_FREQ\n");
311 		return -EPIPE;
312 	}
313 
314 	qm.index = v4l2_ctrl_g_ctrl(link_freq);
315 	r = v4l2_querymenu(q->sensor->ctrl_handler, &qm);
316 	if (r) {
317 		dev_err(dev, "failed to get menu item\n");
318 		return r;
319 	}
320 
321 	if (!qm.value) {
322 		dev_err(dev, "error invalid link_freq\n");
323 		return -EINVAL;
324 	}
325 	freq = qm.value;
326 
327 	timing->clk_termen = cio2_rx_timing(CIO2_CSIRX_DLY_CNT_TERMEN_CLANE_A,
328 					    CIO2_CSIRX_DLY_CNT_TERMEN_CLANE_B,
329 					    freq,
330 					    CIO2_CSIRX_DLY_CNT_TERMEN_DEFAULT);
331 	timing->clk_settle = cio2_rx_timing(CIO2_CSIRX_DLY_CNT_SETTLE_CLANE_A,
332 					    CIO2_CSIRX_DLY_CNT_SETTLE_CLANE_B,
333 					    freq,
334 					    CIO2_CSIRX_DLY_CNT_SETTLE_DEFAULT);
335 	timing->dat_termen = cio2_rx_timing(CIO2_CSIRX_DLY_CNT_TERMEN_DLANE_A,
336 					    CIO2_CSIRX_DLY_CNT_TERMEN_DLANE_B,
337 					    freq,
338 					    CIO2_CSIRX_DLY_CNT_TERMEN_DEFAULT);
339 	timing->dat_settle = cio2_rx_timing(CIO2_CSIRX_DLY_CNT_SETTLE_DLANE_A,
340 					    CIO2_CSIRX_DLY_CNT_SETTLE_DLANE_B,
341 					    freq,
342 					    CIO2_CSIRX_DLY_CNT_SETTLE_DEFAULT);
343 
344 	dev_dbg(dev, "freq ct value is %d\n", timing->clk_termen);
345 	dev_dbg(dev, "freq cs value is %d\n", timing->clk_settle);
346 	dev_dbg(dev, "freq dt value is %d\n", timing->dat_termen);
347 	dev_dbg(dev, "freq ds value is %d\n", timing->dat_settle);
348 
349 	return 0;
350 };
351 
352 static int cio2_hw_init(struct cio2_device *cio2, struct cio2_queue *q)
353 {
354 	static const int NUM_VCS = 4;
355 	static const int SID;	/* Stream id */
356 	static const int ENTRY;
357 	static const int FBPT_WIDTH = DIV_ROUND_UP(CIO2_MAX_LOPS,
358 					CIO2_FBPT_SUBENTRY_UNIT);
359 	const u32 num_buffers1 = CIO2_MAX_BUFFERS - 1;
360 	const struct ipu3_cio2_fmt *fmt;
361 	void __iomem *const base = cio2->base;
362 	u8 lanes, csi2bus = q->csi2.port;
363 	u8 sensor_vc = SENSOR_VIR_CH_DFLT;
364 	struct cio2_csi2_timing timing;
365 	int i, r;
366 
367 	fmt = cio2_find_format(NULL, &q->subdev_fmt.code);
368 	if (!fmt)
369 		return -EINVAL;
370 
371 	lanes = q->csi2.lanes;
372 
373 	r = cio2_csi2_calc_timing(cio2, q, &timing);
374 	if (r)
375 		return r;
376 
377 	writel(timing.clk_termen, q->csi_rx_base +
378 		CIO2_REG_CSIRX_DLY_CNT_TERMEN(CIO2_CSIRX_DLY_CNT_CLANE_IDX));
379 	writel(timing.clk_settle, q->csi_rx_base +
380 		CIO2_REG_CSIRX_DLY_CNT_SETTLE(CIO2_CSIRX_DLY_CNT_CLANE_IDX));
381 
382 	for (i = 0; i < lanes; i++) {
383 		writel(timing.dat_termen, q->csi_rx_base +
384 			CIO2_REG_CSIRX_DLY_CNT_TERMEN(i));
385 		writel(timing.dat_settle, q->csi_rx_base +
386 			CIO2_REG_CSIRX_DLY_CNT_SETTLE(i));
387 	}
388 
389 	writel(CIO2_PBM_WMCTRL1_MIN_2CK |
390 	       CIO2_PBM_WMCTRL1_MID1_2CK |
391 	       CIO2_PBM_WMCTRL1_MID2_2CK, base + CIO2_REG_PBM_WMCTRL1);
392 	writel(CIO2_PBM_WMCTRL2_HWM_2CK << CIO2_PBM_WMCTRL2_HWM_2CK_SHIFT |
393 	       CIO2_PBM_WMCTRL2_LWM_2CK << CIO2_PBM_WMCTRL2_LWM_2CK_SHIFT |
394 	       CIO2_PBM_WMCTRL2_OBFFWM_2CK <<
395 	       CIO2_PBM_WMCTRL2_OBFFWM_2CK_SHIFT |
396 	       CIO2_PBM_WMCTRL2_TRANSDYN << CIO2_PBM_WMCTRL2_TRANSDYN_SHIFT |
397 	       CIO2_PBM_WMCTRL2_OBFF_MEM_EN, base + CIO2_REG_PBM_WMCTRL2);
398 	writel(CIO2_PBM_ARB_CTRL_LANES_DIV <<
399 	       CIO2_PBM_ARB_CTRL_LANES_DIV_SHIFT |
400 	       CIO2_PBM_ARB_CTRL_LE_EN |
401 	       CIO2_PBM_ARB_CTRL_PLL_POST_SHTDN <<
402 	       CIO2_PBM_ARB_CTRL_PLL_POST_SHTDN_SHIFT |
403 	       CIO2_PBM_ARB_CTRL_PLL_AHD_WK_UP <<
404 	       CIO2_PBM_ARB_CTRL_PLL_AHD_WK_UP_SHIFT,
405 	       base + CIO2_REG_PBM_ARB_CTRL);
406 	writel(CIO2_CSIRX_STATUS_DLANE_HS_MASK,
407 	       q->csi_rx_base + CIO2_REG_CSIRX_STATUS_DLANE_HS);
408 	writel(CIO2_CSIRX_STATUS_DLANE_LP_MASK,
409 	       q->csi_rx_base + CIO2_REG_CSIRX_STATUS_DLANE_LP);
410 
411 	writel(CIO2_FB_HPLL_FREQ, base + CIO2_REG_FB_HPLL_FREQ);
412 	writel(CIO2_ISCLK_RATIO, base + CIO2_REG_ISCLK_RATIO);
413 
414 	/* Configure MIPI backend */
415 	for (i = 0; i < NUM_VCS; i++)
416 		writel(1, q->csi_rx_base + CIO2_REG_MIPIBE_SP_LUT_ENTRY(i));
417 
418 	/* There are 16 short packet LUT entry */
419 	for (i = 0; i < 16; i++)
420 		writel(CIO2_MIPIBE_LP_LUT_ENTRY_DISREGARD,
421 		       q->csi_rx_base + CIO2_REG_MIPIBE_LP_LUT_ENTRY(i));
422 	writel(CIO2_MIPIBE_GLOBAL_LUT_DISREGARD,
423 	       q->csi_rx_base + CIO2_REG_MIPIBE_GLOBAL_LUT_DISREGARD);
424 
425 	writel(CIO2_INT_EN_EXT_IE_MASK, base + CIO2_REG_INT_EN_EXT_IE);
426 	writel(CIO2_IRQCTRL_MASK, q->csi_rx_base + CIO2_REG_IRQCTRL_MASK);
427 	writel(CIO2_IRQCTRL_MASK, q->csi_rx_base + CIO2_REG_IRQCTRL_ENABLE);
428 	writel(0, q->csi_rx_base + CIO2_REG_IRQCTRL_EDGE);
429 	writel(0, q->csi_rx_base + CIO2_REG_IRQCTRL_LEVEL_NOT_PULSE);
430 	writel(CIO2_INT_EN_EXT_OE_MASK, base + CIO2_REG_INT_EN_EXT_OE);
431 
432 	writel(CIO2_REG_INT_EN_IRQ | CIO2_INT_IOC(CIO2_DMA_CHAN) |
433 	       CIO2_REG_INT_EN_IOS(CIO2_DMA_CHAN),
434 	       base + CIO2_REG_INT_EN);
435 
436 	writel((CIO2_PXM_PXF_FMT_CFG_BPP_10 | CIO2_PXM_PXF_FMT_CFG_PCK_64B)
437 	       << CIO2_PXM_PXF_FMT_CFG_SID0_SHIFT,
438 	       base + CIO2_REG_PXM_PXF_FMT_CFG0(csi2bus));
439 	writel(SID << CIO2_MIPIBE_LP_LUT_ENTRY_SID_SHIFT |
440 	       sensor_vc << CIO2_MIPIBE_LP_LUT_ENTRY_VC_SHIFT |
441 	       fmt->mipicode << CIO2_MIPIBE_LP_LUT_ENTRY_FORMAT_TYPE_SHIFT,
442 	       q->csi_rx_base + CIO2_REG_MIPIBE_LP_LUT_ENTRY(ENTRY));
443 	writel(0, q->csi_rx_base + CIO2_REG_MIPIBE_COMP_FORMAT(sensor_vc));
444 	writel(0, q->csi_rx_base + CIO2_REG_MIPIBE_FORCE_RAW8);
445 	writel(0, base + CIO2_REG_PXM_SID2BID0(csi2bus));
446 
447 	writel(lanes, q->csi_rx_base + CIO2_REG_CSIRX_NOF_ENABLED_LANES);
448 	writel(CIO2_CGC_PRIM_TGE |
449 	       CIO2_CGC_SIDE_TGE |
450 	       CIO2_CGC_XOSC_TGE |
451 	       CIO2_CGC_D3I3_TGE |
452 	       CIO2_CGC_CSI2_INTERFRAME_TGE |
453 	       CIO2_CGC_CSI2_PORT_DCGE |
454 	       CIO2_CGC_SIDE_DCGE |
455 	       CIO2_CGC_PRIM_DCGE |
456 	       CIO2_CGC_ROSC_DCGE |
457 	       CIO2_CGC_XOSC_DCGE |
458 	       CIO2_CGC_CLKGATE_HOLDOFF << CIO2_CGC_CLKGATE_HOLDOFF_SHIFT |
459 	       CIO2_CGC_CSI_CLKGATE_HOLDOFF
460 	       << CIO2_CGC_CSI_CLKGATE_HOLDOFF_SHIFT, base + CIO2_REG_CGC);
461 	writel(CIO2_LTRCTRL_LTRDYNEN, base + CIO2_REG_LTRCTRL);
462 	writel(CIO2_LTRVAL0_VAL << CIO2_LTRVAL02_VAL_SHIFT |
463 	       CIO2_LTRVAL0_SCALE << CIO2_LTRVAL02_SCALE_SHIFT |
464 	       CIO2_LTRVAL1_VAL << CIO2_LTRVAL13_VAL_SHIFT |
465 	       CIO2_LTRVAL1_SCALE << CIO2_LTRVAL13_SCALE_SHIFT,
466 	       base + CIO2_REG_LTRVAL01);
467 	writel(CIO2_LTRVAL2_VAL << CIO2_LTRVAL02_VAL_SHIFT |
468 	       CIO2_LTRVAL2_SCALE << CIO2_LTRVAL02_SCALE_SHIFT |
469 	       CIO2_LTRVAL3_VAL << CIO2_LTRVAL13_VAL_SHIFT |
470 	       CIO2_LTRVAL3_SCALE << CIO2_LTRVAL13_SCALE_SHIFT,
471 	       base + CIO2_REG_LTRVAL23);
472 
473 	for (i = 0; i < CIO2_NUM_DMA_CHAN; i++) {
474 		writel(0, base + CIO2_REG_CDMABA(i));
475 		writel(0, base + CIO2_REG_CDMAC0(i));
476 		writel(0, base + CIO2_REG_CDMAC1(i));
477 	}
478 
479 	/* Enable DMA */
480 	writel(q->fbpt_bus_addr >> PAGE_SHIFT,
481 	       base + CIO2_REG_CDMABA(CIO2_DMA_CHAN));
482 
483 	writel(num_buffers1 << CIO2_CDMAC0_FBPT_LEN_SHIFT |
484 	       FBPT_WIDTH << CIO2_CDMAC0_FBPT_WIDTH_SHIFT |
485 	       CIO2_CDMAC0_DMA_INTR_ON_FE |
486 	       CIO2_CDMAC0_FBPT_UPDATE_FIFO_FULL |
487 	       CIO2_CDMAC0_DMA_EN |
488 	       CIO2_CDMAC0_DMA_INTR_ON_FS |
489 	       CIO2_CDMAC0_DMA_HALTED, base + CIO2_REG_CDMAC0(CIO2_DMA_CHAN));
490 
491 	writel(1 << CIO2_CDMAC1_LINENUMUPDATE_SHIFT,
492 	       base + CIO2_REG_CDMAC1(CIO2_DMA_CHAN));
493 
494 	writel(0, base + CIO2_REG_PBM_FOPN_ABORT);
495 
496 	writel(CIO2_PXM_FRF_CFG_CRC_TH << CIO2_PXM_FRF_CFG_CRC_TH_SHIFT |
497 	       CIO2_PXM_FRF_CFG_MSK_ECC_DPHY_NR |
498 	       CIO2_PXM_FRF_CFG_MSK_ECC_RE |
499 	       CIO2_PXM_FRF_CFG_MSK_ECC_DPHY_NE,
500 	       base + CIO2_REG_PXM_FRF_CFG(q->csi2.port));
501 
502 	/* Clear interrupts */
503 	writel(CIO2_IRQCTRL_MASK, q->csi_rx_base + CIO2_REG_IRQCTRL_CLEAR);
504 	writel(~0, base + CIO2_REG_INT_STS_EXT_OE);
505 	writel(~0, base + CIO2_REG_INT_STS_EXT_IE);
506 	writel(~0, base + CIO2_REG_INT_STS);
507 
508 	/* Enable devices, starting from the last device in the pipe */
509 	writel(1, q->csi_rx_base + CIO2_REG_MIPIBE_ENABLE);
510 	writel(1, q->csi_rx_base + CIO2_REG_CSIRX_ENABLE);
511 
512 	return 0;
513 }
514 
515 static void cio2_hw_exit(struct cio2_device *cio2, struct cio2_queue *q)
516 {
517 	void __iomem *base = cio2->base;
518 	unsigned int i, maxloops = 1000;
519 
520 	/* Disable CSI receiver and MIPI backend devices */
521 	writel(0, q->csi_rx_base + CIO2_REG_IRQCTRL_MASK);
522 	writel(0, q->csi_rx_base + CIO2_REG_IRQCTRL_ENABLE);
523 	writel(0, q->csi_rx_base + CIO2_REG_CSIRX_ENABLE);
524 	writel(0, q->csi_rx_base + CIO2_REG_MIPIBE_ENABLE);
525 
526 	/* Halt DMA */
527 	writel(0, base + CIO2_REG_CDMAC0(CIO2_DMA_CHAN));
528 	do {
529 		if (readl(base + CIO2_REG_CDMAC0(CIO2_DMA_CHAN)) &
530 		    CIO2_CDMAC0_DMA_HALTED)
531 			break;
532 		usleep_range(1000, 2000);
533 	} while (--maxloops);
534 	if (!maxloops)
535 		dev_err(&cio2->pci_dev->dev,
536 			"DMA %i can not be halted\n", CIO2_DMA_CHAN);
537 
538 	for (i = 0; i < CIO2_NUM_PORTS; i++) {
539 		writel(readl(base + CIO2_REG_PXM_FRF_CFG(i)) |
540 		       CIO2_PXM_FRF_CFG_ABORT, base + CIO2_REG_PXM_FRF_CFG(i));
541 		writel(readl(base + CIO2_REG_PBM_FOPN_ABORT) |
542 		       CIO2_PBM_FOPN_ABORT(i), base + CIO2_REG_PBM_FOPN_ABORT);
543 	}
544 }
545 
546 static void cio2_buffer_done(struct cio2_device *cio2, unsigned int dma_chan)
547 {
548 	struct device *dev = &cio2->pci_dev->dev;
549 	struct cio2_queue *q = cio2->cur_queue;
550 	int buffers_found = 0;
551 	u64 ns = ktime_get_ns();
552 
553 	if (dma_chan >= CIO2_QUEUES) {
554 		dev_err(dev, "bad DMA channel %i\n", dma_chan);
555 		return;
556 	}
557 
558 	/* Find out which buffer(s) are ready */
559 	do {
560 		struct cio2_fbpt_entry *const entry =
561 			&q->fbpt[q->bufs_first * CIO2_MAX_LOPS];
562 		struct cio2_buffer *b;
563 
564 		if (entry->first_entry.ctrl & CIO2_FBPT_CTRL_VALID)
565 			break;
566 
567 		b = q->bufs[q->bufs_first];
568 		if (b) {
569 			unsigned int bytes = entry[1].second_entry.num_of_bytes;
570 
571 			q->bufs[q->bufs_first] = NULL;
572 			atomic_dec(&q->bufs_queued);
573 			dev_dbg(&cio2->pci_dev->dev,
574 				"buffer %i done\n", b->vbb.vb2_buf.index);
575 
576 			b->vbb.vb2_buf.timestamp = ns;
577 			b->vbb.field = V4L2_FIELD_NONE;
578 			b->vbb.sequence = atomic_read(&q->frame_sequence);
579 			if (b->vbb.vb2_buf.planes[0].length != bytes)
580 				dev_warn(dev, "buffer length is %d received %d\n",
581 					 b->vbb.vb2_buf.planes[0].length,
582 					 bytes);
583 			vb2_buffer_done(&b->vbb.vb2_buf, VB2_BUF_STATE_DONE);
584 		}
585 		atomic_inc(&q->frame_sequence);
586 		cio2_fbpt_entry_init_dummy(cio2, entry);
587 		q->bufs_first = (q->bufs_first + 1) % CIO2_MAX_BUFFERS;
588 		buffers_found++;
589 	} while (1);
590 
591 	if (buffers_found == 0)
592 		dev_warn(&cio2->pci_dev->dev,
593 			 "no ready buffers found on DMA channel %u\n",
594 			 dma_chan);
595 }
596 
597 static void cio2_queue_event_sof(struct cio2_device *cio2, struct cio2_queue *q)
598 {
599 	/*
600 	 * For the user space camera control algorithms it is essential
601 	 * to know when the reception of a frame has begun. That's often
602 	 * the best timing information to get from the hardware.
603 	 */
604 	struct v4l2_event event = {
605 		.type = V4L2_EVENT_FRAME_SYNC,
606 		.u.frame_sync.frame_sequence = atomic_read(&q->frame_sequence),
607 	};
608 
609 	v4l2_event_queue(q->subdev.devnode, &event);
610 }
611 
612 static const char *const cio2_irq_errs[] = {
613 	"single packet header error corrected",
614 	"multiple packet header errors detected",
615 	"payload checksum (CRC) error",
616 	"fifo overflow",
617 	"reserved short packet data type detected",
618 	"reserved long packet data type detected",
619 	"incomplete long packet detected",
620 	"frame sync error",
621 	"line sync error",
622 	"DPHY start of transmission error",
623 	"DPHY synchronization error",
624 	"escape mode error",
625 	"escape mode trigger event",
626 	"escape mode ultra-low power state for data lane(s)",
627 	"escape mode ultra-low power state exit for clock lane",
628 	"inter-frame short packet discarded",
629 	"inter-frame long packet discarded",
630 	"non-matching Long Packet stalled",
631 };
632 
633 static const char *const cio2_port_errs[] = {
634 	"ECC recoverable",
635 	"DPHY not recoverable",
636 	"ECC not recoverable",
637 	"CRC error",
638 	"INTERFRAMEDATA",
639 	"PKT2SHORT",
640 	"PKT2LONG",
641 };
642 
643 static void cio2_irq_handle_once(struct cio2_device *cio2, u32 int_status)
644 {
645 	void __iomem *const base = cio2->base;
646 	struct device *dev = &cio2->pci_dev->dev;
647 
648 	if (int_status & CIO2_INT_IOOE) {
649 		/*
650 		 * Interrupt on Output Error:
651 		 * 1) SRAM is full and FS received, or
652 		 * 2) An invalid bit detected by DMA.
653 		 */
654 		u32 oe_status, oe_clear;
655 
656 		oe_clear = readl(base + CIO2_REG_INT_STS_EXT_OE);
657 		oe_status = oe_clear;
658 
659 		if (oe_status & CIO2_INT_EXT_OE_DMAOE_MASK) {
660 			dev_err(dev, "DMA output error: 0x%x\n",
661 				(oe_status & CIO2_INT_EXT_OE_DMAOE_MASK)
662 				>> CIO2_INT_EXT_OE_DMAOE_SHIFT);
663 			oe_status &= ~CIO2_INT_EXT_OE_DMAOE_MASK;
664 		}
665 		if (oe_status & CIO2_INT_EXT_OE_OES_MASK) {
666 			dev_err(dev, "DMA output error on CSI2 buses: 0x%x\n",
667 				(oe_status & CIO2_INT_EXT_OE_OES_MASK)
668 				>> CIO2_INT_EXT_OE_OES_SHIFT);
669 			oe_status &= ~CIO2_INT_EXT_OE_OES_MASK;
670 		}
671 		writel(oe_clear, base + CIO2_REG_INT_STS_EXT_OE);
672 		if (oe_status)
673 			dev_warn(dev, "unknown interrupt 0x%x on OE\n",
674 				 oe_status);
675 		int_status &= ~CIO2_INT_IOOE;
676 	}
677 
678 	if (int_status & CIO2_INT_IOC_MASK) {
679 		/* DMA IO done -- frame ready */
680 		u32 clr = 0;
681 		unsigned int d;
682 
683 		for (d = 0; d < CIO2_NUM_DMA_CHAN; d++)
684 			if (int_status & CIO2_INT_IOC(d)) {
685 				clr |= CIO2_INT_IOC(d);
686 				cio2_buffer_done(cio2, d);
687 			}
688 		int_status &= ~clr;
689 	}
690 
691 	if (int_status & CIO2_INT_IOS_IOLN_MASK) {
692 		/* DMA IO starts or reached specified line */
693 		u32 clr = 0;
694 		unsigned int d;
695 
696 		for (d = 0; d < CIO2_NUM_DMA_CHAN; d++)
697 			if (int_status & CIO2_INT_IOS_IOLN(d)) {
698 				clr |= CIO2_INT_IOS_IOLN(d);
699 				if (d == CIO2_DMA_CHAN)
700 					cio2_queue_event_sof(cio2,
701 							     cio2->cur_queue);
702 			}
703 		int_status &= ~clr;
704 	}
705 
706 	if (int_status & (CIO2_INT_IOIE | CIO2_INT_IOIRQ)) {
707 		/* CSI2 receiver (error) interrupt */
708 		u32 ie_status, ie_clear;
709 		unsigned int port;
710 
711 		ie_clear = readl(base + CIO2_REG_INT_STS_EXT_IE);
712 		ie_status = ie_clear;
713 
714 		for (port = 0; port < CIO2_NUM_PORTS; port++) {
715 			u32 port_status = (ie_status >> (port * 8)) & 0xff;
716 			u32 err_mask = BIT_MASK(ARRAY_SIZE(cio2_port_errs)) - 1;
717 			void __iomem *const csi_rx_base =
718 						base + CIO2_REG_PIPE_BASE(port);
719 			unsigned int i;
720 
721 			while (port_status & err_mask) {
722 				i = ffs(port_status) - 1;
723 				dev_err(dev, "port %i error %s\n",
724 					port, cio2_port_errs[i]);
725 				ie_status &= ~BIT(port * 8 + i);
726 				port_status &= ~BIT(i);
727 			}
728 
729 			if (ie_status & CIO2_INT_EXT_IE_IRQ(port)) {
730 				u32 csi2_status, csi2_clear;
731 
732 				csi2_status = readl(csi_rx_base +
733 						CIO2_REG_IRQCTRL_STATUS);
734 				csi2_clear = csi2_status;
735 				err_mask =
736 					BIT_MASK(ARRAY_SIZE(cio2_irq_errs)) - 1;
737 
738 				while (csi2_status & err_mask) {
739 					i = ffs(csi2_status) - 1;
740 					dev_err(dev,
741 						"CSI-2 receiver port %i: %s\n",
742 							port, cio2_irq_errs[i]);
743 					csi2_status &= ~BIT(i);
744 				}
745 
746 				writel(csi2_clear,
747 				       csi_rx_base + CIO2_REG_IRQCTRL_CLEAR);
748 				if (csi2_status)
749 					dev_warn(dev,
750 						 "unknown CSI2 error 0x%x on port %i\n",
751 						 csi2_status, port);
752 
753 				ie_status &= ~CIO2_INT_EXT_IE_IRQ(port);
754 			}
755 		}
756 
757 		writel(ie_clear, base + CIO2_REG_INT_STS_EXT_IE);
758 		if (ie_status)
759 			dev_warn(dev, "unknown interrupt 0x%x on IE\n",
760 				 ie_status);
761 
762 		int_status &= ~(CIO2_INT_IOIE | CIO2_INT_IOIRQ);
763 	}
764 
765 	if (int_status)
766 		dev_warn(dev, "unknown interrupt 0x%x on INT\n", int_status);
767 }
768 
769 static irqreturn_t cio2_irq(int irq, void *cio2_ptr)
770 {
771 	struct cio2_device *cio2 = cio2_ptr;
772 	void __iomem *const base = cio2->base;
773 	struct device *dev = &cio2->pci_dev->dev;
774 	u32 int_status;
775 
776 	int_status = readl(base + CIO2_REG_INT_STS);
777 	dev_dbg(dev, "isr enter - interrupt status 0x%x\n", int_status);
778 	if (!int_status)
779 		return IRQ_NONE;
780 
781 	do {
782 		writel(int_status, base + CIO2_REG_INT_STS);
783 		cio2_irq_handle_once(cio2, int_status);
784 		int_status = readl(base + CIO2_REG_INT_STS);
785 		if (int_status)
786 			dev_dbg(dev, "pending status 0x%x\n", int_status);
787 	} while (int_status);
788 
789 	return IRQ_HANDLED;
790 }
791 
792 /**************** Videobuf2 interface ****************/
793 
794 static void cio2_vb2_return_all_buffers(struct cio2_queue *q,
795 					enum vb2_buffer_state state)
796 {
797 	unsigned int i;
798 
799 	for (i = 0; i < CIO2_MAX_BUFFERS; i++) {
800 		if (q->bufs[i]) {
801 			atomic_dec(&q->bufs_queued);
802 			vb2_buffer_done(&q->bufs[i]->vbb.vb2_buf,
803 					state);
804 		}
805 	}
806 }
807 
808 static int cio2_vb2_queue_setup(struct vb2_queue *vq,
809 				unsigned int *num_buffers,
810 				unsigned int *num_planes,
811 				unsigned int sizes[],
812 				struct device *alloc_devs[])
813 {
814 	struct cio2_device *cio2 = vb2_get_drv_priv(vq);
815 	struct cio2_queue *q = vb2q_to_cio2_queue(vq);
816 	unsigned int i;
817 
818 	*num_planes = q->format.num_planes;
819 
820 	for (i = 0; i < *num_planes; ++i) {
821 		sizes[i] = q->format.plane_fmt[i].sizeimage;
822 		alloc_devs[i] = &cio2->pci_dev->dev;
823 	}
824 
825 	*num_buffers = clamp_val(*num_buffers, 1, CIO2_MAX_BUFFERS);
826 
827 	/* Initialize buffer queue */
828 	for (i = 0; i < CIO2_MAX_BUFFERS; i++) {
829 		q->bufs[i] = NULL;
830 		cio2_fbpt_entry_init_dummy(cio2, &q->fbpt[i * CIO2_MAX_LOPS]);
831 	}
832 	atomic_set(&q->bufs_queued, 0);
833 	q->bufs_first = 0;
834 	q->bufs_next = 0;
835 
836 	return 0;
837 }
838 
839 /* Called after each buffer is allocated */
840 static int cio2_vb2_buf_init(struct vb2_buffer *vb)
841 {
842 	struct cio2_device *cio2 = vb2_get_drv_priv(vb->vb2_queue);
843 	struct device *dev = &cio2->pci_dev->dev;
844 	struct cio2_buffer *b =
845 		container_of(vb, struct cio2_buffer, vbb.vb2_buf);
846 	static const unsigned int entries_per_page =
847 		CIO2_PAGE_SIZE / sizeof(u32);
848 	unsigned int pages = DIV_ROUND_UP(vb->planes[0].length, CIO2_PAGE_SIZE);
849 	unsigned int lops = DIV_ROUND_UP(pages + 1, entries_per_page);
850 	struct sg_table *sg;
851 	struct sg_page_iter sg_iter;
852 	int i, j;
853 
854 	if (lops <= 0 || lops > CIO2_MAX_LOPS) {
855 		dev_err(dev, "%s: bad buffer size (%i)\n", __func__,
856 			vb->planes[0].length);
857 		return -ENOSPC;		/* Should never happen */
858 	}
859 
860 	memset(b->lop, 0, sizeof(b->lop));
861 	/* Allocate LOP table */
862 	for (i = 0; i < lops; i++) {
863 		b->lop[i] = dma_alloc_coherent(dev, CIO2_PAGE_SIZE,
864 					       &b->lop_bus_addr[i], GFP_KERNEL);
865 		if (!b->lop[i])
866 			goto fail;
867 	}
868 
869 	/* Fill LOP */
870 	sg = vb2_dma_sg_plane_desc(vb, 0);
871 	if (!sg)
872 		return -ENOMEM;
873 
874 	if (sg->nents && sg->sgl)
875 		b->offset = sg->sgl->offset;
876 
877 	i = j = 0;
878 	for_each_sg_page(sg->sgl, &sg_iter, sg->nents, 0) {
879 		if (!pages--)
880 			break;
881 		b->lop[i][j] = sg_page_iter_dma_address(&sg_iter) >> PAGE_SHIFT;
882 		j++;
883 		if (j == entries_per_page) {
884 			i++;
885 			j = 0;
886 		}
887 	}
888 
889 	b->lop[i][j] = cio2->dummy_page_bus_addr >> PAGE_SHIFT;
890 	return 0;
891 fail:
892 	for (i--; i >= 0; i--)
893 		dma_free_coherent(dev, CIO2_PAGE_SIZE,
894 				  b->lop[i], b->lop_bus_addr[i]);
895 	return -ENOMEM;
896 }
897 
898 /* Transfer buffer ownership to cio2 */
899 static void cio2_vb2_buf_queue(struct vb2_buffer *vb)
900 {
901 	struct cio2_device *cio2 = vb2_get_drv_priv(vb->vb2_queue);
902 	struct cio2_queue *q =
903 		container_of(vb->vb2_queue, struct cio2_queue, vbq);
904 	struct cio2_buffer *b =
905 		container_of(vb, struct cio2_buffer, vbb.vb2_buf);
906 	struct cio2_fbpt_entry *entry;
907 	unsigned long flags;
908 	unsigned int i, j, next = q->bufs_next;
909 	int bufs_queued = atomic_inc_return(&q->bufs_queued);
910 	u32 fbpt_rp;
911 
912 	dev_dbg(&cio2->pci_dev->dev, "queue buffer %d\n", vb->index);
913 
914 	/*
915 	 * This code queues the buffer to the CIO2 DMA engine, which starts
916 	 * running once streaming has started. It is possible that this code
917 	 * gets pre-empted due to increased CPU load. Upon this, the driver
918 	 * does not get an opportunity to queue new buffers to the CIO2 DMA
919 	 * engine. When the DMA engine encounters an FBPT entry without the
920 	 * VALID bit set, the DMA engine halts, which requires a restart of
921 	 * the DMA engine and sensor, to continue streaming.
922 	 * This is not desired and is highly unlikely given that there are
923 	 * 32 FBPT entries that the DMA engine needs to process, to run into
924 	 * an FBPT entry, without the VALID bit set. We try to mitigate this
925 	 * by disabling interrupts for the duration of this queueing.
926 	 */
927 	local_irq_save(flags);
928 
929 	fbpt_rp = (readl(cio2->base + CIO2_REG_CDMARI(CIO2_DMA_CHAN))
930 		   >> CIO2_CDMARI_FBPT_RP_SHIFT)
931 		   & CIO2_CDMARI_FBPT_RP_MASK;
932 
933 	/*
934 	 * fbpt_rp is the fbpt entry that the dma is currently working
935 	 * on, but since it could jump to next entry at any time,
936 	 * assume that we might already be there.
937 	 */
938 	fbpt_rp = (fbpt_rp + 1) % CIO2_MAX_BUFFERS;
939 
940 	if (bufs_queued <= 1 || fbpt_rp == next)
941 		/* Buffers were drained */
942 		next = (fbpt_rp + 1) % CIO2_MAX_BUFFERS;
943 
944 	for (i = 0; i < CIO2_MAX_BUFFERS; i++) {
945 		/*
946 		 * We have allocated CIO2_MAX_BUFFERS circularly for the
947 		 * hw, the user has requested N buffer queue. The driver
948 		 * ensures N <= CIO2_MAX_BUFFERS and guarantees that whenever
949 		 * user queues a buffer, there necessarily is a free buffer.
950 		 */
951 		if (!q->bufs[next]) {
952 			q->bufs[next] = b;
953 			entry = &q->fbpt[next * CIO2_MAX_LOPS];
954 			cio2_fbpt_entry_init_buf(cio2, b, entry);
955 			local_irq_restore(flags);
956 			q->bufs_next = (next + 1) % CIO2_MAX_BUFFERS;
957 			for (j = 0; j < vb->num_planes; j++)
958 				vb2_set_plane_payload(vb, j,
959 					q->format.plane_fmt[j].sizeimage);
960 			return;
961 		}
962 
963 		dev_dbg(&cio2->pci_dev->dev, "entry %i was full!\n", next);
964 		next = (next + 1) % CIO2_MAX_BUFFERS;
965 	}
966 
967 	local_irq_restore(flags);
968 	dev_err(&cio2->pci_dev->dev, "error: all cio2 entries were full!\n");
969 	atomic_dec(&q->bufs_queued);
970 	vb2_buffer_done(vb, VB2_BUF_STATE_ERROR);
971 }
972 
973 /* Called when each buffer is freed */
974 static void cio2_vb2_buf_cleanup(struct vb2_buffer *vb)
975 {
976 	struct cio2_device *cio2 = vb2_get_drv_priv(vb->vb2_queue);
977 	struct cio2_buffer *b =
978 		container_of(vb, struct cio2_buffer, vbb.vb2_buf);
979 	unsigned int i;
980 
981 	/* Free LOP table */
982 	for (i = 0; i < CIO2_MAX_LOPS; i++) {
983 		if (b->lop[i])
984 			dma_free_coherent(&cio2->pci_dev->dev, CIO2_PAGE_SIZE,
985 					  b->lop[i], b->lop_bus_addr[i]);
986 	}
987 }
988 
989 static int cio2_vb2_start_streaming(struct vb2_queue *vq, unsigned int count)
990 {
991 	struct cio2_queue *q = vb2q_to_cio2_queue(vq);
992 	struct cio2_device *cio2 = vb2_get_drv_priv(vq);
993 	int r;
994 
995 	cio2->cur_queue = q;
996 	atomic_set(&q->frame_sequence, 0);
997 
998 	r = pm_runtime_get_sync(&cio2->pci_dev->dev);
999 	if (r < 0) {
1000 		dev_info(&cio2->pci_dev->dev, "failed to set power %d\n", r);
1001 		pm_runtime_put_noidle(&cio2->pci_dev->dev);
1002 		return r;
1003 	}
1004 
1005 	r = media_pipeline_start(&q->vdev.entity, &q->pipe);
1006 	if (r)
1007 		goto fail_pipeline;
1008 
1009 	r = cio2_hw_init(cio2, q);
1010 	if (r)
1011 		goto fail_hw;
1012 
1013 	/* Start streaming on sensor */
1014 	r = v4l2_subdev_call(q->sensor, video, s_stream, 1);
1015 	if (r)
1016 		goto fail_csi2_subdev;
1017 
1018 	cio2->streaming = true;
1019 
1020 	return 0;
1021 
1022 fail_csi2_subdev:
1023 	cio2_hw_exit(cio2, q);
1024 fail_hw:
1025 	media_pipeline_stop(&q->vdev.entity);
1026 fail_pipeline:
1027 	dev_dbg(&cio2->pci_dev->dev, "failed to start streaming (%d)\n", r);
1028 	cio2_vb2_return_all_buffers(q, VB2_BUF_STATE_QUEUED);
1029 	pm_runtime_put(&cio2->pci_dev->dev);
1030 
1031 	return r;
1032 }
1033 
1034 static void cio2_vb2_stop_streaming(struct vb2_queue *vq)
1035 {
1036 	struct cio2_queue *q = vb2q_to_cio2_queue(vq);
1037 	struct cio2_device *cio2 = vb2_get_drv_priv(vq);
1038 
1039 	if (v4l2_subdev_call(q->sensor, video, s_stream, 0))
1040 		dev_err(&cio2->pci_dev->dev,
1041 			"failed to stop sensor streaming\n");
1042 
1043 	cio2_hw_exit(cio2, q);
1044 	synchronize_irq(cio2->pci_dev->irq);
1045 	cio2_vb2_return_all_buffers(q, VB2_BUF_STATE_ERROR);
1046 	media_pipeline_stop(&q->vdev.entity);
1047 	pm_runtime_put(&cio2->pci_dev->dev);
1048 	cio2->streaming = false;
1049 }
1050 
1051 static const struct vb2_ops cio2_vb2_ops = {
1052 	.buf_init = cio2_vb2_buf_init,
1053 	.buf_queue = cio2_vb2_buf_queue,
1054 	.buf_cleanup = cio2_vb2_buf_cleanup,
1055 	.queue_setup = cio2_vb2_queue_setup,
1056 	.start_streaming = cio2_vb2_start_streaming,
1057 	.stop_streaming = cio2_vb2_stop_streaming,
1058 	.wait_prepare = vb2_ops_wait_prepare,
1059 	.wait_finish = vb2_ops_wait_finish,
1060 };
1061 
1062 /**************** V4L2 interface ****************/
1063 
1064 static int cio2_v4l2_querycap(struct file *file, void *fh,
1065 			      struct v4l2_capability *cap)
1066 {
1067 	struct cio2_device *cio2 = video_drvdata(file);
1068 
1069 	strlcpy(cap->driver, CIO2_NAME, sizeof(cap->driver));
1070 	strlcpy(cap->card, CIO2_DEVICE_NAME, sizeof(cap->card));
1071 	snprintf(cap->bus_info, sizeof(cap->bus_info),
1072 		 "PCI:%s", pci_name(cio2->pci_dev));
1073 
1074 	return 0;
1075 }
1076 
1077 static int cio2_v4l2_enum_fmt(struct file *file, void *fh,
1078 			      struct v4l2_fmtdesc *f)
1079 {
1080 	if (f->index >= ARRAY_SIZE(formats))
1081 		return -EINVAL;
1082 
1083 	f->pixelformat = formats[f->index].fourcc;
1084 
1085 	return 0;
1086 }
1087 
1088 /* The format is validated in cio2_video_link_validate() */
1089 static int cio2_v4l2_g_fmt(struct file *file, void *fh, struct v4l2_format *f)
1090 {
1091 	struct cio2_queue *q = file_to_cio2_queue(file);
1092 
1093 	f->fmt.pix_mp = q->format;
1094 
1095 	return 0;
1096 }
1097 
1098 static int cio2_v4l2_try_fmt(struct file *file, void *fh, struct v4l2_format *f)
1099 {
1100 	const struct ipu3_cio2_fmt *fmt;
1101 	struct v4l2_pix_format_mplane *mpix = &f->fmt.pix_mp;
1102 
1103 	fmt = cio2_find_format(&mpix->pixelformat, NULL);
1104 	if (!fmt)
1105 		fmt = &formats[0];
1106 
1107 	/* Only supports up to 4224x3136 */
1108 	if (mpix->width > CIO2_IMAGE_MAX_WIDTH)
1109 		mpix->width = CIO2_IMAGE_MAX_WIDTH;
1110 	if (mpix->height > CIO2_IMAGE_MAX_LENGTH)
1111 		mpix->height = CIO2_IMAGE_MAX_LENGTH;
1112 
1113 	mpix->num_planes = 1;
1114 	mpix->pixelformat = fmt->fourcc;
1115 	mpix->colorspace = V4L2_COLORSPACE_RAW;
1116 	mpix->field = V4L2_FIELD_NONE;
1117 	memset(mpix->reserved, 0, sizeof(mpix->reserved));
1118 	mpix->plane_fmt[0].bytesperline = cio2_bytesperline(mpix->width);
1119 	mpix->plane_fmt[0].sizeimage = mpix->plane_fmt[0].bytesperline *
1120 							mpix->height;
1121 	memset(mpix->plane_fmt[0].reserved, 0,
1122 	       sizeof(mpix->plane_fmt[0].reserved));
1123 
1124 	/* use default */
1125 	mpix->ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
1126 	mpix->quantization = V4L2_QUANTIZATION_DEFAULT;
1127 	mpix->xfer_func = V4L2_XFER_FUNC_DEFAULT;
1128 
1129 	return 0;
1130 }
1131 
1132 static int cio2_v4l2_s_fmt(struct file *file, void *fh, struct v4l2_format *f)
1133 {
1134 	struct cio2_queue *q = file_to_cio2_queue(file);
1135 
1136 	cio2_v4l2_try_fmt(file, fh, f);
1137 	q->format = f->fmt.pix_mp;
1138 
1139 	return 0;
1140 }
1141 
1142 static int
1143 cio2_video_enum_input(struct file *file, void *fh, struct v4l2_input *input)
1144 {
1145 	if (input->index > 0)
1146 		return -EINVAL;
1147 
1148 	strlcpy(input->name, "camera", sizeof(input->name));
1149 	input->type = V4L2_INPUT_TYPE_CAMERA;
1150 
1151 	return 0;
1152 }
1153 
1154 static int
1155 cio2_video_g_input(struct file *file, void *fh, unsigned int *input)
1156 {
1157 	*input = 0;
1158 
1159 	return 0;
1160 }
1161 
1162 static int
1163 cio2_video_s_input(struct file *file, void *fh, unsigned int input)
1164 {
1165 	return input == 0 ? 0 : -EINVAL;
1166 }
1167 
1168 static const struct v4l2_file_operations cio2_v4l2_fops = {
1169 	.owner = THIS_MODULE,
1170 	.unlocked_ioctl = video_ioctl2,
1171 	.open = v4l2_fh_open,
1172 	.release = vb2_fop_release,
1173 	.poll = vb2_fop_poll,
1174 	.mmap = vb2_fop_mmap,
1175 };
1176 
1177 static const struct v4l2_ioctl_ops cio2_v4l2_ioctl_ops = {
1178 	.vidioc_querycap = cio2_v4l2_querycap,
1179 	.vidioc_enum_fmt_vid_cap_mplane = cio2_v4l2_enum_fmt,
1180 	.vidioc_g_fmt_vid_cap_mplane = cio2_v4l2_g_fmt,
1181 	.vidioc_s_fmt_vid_cap_mplane = cio2_v4l2_s_fmt,
1182 	.vidioc_try_fmt_vid_cap_mplane = cio2_v4l2_try_fmt,
1183 	.vidioc_reqbufs = vb2_ioctl_reqbufs,
1184 	.vidioc_create_bufs = vb2_ioctl_create_bufs,
1185 	.vidioc_prepare_buf = vb2_ioctl_prepare_buf,
1186 	.vidioc_querybuf = vb2_ioctl_querybuf,
1187 	.vidioc_qbuf = vb2_ioctl_qbuf,
1188 	.vidioc_dqbuf = vb2_ioctl_dqbuf,
1189 	.vidioc_streamon = vb2_ioctl_streamon,
1190 	.vidioc_streamoff = vb2_ioctl_streamoff,
1191 	.vidioc_expbuf = vb2_ioctl_expbuf,
1192 	.vidioc_enum_input = cio2_video_enum_input,
1193 	.vidioc_g_input	= cio2_video_g_input,
1194 	.vidioc_s_input	= cio2_video_s_input,
1195 };
1196 
1197 static int cio2_subdev_subscribe_event(struct v4l2_subdev *sd,
1198 				       struct v4l2_fh *fh,
1199 				       struct v4l2_event_subscription *sub)
1200 {
1201 	if (sub->type != V4L2_EVENT_FRAME_SYNC)
1202 		return -EINVAL;
1203 
1204 	/* Line number. For now only zero accepted. */
1205 	if (sub->id != 0)
1206 		return -EINVAL;
1207 
1208 	return v4l2_event_subscribe(fh, sub, 0, NULL);
1209 }
1210 
1211 static int cio2_subdev_open(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh)
1212 {
1213 	struct v4l2_mbus_framefmt *format;
1214 	const struct v4l2_mbus_framefmt fmt_default = {
1215 		.width = 1936,
1216 		.height = 1096,
1217 		.code = formats[0].mbus_code,
1218 		.field = V4L2_FIELD_NONE,
1219 		.colorspace = V4L2_COLORSPACE_RAW,
1220 		.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT,
1221 		.quantization = V4L2_QUANTIZATION_DEFAULT,
1222 		.xfer_func = V4L2_XFER_FUNC_DEFAULT,
1223 	};
1224 
1225 	/* Initialize try_fmt */
1226 	format = v4l2_subdev_get_try_format(sd, fh->pad, CIO2_PAD_SINK);
1227 	*format = fmt_default;
1228 
1229 	/* same as sink */
1230 	format = v4l2_subdev_get_try_format(sd, fh->pad, CIO2_PAD_SOURCE);
1231 	*format = fmt_default;
1232 
1233 	return 0;
1234 }
1235 
1236 /*
1237  * cio2_subdev_get_fmt - Handle get format by pads subdev method
1238  * @sd : pointer to v4l2 subdev structure
1239  * @cfg: V4L2 subdev pad config
1240  * @fmt: pointer to v4l2 subdev format structure
1241  * return -EINVAL or zero on success
1242  */
1243 static int cio2_subdev_get_fmt(struct v4l2_subdev *sd,
1244 			       struct v4l2_subdev_pad_config *cfg,
1245 			       struct v4l2_subdev_format *fmt)
1246 {
1247 	struct cio2_queue *q = container_of(sd, struct cio2_queue, subdev);
1248 	struct v4l2_subdev_format format;
1249 	int ret;
1250 
1251 	if (fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
1252 		fmt->format = *v4l2_subdev_get_try_format(sd, cfg, fmt->pad);
1253 		return 0;
1254 	}
1255 
1256 	if (fmt->pad == CIO2_PAD_SINK) {
1257 		format.which = V4L2_SUBDEV_FORMAT_ACTIVE;
1258 		ret = v4l2_subdev_call(sd, pad, get_fmt, NULL,
1259 				       &format);
1260 
1261 		if (ret)
1262 			return ret;
1263 		/* update colorspace etc */
1264 		q->subdev_fmt.colorspace = format.format.colorspace;
1265 		q->subdev_fmt.ycbcr_enc = format.format.ycbcr_enc;
1266 		q->subdev_fmt.quantization = format.format.quantization;
1267 		q->subdev_fmt.xfer_func = format.format.xfer_func;
1268 	}
1269 
1270 	fmt->format = q->subdev_fmt;
1271 
1272 	return 0;
1273 }
1274 
1275 /*
1276  * cio2_subdev_set_fmt - Handle set format by pads subdev method
1277  * @sd : pointer to v4l2 subdev structure
1278  * @cfg: V4L2 subdev pad config
1279  * @fmt: pointer to v4l2 subdev format structure
1280  * return -EINVAL or zero on success
1281  */
1282 static int cio2_subdev_set_fmt(struct v4l2_subdev *sd,
1283 			       struct v4l2_subdev_pad_config *cfg,
1284 			       struct v4l2_subdev_format *fmt)
1285 {
1286 	struct cio2_queue *q = container_of(sd, struct cio2_queue, subdev);
1287 
1288 	/*
1289 	 * Only allow setting sink pad format;
1290 	 * source always propagates from sink
1291 	 */
1292 	if (fmt->pad == CIO2_PAD_SOURCE)
1293 		return cio2_subdev_get_fmt(sd, cfg, fmt);
1294 
1295 	if (fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
1296 		*v4l2_subdev_get_try_format(sd, cfg, fmt->pad) = fmt->format;
1297 	} else {
1298 		/* It's the sink, allow changing frame size */
1299 		q->subdev_fmt.width = fmt->format.width;
1300 		q->subdev_fmt.height = fmt->format.height;
1301 		q->subdev_fmt.code = fmt->format.code;
1302 		fmt->format = q->subdev_fmt;
1303 	}
1304 
1305 	return 0;
1306 }
1307 
1308 static int cio2_subdev_enum_mbus_code(struct v4l2_subdev *sd,
1309 				      struct v4l2_subdev_pad_config *cfg,
1310 				      struct v4l2_subdev_mbus_code_enum *code)
1311 {
1312 	if (code->index >= ARRAY_SIZE(formats))
1313 		return -EINVAL;
1314 
1315 	code->code = formats[code->index].mbus_code;
1316 	return 0;
1317 }
1318 
1319 static int cio2_subdev_link_validate_get_format(struct media_pad *pad,
1320 						struct v4l2_subdev_format *fmt)
1321 {
1322 	if (is_media_entity_v4l2_subdev(pad->entity)) {
1323 		struct v4l2_subdev *sd =
1324 			media_entity_to_v4l2_subdev(pad->entity);
1325 
1326 		fmt->which = V4L2_SUBDEV_FORMAT_ACTIVE;
1327 		fmt->pad = pad->index;
1328 		return v4l2_subdev_call(sd, pad, get_fmt, NULL, fmt);
1329 	}
1330 
1331 	return -EINVAL;
1332 }
1333 
1334 static int cio2_video_link_validate(struct media_link *link)
1335 {
1336 	struct video_device *vd = container_of(link->sink->entity,
1337 						struct video_device, entity);
1338 	struct cio2_queue *q = container_of(vd, struct cio2_queue, vdev);
1339 	struct cio2_device *cio2 = video_get_drvdata(vd);
1340 	struct v4l2_subdev_format source_fmt;
1341 	int ret;
1342 
1343 	if (!media_entity_remote_pad(link->sink->entity->pads)) {
1344 		dev_info(&cio2->pci_dev->dev,
1345 			 "video node %s pad not connected\n", vd->name);
1346 		return -ENOTCONN;
1347 	}
1348 
1349 	ret = cio2_subdev_link_validate_get_format(link->source, &source_fmt);
1350 	if (ret < 0)
1351 		return 0;
1352 
1353 	if (source_fmt.format.width != q->format.width ||
1354 	    source_fmt.format.height != q->format.height) {
1355 		dev_err(&cio2->pci_dev->dev,
1356 			"Wrong width or height %ux%u (%ux%u expected)\n",
1357 			q->format.width, q->format.height,
1358 			source_fmt.format.width, source_fmt.format.height);
1359 		return -EINVAL;
1360 	}
1361 
1362 	if (!cio2_find_format(&q->format.pixelformat, &source_fmt.format.code))
1363 		return -EINVAL;
1364 
1365 	return 0;
1366 }
1367 
1368 static const struct v4l2_subdev_core_ops cio2_subdev_core_ops = {
1369 	.subscribe_event = cio2_subdev_subscribe_event,
1370 	.unsubscribe_event = v4l2_event_subdev_unsubscribe,
1371 };
1372 
1373 static const struct v4l2_subdev_internal_ops cio2_subdev_internal_ops = {
1374 	.open = cio2_subdev_open,
1375 };
1376 
1377 static const struct v4l2_subdev_pad_ops cio2_subdev_pad_ops = {
1378 	.link_validate = v4l2_subdev_link_validate_default,
1379 	.get_fmt = cio2_subdev_get_fmt,
1380 	.set_fmt = cio2_subdev_set_fmt,
1381 	.enum_mbus_code = cio2_subdev_enum_mbus_code,
1382 };
1383 
1384 static const struct v4l2_subdev_ops cio2_subdev_ops = {
1385 	.core = &cio2_subdev_core_ops,
1386 	.pad = &cio2_subdev_pad_ops,
1387 };
1388 
1389 /******* V4L2 sub-device asynchronous registration callbacks***********/
1390 
1391 struct sensor_async_subdev {
1392 	struct v4l2_async_subdev asd;
1393 	struct csi2_bus_info csi2;
1394 };
1395 
1396 /* The .bound() notifier callback when a match is found */
1397 static int cio2_notifier_bound(struct v4l2_async_notifier *notifier,
1398 			       struct v4l2_subdev *sd,
1399 			       struct v4l2_async_subdev *asd)
1400 {
1401 	struct cio2_device *cio2 = container_of(notifier,
1402 					struct cio2_device, notifier);
1403 	struct sensor_async_subdev *s_asd = container_of(asd,
1404 					struct sensor_async_subdev, asd);
1405 	struct cio2_queue *q;
1406 
1407 	if (cio2->queue[s_asd->csi2.port].sensor)
1408 		return -EBUSY;
1409 
1410 	q = &cio2->queue[s_asd->csi2.port];
1411 
1412 	q->csi2 = s_asd->csi2;
1413 	q->sensor = sd;
1414 	q->csi_rx_base = cio2->base + CIO2_REG_PIPE_BASE(q->csi2.port);
1415 
1416 	return 0;
1417 }
1418 
1419 /* The .unbind callback */
1420 static void cio2_notifier_unbind(struct v4l2_async_notifier *notifier,
1421 				 struct v4l2_subdev *sd,
1422 				 struct v4l2_async_subdev *asd)
1423 {
1424 	struct cio2_device *cio2 = container_of(notifier,
1425 						struct cio2_device, notifier);
1426 	struct sensor_async_subdev *s_asd = container_of(asd,
1427 					struct sensor_async_subdev, asd);
1428 
1429 	cio2->queue[s_asd->csi2.port].sensor = NULL;
1430 }
1431 
1432 /* .complete() is called after all subdevices have been located */
1433 static int cio2_notifier_complete(struct v4l2_async_notifier *notifier)
1434 {
1435 	struct cio2_device *cio2 = container_of(notifier, struct cio2_device,
1436 						notifier);
1437 	struct sensor_async_subdev *s_asd;
1438 	struct cio2_queue *q;
1439 	unsigned int i, pad;
1440 	int ret;
1441 
1442 	for (i = 0; i < notifier->num_subdevs; i++) {
1443 		s_asd = container_of(cio2->notifier.subdevs[i],
1444 				     struct sensor_async_subdev, asd);
1445 		q = &cio2->queue[s_asd->csi2.port];
1446 
1447 		for (pad = 0; pad < q->sensor->entity.num_pads; pad++)
1448 			if (q->sensor->entity.pads[pad].flags &
1449 						MEDIA_PAD_FL_SOURCE)
1450 				break;
1451 
1452 		if (pad == q->sensor->entity.num_pads) {
1453 			dev_err(&cio2->pci_dev->dev,
1454 				"failed to find src pad for %s\n",
1455 				q->sensor->name);
1456 			return -ENXIO;
1457 		}
1458 
1459 		ret = media_create_pad_link(
1460 				&q->sensor->entity, pad,
1461 				&q->subdev.entity, CIO2_PAD_SINK,
1462 				0);
1463 		if (ret) {
1464 			dev_err(&cio2->pci_dev->dev,
1465 				"failed to create link for %s\n",
1466 				cio2->queue[i].sensor->name);
1467 			return ret;
1468 		}
1469 	}
1470 
1471 	return v4l2_device_register_subdev_nodes(&cio2->v4l2_dev);
1472 }
1473 
1474 static const struct v4l2_async_notifier_operations cio2_async_ops = {
1475 	.bound = cio2_notifier_bound,
1476 	.unbind = cio2_notifier_unbind,
1477 	.complete = cio2_notifier_complete,
1478 };
1479 
1480 static int cio2_fwnode_parse(struct device *dev,
1481 			     struct v4l2_fwnode_endpoint *vep,
1482 			     struct v4l2_async_subdev *asd)
1483 {
1484 	struct sensor_async_subdev *s_asd =
1485 			container_of(asd, struct sensor_async_subdev, asd);
1486 
1487 	if (vep->bus_type != V4L2_MBUS_CSI2) {
1488 		dev_err(dev, "Only CSI2 bus type is currently supported\n");
1489 		return -EINVAL;
1490 	}
1491 
1492 	s_asd->csi2.port = vep->base.port;
1493 	s_asd->csi2.lanes = vep->bus.mipi_csi2.num_data_lanes;
1494 
1495 	return 0;
1496 }
1497 
1498 static int cio2_notifier_init(struct cio2_device *cio2)
1499 {
1500 	int ret;
1501 
1502 	ret = v4l2_async_notifier_parse_fwnode_endpoints(
1503 		&cio2->pci_dev->dev, &cio2->notifier,
1504 		sizeof(struct sensor_async_subdev),
1505 		cio2_fwnode_parse);
1506 	if (ret < 0)
1507 		return ret;
1508 
1509 	if (!cio2->notifier.num_subdevs)
1510 		return -ENODEV;	/* no endpoint */
1511 
1512 	cio2->notifier.ops = &cio2_async_ops;
1513 	ret = v4l2_async_notifier_register(&cio2->v4l2_dev, &cio2->notifier);
1514 	if (ret) {
1515 		dev_err(&cio2->pci_dev->dev,
1516 			"failed to register async notifier : %d\n", ret);
1517 		v4l2_async_notifier_cleanup(&cio2->notifier);
1518 	}
1519 
1520 	return ret;
1521 }
1522 
1523 static void cio2_notifier_exit(struct cio2_device *cio2)
1524 {
1525 	v4l2_async_notifier_unregister(&cio2->notifier);
1526 	v4l2_async_notifier_cleanup(&cio2->notifier);
1527 }
1528 
1529 /**************** Queue initialization ****************/
1530 static const struct media_entity_operations cio2_media_ops = {
1531 	.link_validate = v4l2_subdev_link_validate,
1532 };
1533 
1534 static const struct media_entity_operations cio2_video_entity_ops = {
1535 	.link_validate = cio2_video_link_validate,
1536 };
1537 
1538 static int cio2_queue_init(struct cio2_device *cio2, struct cio2_queue *q)
1539 {
1540 	static const u32 default_width = 1936;
1541 	static const u32 default_height = 1096;
1542 	const struct ipu3_cio2_fmt dflt_fmt = formats[0];
1543 
1544 	struct video_device *vdev = &q->vdev;
1545 	struct vb2_queue *vbq = &q->vbq;
1546 	struct v4l2_subdev *subdev = &q->subdev;
1547 	struct v4l2_mbus_framefmt *fmt;
1548 	int r;
1549 
1550 	/* Initialize miscellaneous variables */
1551 	mutex_init(&q->lock);
1552 
1553 	/* Initialize formats to default values */
1554 	fmt = &q->subdev_fmt;
1555 	fmt->width = default_width;
1556 	fmt->height = default_height;
1557 	fmt->code = dflt_fmt.mbus_code;
1558 	fmt->field = V4L2_FIELD_NONE;
1559 
1560 	q->format.width = default_width;
1561 	q->format.height = default_height;
1562 	q->format.pixelformat = dflt_fmt.fourcc;
1563 	q->format.colorspace = V4L2_COLORSPACE_RAW;
1564 	q->format.field = V4L2_FIELD_NONE;
1565 	q->format.num_planes = 1;
1566 	q->format.plane_fmt[0].bytesperline =
1567 				cio2_bytesperline(q->format.width);
1568 	q->format.plane_fmt[0].sizeimage = q->format.plane_fmt[0].bytesperline *
1569 						q->format.height;
1570 
1571 	/* Initialize fbpt */
1572 	r = cio2_fbpt_init(cio2, q);
1573 	if (r)
1574 		goto fail_fbpt;
1575 
1576 	/* Initialize media entities */
1577 	q->subdev_pads[CIO2_PAD_SINK].flags = MEDIA_PAD_FL_SINK |
1578 		MEDIA_PAD_FL_MUST_CONNECT;
1579 	q->subdev_pads[CIO2_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
1580 	subdev->entity.ops = &cio2_media_ops;
1581 	subdev->internal_ops = &cio2_subdev_internal_ops;
1582 	r = media_entity_pads_init(&subdev->entity, CIO2_PADS, q->subdev_pads);
1583 	if (r) {
1584 		dev_err(&cio2->pci_dev->dev,
1585 			"failed initialize subdev media entity (%d)\n", r);
1586 		goto fail_subdev_media_entity;
1587 	}
1588 
1589 	q->vdev_pad.flags = MEDIA_PAD_FL_SINK | MEDIA_PAD_FL_MUST_CONNECT;
1590 	vdev->entity.ops = &cio2_video_entity_ops;
1591 	r = media_entity_pads_init(&vdev->entity, 1, &q->vdev_pad);
1592 	if (r) {
1593 		dev_err(&cio2->pci_dev->dev,
1594 			"failed initialize videodev media entity (%d)\n", r);
1595 		goto fail_vdev_media_entity;
1596 	}
1597 
1598 	/* Initialize subdev */
1599 	v4l2_subdev_init(subdev, &cio2_subdev_ops);
1600 	subdev->flags = V4L2_SUBDEV_FL_HAS_DEVNODE | V4L2_SUBDEV_FL_HAS_EVENTS;
1601 	subdev->owner = THIS_MODULE;
1602 	snprintf(subdev->name, sizeof(subdev->name),
1603 		 CIO2_ENTITY_NAME " %td", q - cio2->queue);
1604 	v4l2_set_subdevdata(subdev, cio2);
1605 	r = v4l2_device_register_subdev(&cio2->v4l2_dev, subdev);
1606 	if (r) {
1607 		dev_err(&cio2->pci_dev->dev,
1608 			"failed initialize subdev (%d)\n", r);
1609 		goto fail_subdev;
1610 	}
1611 
1612 	/* Initialize vbq */
1613 	vbq->type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
1614 	vbq->io_modes = VB2_USERPTR | VB2_MMAP | VB2_DMABUF;
1615 	vbq->ops = &cio2_vb2_ops;
1616 	vbq->mem_ops = &vb2_dma_sg_memops;
1617 	vbq->buf_struct_size = sizeof(struct cio2_buffer);
1618 	vbq->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_MONOTONIC;
1619 	vbq->min_buffers_needed = 1;
1620 	vbq->drv_priv = cio2;
1621 	vbq->lock = &q->lock;
1622 	r = vb2_queue_init(vbq);
1623 	if (r) {
1624 		dev_err(&cio2->pci_dev->dev,
1625 			"failed to initialize videobuf2 queue (%d)\n", r);
1626 		goto fail_vbq;
1627 	}
1628 
1629 	/* Initialize vdev */
1630 	snprintf(vdev->name, sizeof(vdev->name),
1631 		 "%s %td", CIO2_NAME, q - cio2->queue);
1632 	vdev->release = video_device_release_empty;
1633 	vdev->fops = &cio2_v4l2_fops;
1634 	vdev->ioctl_ops = &cio2_v4l2_ioctl_ops;
1635 	vdev->lock = &cio2->lock;
1636 	vdev->v4l2_dev = &cio2->v4l2_dev;
1637 	vdev->queue = &q->vbq;
1638 	vdev->device_caps = V4L2_CAP_VIDEO_CAPTURE_MPLANE | V4L2_CAP_STREAMING;
1639 	video_set_drvdata(vdev, cio2);
1640 	r = video_register_device(vdev, VFL_TYPE_GRABBER, -1);
1641 	if (r) {
1642 		dev_err(&cio2->pci_dev->dev,
1643 			"failed to register video device (%d)\n", r);
1644 		goto fail_vdev;
1645 	}
1646 
1647 	/* Create link from CIO2 subdev to output node */
1648 	r = media_create_pad_link(
1649 		&subdev->entity, CIO2_PAD_SOURCE, &vdev->entity, 0,
1650 		MEDIA_LNK_FL_ENABLED | MEDIA_LNK_FL_IMMUTABLE);
1651 	if (r)
1652 		goto fail_link;
1653 
1654 	return 0;
1655 
1656 fail_link:
1657 	video_unregister_device(&q->vdev);
1658 fail_vdev:
1659 	vb2_queue_release(vbq);
1660 fail_vbq:
1661 	v4l2_device_unregister_subdev(subdev);
1662 fail_subdev:
1663 	media_entity_cleanup(&vdev->entity);
1664 fail_vdev_media_entity:
1665 	media_entity_cleanup(&subdev->entity);
1666 fail_subdev_media_entity:
1667 	cio2_fbpt_exit(q, &cio2->pci_dev->dev);
1668 fail_fbpt:
1669 	mutex_destroy(&q->lock);
1670 
1671 	return r;
1672 }
1673 
1674 static void cio2_queue_exit(struct cio2_device *cio2, struct cio2_queue *q)
1675 {
1676 	video_unregister_device(&q->vdev);
1677 	media_entity_cleanup(&q->vdev.entity);
1678 	vb2_queue_release(&q->vbq);
1679 	v4l2_device_unregister_subdev(&q->subdev);
1680 	media_entity_cleanup(&q->subdev.entity);
1681 	cio2_fbpt_exit(q, &cio2->pci_dev->dev);
1682 	mutex_destroy(&q->lock);
1683 }
1684 
1685 static int cio2_queues_init(struct cio2_device *cio2)
1686 {
1687 	int i, r;
1688 
1689 	for (i = 0; i < CIO2_QUEUES; i++) {
1690 		r = cio2_queue_init(cio2, &cio2->queue[i]);
1691 		if (r)
1692 			break;
1693 	}
1694 
1695 	if (i == CIO2_QUEUES)
1696 		return 0;
1697 
1698 	for (i--; i >= 0; i--)
1699 		cio2_queue_exit(cio2, &cio2->queue[i]);
1700 
1701 	return r;
1702 }
1703 
1704 static void cio2_queues_exit(struct cio2_device *cio2)
1705 {
1706 	unsigned int i;
1707 
1708 	for (i = 0; i < CIO2_QUEUES; i++)
1709 		cio2_queue_exit(cio2, &cio2->queue[i]);
1710 }
1711 
1712 /**************** PCI interface ****************/
1713 
1714 static int cio2_pci_config_setup(struct pci_dev *dev)
1715 {
1716 	u16 pci_command;
1717 	int r = pci_enable_msi(dev);
1718 
1719 	if (r) {
1720 		dev_err(&dev->dev, "failed to enable MSI (%d)\n", r);
1721 		return r;
1722 	}
1723 
1724 	pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1725 	pci_command |= PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
1726 		PCI_COMMAND_INTX_DISABLE;
1727 	pci_write_config_word(dev, PCI_COMMAND, pci_command);
1728 
1729 	return 0;
1730 }
1731 
1732 static int cio2_pci_probe(struct pci_dev *pci_dev,
1733 			  const struct pci_device_id *id)
1734 {
1735 	struct cio2_device *cio2;
1736 	void __iomem *const *iomap;
1737 	int r;
1738 
1739 	cio2 = devm_kzalloc(&pci_dev->dev, sizeof(*cio2), GFP_KERNEL);
1740 	if (!cio2)
1741 		return -ENOMEM;
1742 	cio2->pci_dev = pci_dev;
1743 
1744 	r = pcim_enable_device(pci_dev);
1745 	if (r) {
1746 		dev_err(&pci_dev->dev, "failed to enable device (%d)\n", r);
1747 		return r;
1748 	}
1749 
1750 	dev_info(&pci_dev->dev, "device 0x%x (rev: 0x%x)\n",
1751 		 pci_dev->device, pci_dev->revision);
1752 
1753 	r = pcim_iomap_regions(pci_dev, 1 << CIO2_PCI_BAR, pci_name(pci_dev));
1754 	if (r) {
1755 		dev_err(&pci_dev->dev, "failed to remap I/O memory (%d)\n", r);
1756 		return -ENODEV;
1757 	}
1758 
1759 	iomap = pcim_iomap_table(pci_dev);
1760 	if (!iomap) {
1761 		dev_err(&pci_dev->dev, "failed to iomap table\n");
1762 		return -ENODEV;
1763 	}
1764 
1765 	cio2->base = iomap[CIO2_PCI_BAR];
1766 
1767 	pci_set_drvdata(pci_dev, cio2);
1768 
1769 	pci_set_master(pci_dev);
1770 
1771 	r = pci_set_dma_mask(pci_dev, CIO2_DMA_MASK);
1772 	if (r) {
1773 		dev_err(&pci_dev->dev, "failed to set DMA mask (%d)\n", r);
1774 		return -ENODEV;
1775 	}
1776 
1777 	r = cio2_pci_config_setup(pci_dev);
1778 	if (r)
1779 		return -ENODEV;
1780 
1781 	r = cio2_fbpt_init_dummy(cio2);
1782 	if (r)
1783 		return r;
1784 
1785 	mutex_init(&cio2->lock);
1786 
1787 	cio2->media_dev.dev = &cio2->pci_dev->dev;
1788 	strlcpy(cio2->media_dev.model, CIO2_DEVICE_NAME,
1789 		sizeof(cio2->media_dev.model));
1790 	snprintf(cio2->media_dev.bus_info, sizeof(cio2->media_dev.bus_info),
1791 		 "PCI:%s", pci_name(cio2->pci_dev));
1792 	cio2->media_dev.hw_revision = 0;
1793 
1794 	media_device_init(&cio2->media_dev);
1795 	r = media_device_register(&cio2->media_dev);
1796 	if (r < 0)
1797 		goto fail_mutex_destroy;
1798 
1799 	cio2->v4l2_dev.mdev = &cio2->media_dev;
1800 	r = v4l2_device_register(&pci_dev->dev, &cio2->v4l2_dev);
1801 	if (r) {
1802 		dev_err(&pci_dev->dev,
1803 			"failed to register V4L2 device (%d)\n", r);
1804 		goto fail_media_device_unregister;
1805 	}
1806 
1807 	r = cio2_queues_init(cio2);
1808 	if (r)
1809 		goto fail_v4l2_device_unregister;
1810 
1811 	/* Register notifier for subdevices we care */
1812 	r = cio2_notifier_init(cio2);
1813 	if (r)
1814 		goto fail_cio2_queue_exit;
1815 
1816 	r = devm_request_irq(&pci_dev->dev, pci_dev->irq, cio2_irq,
1817 			     IRQF_SHARED, CIO2_NAME, cio2);
1818 	if (r) {
1819 		dev_err(&pci_dev->dev, "failed to request IRQ (%d)\n", r);
1820 		goto fail;
1821 	}
1822 
1823 	pm_runtime_put_noidle(&pci_dev->dev);
1824 	pm_runtime_allow(&pci_dev->dev);
1825 
1826 	return 0;
1827 
1828 fail:
1829 	cio2_notifier_exit(cio2);
1830 fail_cio2_queue_exit:
1831 	cio2_queues_exit(cio2);
1832 fail_v4l2_device_unregister:
1833 	v4l2_device_unregister(&cio2->v4l2_dev);
1834 fail_media_device_unregister:
1835 	media_device_unregister(&cio2->media_dev);
1836 	media_device_cleanup(&cio2->media_dev);
1837 fail_mutex_destroy:
1838 	mutex_destroy(&cio2->lock);
1839 	cio2_fbpt_exit_dummy(cio2);
1840 
1841 	return r;
1842 }
1843 
1844 static void cio2_pci_remove(struct pci_dev *pci_dev)
1845 {
1846 	struct cio2_device *cio2 = pci_get_drvdata(pci_dev);
1847 	unsigned int i;
1848 
1849 	cio2_notifier_exit(cio2);
1850 	cio2_fbpt_exit_dummy(cio2);
1851 	for (i = 0; i < CIO2_QUEUES; i++)
1852 		cio2_queue_exit(cio2, &cio2->queue[i]);
1853 	v4l2_device_unregister(&cio2->v4l2_dev);
1854 	media_device_unregister(&cio2->media_dev);
1855 	media_device_cleanup(&cio2->media_dev);
1856 	mutex_destroy(&cio2->lock);
1857 }
1858 
1859 static int __maybe_unused cio2_runtime_suspend(struct device *dev)
1860 {
1861 	struct pci_dev *pci_dev = to_pci_dev(dev);
1862 	struct cio2_device *cio2 = pci_get_drvdata(pci_dev);
1863 	void __iomem *const base = cio2->base;
1864 	u16 pm;
1865 
1866 	writel(CIO2_D0I3C_I3, base + CIO2_REG_D0I3C);
1867 	dev_dbg(dev, "cio2 runtime suspend.\n");
1868 
1869 	pci_read_config_word(pci_dev, pci_dev->pm_cap + CIO2_PMCSR_OFFSET, &pm);
1870 	pm = (pm >> CIO2_PMCSR_D0D3_SHIFT) << CIO2_PMCSR_D0D3_SHIFT;
1871 	pm |= CIO2_PMCSR_D3;
1872 	pci_write_config_word(pci_dev, pci_dev->pm_cap + CIO2_PMCSR_OFFSET, pm);
1873 
1874 	return 0;
1875 }
1876 
1877 static int __maybe_unused cio2_runtime_resume(struct device *dev)
1878 {
1879 	struct pci_dev *pci_dev = to_pci_dev(dev);
1880 	struct cio2_device *cio2 = pci_get_drvdata(pci_dev);
1881 	void __iomem *const base = cio2->base;
1882 	u16 pm;
1883 
1884 	writel(CIO2_D0I3C_RR, base + CIO2_REG_D0I3C);
1885 	dev_dbg(dev, "cio2 runtime resume.\n");
1886 
1887 	pci_read_config_word(pci_dev, pci_dev->pm_cap + CIO2_PMCSR_OFFSET, &pm);
1888 	pm = (pm >> CIO2_PMCSR_D0D3_SHIFT) << CIO2_PMCSR_D0D3_SHIFT;
1889 	pci_write_config_word(pci_dev, pci_dev->pm_cap + CIO2_PMCSR_OFFSET, pm);
1890 
1891 	return 0;
1892 }
1893 
1894 /*
1895  * Helper function to advance all the elements of a circular buffer by "start"
1896  * positions
1897  */
1898 static void arrange(void *ptr, size_t elem_size, size_t elems, size_t start)
1899 {
1900 	struct {
1901 		size_t begin, end;
1902 	} arr[2] = {
1903 		{ 0, start - 1 },
1904 		{ start, elems - 1 },
1905 	};
1906 
1907 #define CHUNK_SIZE(a) ((a)->end - (a)->begin + 1)
1908 
1909 	/* Loop as long as we have out-of-place entries */
1910 	while (CHUNK_SIZE(&arr[0]) && CHUNK_SIZE(&arr[1])) {
1911 		size_t size0, i;
1912 
1913 		/*
1914 		 * Find the number of entries that can be arranged on this
1915 		 * iteration.
1916 		 */
1917 		size0 = min(CHUNK_SIZE(&arr[0]), CHUNK_SIZE(&arr[1]));
1918 
1919 		/* Swap the entries in two parts of the array. */
1920 		for (i = 0; i < size0; i++) {
1921 			u8 *d = ptr + elem_size * (arr[1].begin + i);
1922 			u8 *s = ptr + elem_size * (arr[0].begin + i);
1923 			size_t j;
1924 
1925 			for (j = 0; j < elem_size; j++)
1926 				swap(d[j], s[j]);
1927 		}
1928 
1929 		if (CHUNK_SIZE(&arr[0]) > CHUNK_SIZE(&arr[1])) {
1930 			/* The end of the first array remains unarranged. */
1931 			arr[0].begin += size0;
1932 		} else {
1933 			/*
1934 			 * The first array is fully arranged so we proceed
1935 			 * handling the next one.
1936 			 */
1937 			arr[0].begin = arr[1].begin;
1938 			arr[0].end = arr[1].begin + size0 - 1;
1939 			arr[1].begin += size0;
1940 		}
1941 	}
1942 }
1943 
1944 static void cio2_fbpt_rearrange(struct cio2_device *cio2, struct cio2_queue *q)
1945 {
1946 	unsigned int i, j;
1947 
1948 	for (i = 0, j = q->bufs_first; i < CIO2_MAX_BUFFERS;
1949 		i++, j = (j + 1) % CIO2_MAX_BUFFERS)
1950 		if (q->bufs[j])
1951 			break;
1952 
1953 	if (i == CIO2_MAX_BUFFERS)
1954 		return;
1955 
1956 	if (j) {
1957 		arrange(q->fbpt, sizeof(struct cio2_fbpt_entry) * CIO2_MAX_LOPS,
1958 			CIO2_MAX_BUFFERS, j);
1959 		arrange(q->bufs, sizeof(struct cio2_buffer *),
1960 			CIO2_MAX_BUFFERS, j);
1961 	}
1962 
1963 	/*
1964 	 * DMA clears the valid bit when accessing the buffer.
1965 	 * When stopping stream in suspend callback, some of the buffers
1966 	 * may be in invalid state. After resume, when DMA meets the invalid
1967 	 * buffer, it will halt and stop receiving new data.
1968 	 * To avoid DMA halting, set the valid bit for all buffers in FBPT.
1969 	 */
1970 	for (i = 0; i < CIO2_MAX_BUFFERS; i++)
1971 		cio2_fbpt_entry_enable(cio2, q->fbpt + i * CIO2_MAX_LOPS);
1972 }
1973 
1974 static int __maybe_unused cio2_suspend(struct device *dev)
1975 {
1976 	struct pci_dev *pci_dev = to_pci_dev(dev);
1977 	struct cio2_device *cio2 = pci_get_drvdata(pci_dev);
1978 	struct cio2_queue *q = cio2->cur_queue;
1979 
1980 	dev_dbg(dev, "cio2 suspend\n");
1981 	if (!cio2->streaming)
1982 		return 0;
1983 
1984 	/* Stop stream */
1985 	cio2_hw_exit(cio2, q);
1986 	synchronize_irq(pci_dev->irq);
1987 
1988 	pm_runtime_force_suspend(dev);
1989 
1990 	/*
1991 	 * Upon resume, hw starts to process the fbpt entries from beginning,
1992 	 * so relocate the queued buffs to the fbpt head before suspend.
1993 	 */
1994 	cio2_fbpt_rearrange(cio2, q);
1995 	q->bufs_first = 0;
1996 	q->bufs_next = 0;
1997 
1998 	return 0;
1999 }
2000 
2001 static int __maybe_unused cio2_resume(struct device *dev)
2002 {
2003 	struct pci_dev *pci_dev = to_pci_dev(dev);
2004 	struct cio2_device *cio2 = pci_get_drvdata(pci_dev);
2005 	int r = 0;
2006 	struct cio2_queue *q = cio2->cur_queue;
2007 
2008 	dev_dbg(dev, "cio2 resume\n");
2009 	if (!cio2->streaming)
2010 		return 0;
2011 	/* Start stream */
2012 	r = pm_runtime_force_resume(&cio2->pci_dev->dev);
2013 	if (r < 0) {
2014 		dev_err(&cio2->pci_dev->dev,
2015 			"failed to set power %d\n", r);
2016 		return r;
2017 	}
2018 
2019 	r = cio2_hw_init(cio2, q);
2020 	if (r)
2021 		dev_err(dev, "fail to init cio2 hw\n");
2022 
2023 	return r;
2024 }
2025 
2026 static const struct dev_pm_ops cio2_pm_ops = {
2027 	SET_RUNTIME_PM_OPS(&cio2_runtime_suspend, &cio2_runtime_resume, NULL)
2028 	SET_SYSTEM_SLEEP_PM_OPS(&cio2_suspend, &cio2_resume)
2029 };
2030 
2031 static const struct pci_device_id cio2_pci_id_table[] = {
2032 	{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, CIO2_PCI_ID) },
2033 	{ 0 }
2034 };
2035 
2036 MODULE_DEVICE_TABLE(pci, cio2_pci_id_table);
2037 
2038 static struct pci_driver cio2_pci_driver = {
2039 	.name = CIO2_NAME,
2040 	.id_table = cio2_pci_id_table,
2041 	.probe = cio2_pci_probe,
2042 	.remove = cio2_pci_remove,
2043 	.driver = {
2044 		.pm = &cio2_pm_ops,
2045 	},
2046 };
2047 
2048 module_pci_driver(cio2_pci_driver);
2049 
2050 MODULE_AUTHOR("Tuukka Toivonen <tuukka.toivonen@intel.com>");
2051 MODULE_AUTHOR("Tianshu Qiu <tian.shu.qiu@intel.com>");
2052 MODULE_AUTHOR("Jian Xu Zheng <jian.xu.zheng@intel.com>");
2053 MODULE_AUTHOR("Yuning Pu <yuning.pu@intel.com>");
2054 MODULE_AUTHOR("Yong Zhi <yong.zhi@intel.com>");
2055 MODULE_LICENSE("GPL v2");
2056 MODULE_DESCRIPTION("IPU3 CIO2 driver");
2057