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 	return 0;
227 }
228 
229 static void cio2_fbpt_exit(struct cio2_queue *q, struct device *dev)
230 {
231 	dma_free_coherent(dev, CIO2_FBPT_SIZE, q->fbpt, q->fbpt_bus_addr);
232 }
233 
234 /**************** CSI2 hardware setup ****************/
235 
236 /*
237  * The CSI2 receiver has several parameters affecting
238  * the receiver timings. These depend on the MIPI bus frequency
239  * F in Hz (sensor transmitter rate) as follows:
240  *     register value = (A/1e9 + B * UI) / COUNT_ACC
241  * where
242  *      UI = 1 / (2 * F) in seconds
243  *      COUNT_ACC = counter accuracy in seconds
244  *      For IPU3 COUNT_ACC = 0.0625
245  *
246  * A and B are coefficients from the table below,
247  * depending whether the register minimum or maximum value is
248  * calculated.
249  *                                     Minimum     Maximum
250  * Clock lane                          A     B     A     B
251  * reg_rx_csi_dly_cnt_termen_clane     0     0    38     0
252  * reg_rx_csi_dly_cnt_settle_clane    95    -8   300   -16
253  * Data lanes
254  * reg_rx_csi_dly_cnt_termen_dlane0    0     0    35     4
255  * reg_rx_csi_dly_cnt_settle_dlane0   85    -2   145    -6
256  * reg_rx_csi_dly_cnt_termen_dlane1    0     0    35     4
257  * reg_rx_csi_dly_cnt_settle_dlane1   85    -2   145    -6
258  * reg_rx_csi_dly_cnt_termen_dlane2    0     0    35     4
259  * reg_rx_csi_dly_cnt_settle_dlane2   85    -2   145    -6
260  * reg_rx_csi_dly_cnt_termen_dlane3    0     0    35     4
261  * reg_rx_csi_dly_cnt_settle_dlane3   85    -2   145    -6
262  *
263  * We use the minimum values of both A and B.
264  */
265 
266 /*
267  * shift for keeping value range suitable for 32-bit integer arithmetic
268  */
269 #define LIMIT_SHIFT	8
270 
271 static s32 cio2_rx_timing(s32 a, s32 b, s64 freq, int def)
272 {
273 	const u32 accinv = 16; /* invert of counter resolution */
274 	const u32 uiinv = 500000000; /* 1e9 / 2 */
275 	s32 r;
276 
277 	freq >>= LIMIT_SHIFT;
278 
279 	if (WARN_ON(freq <= 0 || freq > S32_MAX))
280 		return def;
281 	/*
282 	 * b could be 0, -2 or -8, so |accinv * b| is always
283 	 * less than (1 << ds) and thus |r| < 500000000.
284 	 */
285 	r = accinv * b * (uiinv >> LIMIT_SHIFT);
286 	r = r / (s32)freq;
287 	/* max value of a is 95 */
288 	r += accinv * a;
289 
290 	return r;
291 };
292 
293 /* Calculate the the delay value for termination enable of clock lane HS Rx */
294 static int cio2_csi2_calc_timing(struct cio2_device *cio2, struct cio2_queue *q,
295 				 struct cio2_csi2_timing *timing)
296 {
297 	struct device *dev = &cio2->pci_dev->dev;
298 	struct v4l2_querymenu qm = {.id = V4L2_CID_LINK_FREQ, };
299 	struct v4l2_ctrl *link_freq;
300 	s64 freq;
301 	int r;
302 
303 	if (!q->sensor)
304 		return -ENODEV;
305 
306 	link_freq = v4l2_ctrl_find(q->sensor->ctrl_handler, V4L2_CID_LINK_FREQ);
307 	if (!link_freq) {
308 		dev_err(dev, "failed to find LINK_FREQ\n");
309 		return -EPIPE;
310 	}
311 
312 	qm.index = v4l2_ctrl_g_ctrl(link_freq);
313 	r = v4l2_querymenu(q->sensor->ctrl_handler, &qm);
314 	if (r) {
315 		dev_err(dev, "failed to get menu item\n");
316 		return r;
317 	}
318 
319 	if (!qm.value) {
320 		dev_err(dev, "error invalid link_freq\n");
321 		return -EINVAL;
322 	}
323 	freq = qm.value;
324 
325 	timing->clk_termen = cio2_rx_timing(CIO2_CSIRX_DLY_CNT_TERMEN_CLANE_A,
326 					    CIO2_CSIRX_DLY_CNT_TERMEN_CLANE_B,
327 					    freq,
328 					    CIO2_CSIRX_DLY_CNT_TERMEN_DEFAULT);
329 	timing->clk_settle = cio2_rx_timing(CIO2_CSIRX_DLY_CNT_SETTLE_CLANE_A,
330 					    CIO2_CSIRX_DLY_CNT_SETTLE_CLANE_B,
331 					    freq,
332 					    CIO2_CSIRX_DLY_CNT_SETTLE_DEFAULT);
333 	timing->dat_termen = cio2_rx_timing(CIO2_CSIRX_DLY_CNT_TERMEN_DLANE_A,
334 					    CIO2_CSIRX_DLY_CNT_TERMEN_DLANE_B,
335 					    freq,
336 					    CIO2_CSIRX_DLY_CNT_TERMEN_DEFAULT);
337 	timing->dat_settle = cio2_rx_timing(CIO2_CSIRX_DLY_CNT_SETTLE_DLANE_A,
338 					    CIO2_CSIRX_DLY_CNT_SETTLE_DLANE_B,
339 					    freq,
340 					    CIO2_CSIRX_DLY_CNT_SETTLE_DEFAULT);
341 
342 	dev_dbg(dev, "freq ct value is %d\n", timing->clk_termen);
343 	dev_dbg(dev, "freq cs value is %d\n", timing->clk_settle);
344 	dev_dbg(dev, "freq dt value is %d\n", timing->dat_termen);
345 	dev_dbg(dev, "freq ds value is %d\n", timing->dat_settle);
346 
347 	return 0;
348 };
349 
350 static int cio2_hw_init(struct cio2_device *cio2, struct cio2_queue *q)
351 {
352 	static const int NUM_VCS = 4;
353 	static const int SID;	/* Stream id */
354 	static const int ENTRY;
355 	static const int FBPT_WIDTH = DIV_ROUND_UP(CIO2_MAX_LOPS,
356 					CIO2_FBPT_SUBENTRY_UNIT);
357 	const u32 num_buffers1 = CIO2_MAX_BUFFERS - 1;
358 	const struct ipu3_cio2_fmt *fmt;
359 	void __iomem *const base = cio2->base;
360 	u8 lanes, csi2bus = q->csi2.port;
361 	u8 sensor_vc = SENSOR_VIR_CH_DFLT;
362 	struct cio2_csi2_timing timing;
363 	int i, r;
364 
365 	fmt = cio2_find_format(NULL, &q->subdev_fmt.code);
366 	if (!fmt)
367 		return -EINVAL;
368 
369 	lanes = q->csi2.lanes;
370 
371 	r = cio2_csi2_calc_timing(cio2, q, &timing);
372 	if (r)
373 		return r;
374 
375 	writel(timing.clk_termen, q->csi_rx_base +
376 		CIO2_REG_CSIRX_DLY_CNT_TERMEN(CIO2_CSIRX_DLY_CNT_CLANE_IDX));
377 	writel(timing.clk_settle, q->csi_rx_base +
378 		CIO2_REG_CSIRX_DLY_CNT_SETTLE(CIO2_CSIRX_DLY_CNT_CLANE_IDX));
379 
380 	for (i = 0; i < lanes; i++) {
381 		writel(timing.dat_termen, q->csi_rx_base +
382 			CIO2_REG_CSIRX_DLY_CNT_TERMEN(i));
383 		writel(timing.dat_settle, q->csi_rx_base +
384 			CIO2_REG_CSIRX_DLY_CNT_SETTLE(i));
385 	}
386 
387 	writel(CIO2_PBM_WMCTRL1_MIN_2CK |
388 	       CIO2_PBM_WMCTRL1_MID1_2CK |
389 	       CIO2_PBM_WMCTRL1_MID2_2CK, base + CIO2_REG_PBM_WMCTRL1);
390 	writel(CIO2_PBM_WMCTRL2_HWM_2CK << CIO2_PBM_WMCTRL2_HWM_2CK_SHIFT |
391 	       CIO2_PBM_WMCTRL2_LWM_2CK << CIO2_PBM_WMCTRL2_LWM_2CK_SHIFT |
392 	       CIO2_PBM_WMCTRL2_OBFFWM_2CK <<
393 	       CIO2_PBM_WMCTRL2_OBFFWM_2CK_SHIFT |
394 	       CIO2_PBM_WMCTRL2_TRANSDYN << CIO2_PBM_WMCTRL2_TRANSDYN_SHIFT |
395 	       CIO2_PBM_WMCTRL2_OBFF_MEM_EN, base + CIO2_REG_PBM_WMCTRL2);
396 	writel(CIO2_PBM_ARB_CTRL_LANES_DIV <<
397 	       CIO2_PBM_ARB_CTRL_LANES_DIV_SHIFT |
398 	       CIO2_PBM_ARB_CTRL_LE_EN |
399 	       CIO2_PBM_ARB_CTRL_PLL_POST_SHTDN <<
400 	       CIO2_PBM_ARB_CTRL_PLL_POST_SHTDN_SHIFT |
401 	       CIO2_PBM_ARB_CTRL_PLL_AHD_WK_UP <<
402 	       CIO2_PBM_ARB_CTRL_PLL_AHD_WK_UP_SHIFT,
403 	       base + CIO2_REG_PBM_ARB_CTRL);
404 	writel(CIO2_CSIRX_STATUS_DLANE_HS_MASK,
405 	       q->csi_rx_base + CIO2_REG_CSIRX_STATUS_DLANE_HS);
406 	writel(CIO2_CSIRX_STATUS_DLANE_LP_MASK,
407 	       q->csi_rx_base + CIO2_REG_CSIRX_STATUS_DLANE_LP);
408 
409 	writel(CIO2_FB_HPLL_FREQ, base + CIO2_REG_FB_HPLL_FREQ);
410 	writel(CIO2_ISCLK_RATIO, base + CIO2_REG_ISCLK_RATIO);
411 
412 	/* Configure MIPI backend */
413 	for (i = 0; i < NUM_VCS; i++)
414 		writel(1, q->csi_rx_base + CIO2_REG_MIPIBE_SP_LUT_ENTRY(i));
415 
416 	/* There are 16 short packet LUT entry */
417 	for (i = 0; i < 16; i++)
418 		writel(CIO2_MIPIBE_LP_LUT_ENTRY_DISREGARD,
419 		       q->csi_rx_base + CIO2_REG_MIPIBE_LP_LUT_ENTRY(i));
420 	writel(CIO2_MIPIBE_GLOBAL_LUT_DISREGARD,
421 	       q->csi_rx_base + CIO2_REG_MIPIBE_GLOBAL_LUT_DISREGARD);
422 
423 	writel(CIO2_INT_EN_EXT_IE_MASK, base + CIO2_REG_INT_EN_EXT_IE);
424 	writel(CIO2_IRQCTRL_MASK, q->csi_rx_base + CIO2_REG_IRQCTRL_MASK);
425 	writel(CIO2_IRQCTRL_MASK, q->csi_rx_base + CIO2_REG_IRQCTRL_ENABLE);
426 	writel(0, q->csi_rx_base + CIO2_REG_IRQCTRL_EDGE);
427 	writel(0, q->csi_rx_base + CIO2_REG_IRQCTRL_LEVEL_NOT_PULSE);
428 	writel(CIO2_INT_EN_EXT_OE_MASK, base + CIO2_REG_INT_EN_EXT_OE);
429 
430 	writel(CIO2_REG_INT_EN_IRQ | CIO2_INT_IOC(CIO2_DMA_CHAN) |
431 	       CIO2_REG_INT_EN_IOS(CIO2_DMA_CHAN),
432 	       base + CIO2_REG_INT_EN);
433 
434 	writel((CIO2_PXM_PXF_FMT_CFG_BPP_10 | CIO2_PXM_PXF_FMT_CFG_PCK_64B)
435 	       << CIO2_PXM_PXF_FMT_CFG_SID0_SHIFT,
436 	       base + CIO2_REG_PXM_PXF_FMT_CFG0(csi2bus));
437 	writel(SID << CIO2_MIPIBE_LP_LUT_ENTRY_SID_SHIFT |
438 	       sensor_vc << CIO2_MIPIBE_LP_LUT_ENTRY_VC_SHIFT |
439 	       fmt->mipicode << CIO2_MIPIBE_LP_LUT_ENTRY_FORMAT_TYPE_SHIFT,
440 	       q->csi_rx_base + CIO2_REG_MIPIBE_LP_LUT_ENTRY(ENTRY));
441 	writel(0, q->csi_rx_base + CIO2_REG_MIPIBE_COMP_FORMAT(sensor_vc));
442 	writel(0, q->csi_rx_base + CIO2_REG_MIPIBE_FORCE_RAW8);
443 	writel(0, base + CIO2_REG_PXM_SID2BID0(csi2bus));
444 
445 	writel(lanes, q->csi_rx_base + CIO2_REG_CSIRX_NOF_ENABLED_LANES);
446 	writel(CIO2_CGC_PRIM_TGE |
447 	       CIO2_CGC_SIDE_TGE |
448 	       CIO2_CGC_XOSC_TGE |
449 	       CIO2_CGC_D3I3_TGE |
450 	       CIO2_CGC_CSI2_INTERFRAME_TGE |
451 	       CIO2_CGC_CSI2_PORT_DCGE |
452 	       CIO2_CGC_SIDE_DCGE |
453 	       CIO2_CGC_PRIM_DCGE |
454 	       CIO2_CGC_ROSC_DCGE |
455 	       CIO2_CGC_XOSC_DCGE |
456 	       CIO2_CGC_CLKGATE_HOLDOFF << CIO2_CGC_CLKGATE_HOLDOFF_SHIFT |
457 	       CIO2_CGC_CSI_CLKGATE_HOLDOFF
458 	       << CIO2_CGC_CSI_CLKGATE_HOLDOFF_SHIFT, base + CIO2_REG_CGC);
459 	writel(CIO2_LTRCTRL_LTRDYNEN, base + CIO2_REG_LTRCTRL);
460 	writel(CIO2_LTRVAL0_VAL << CIO2_LTRVAL02_VAL_SHIFT |
461 	       CIO2_LTRVAL0_SCALE << CIO2_LTRVAL02_SCALE_SHIFT |
462 	       CIO2_LTRVAL1_VAL << CIO2_LTRVAL13_VAL_SHIFT |
463 	       CIO2_LTRVAL1_SCALE << CIO2_LTRVAL13_SCALE_SHIFT,
464 	       base + CIO2_REG_LTRVAL01);
465 	writel(CIO2_LTRVAL2_VAL << CIO2_LTRVAL02_VAL_SHIFT |
466 	       CIO2_LTRVAL2_SCALE << CIO2_LTRVAL02_SCALE_SHIFT |
467 	       CIO2_LTRVAL3_VAL << CIO2_LTRVAL13_VAL_SHIFT |
468 	       CIO2_LTRVAL3_SCALE << CIO2_LTRVAL13_SCALE_SHIFT,
469 	       base + CIO2_REG_LTRVAL23);
470 
471 	for (i = 0; i < CIO2_NUM_DMA_CHAN; i++) {
472 		writel(0, base + CIO2_REG_CDMABA(i));
473 		writel(0, base + CIO2_REG_CDMAC0(i));
474 		writel(0, base + CIO2_REG_CDMAC1(i));
475 	}
476 
477 	/* Enable DMA */
478 	writel(q->fbpt_bus_addr >> PAGE_SHIFT,
479 	       base + CIO2_REG_CDMABA(CIO2_DMA_CHAN));
480 
481 	writel(num_buffers1 << CIO2_CDMAC0_FBPT_LEN_SHIFT |
482 	       FBPT_WIDTH << CIO2_CDMAC0_FBPT_WIDTH_SHIFT |
483 	       CIO2_CDMAC0_DMA_INTR_ON_FE |
484 	       CIO2_CDMAC0_FBPT_UPDATE_FIFO_FULL |
485 	       CIO2_CDMAC0_DMA_EN |
486 	       CIO2_CDMAC0_DMA_INTR_ON_FS |
487 	       CIO2_CDMAC0_DMA_HALTED, base + CIO2_REG_CDMAC0(CIO2_DMA_CHAN));
488 
489 	writel(1 << CIO2_CDMAC1_LINENUMUPDATE_SHIFT,
490 	       base + CIO2_REG_CDMAC1(CIO2_DMA_CHAN));
491 
492 	writel(0, base + CIO2_REG_PBM_FOPN_ABORT);
493 
494 	writel(CIO2_PXM_FRF_CFG_CRC_TH << CIO2_PXM_FRF_CFG_CRC_TH_SHIFT |
495 	       CIO2_PXM_FRF_CFG_MSK_ECC_DPHY_NR |
496 	       CIO2_PXM_FRF_CFG_MSK_ECC_RE |
497 	       CIO2_PXM_FRF_CFG_MSK_ECC_DPHY_NE,
498 	       base + CIO2_REG_PXM_FRF_CFG(q->csi2.port));
499 
500 	/* Clear interrupts */
501 	writel(CIO2_IRQCTRL_MASK, q->csi_rx_base + CIO2_REG_IRQCTRL_CLEAR);
502 	writel(~0, base + CIO2_REG_INT_STS_EXT_OE);
503 	writel(~0, base + CIO2_REG_INT_STS_EXT_IE);
504 	writel(~0, base + CIO2_REG_INT_STS);
505 
506 	/* Enable devices, starting from the last device in the pipe */
507 	writel(1, q->csi_rx_base + CIO2_REG_MIPIBE_ENABLE);
508 	writel(1, q->csi_rx_base + CIO2_REG_CSIRX_ENABLE);
509 
510 	return 0;
511 }
512 
513 static void cio2_hw_exit(struct cio2_device *cio2, struct cio2_queue *q)
514 {
515 	void __iomem *base = cio2->base;
516 	unsigned int i, maxloops = 1000;
517 
518 	/* Disable CSI receiver and MIPI backend devices */
519 	writel(0, q->csi_rx_base + CIO2_REG_IRQCTRL_MASK);
520 	writel(0, q->csi_rx_base + CIO2_REG_IRQCTRL_ENABLE);
521 	writel(0, q->csi_rx_base + CIO2_REG_CSIRX_ENABLE);
522 	writel(0, q->csi_rx_base + CIO2_REG_MIPIBE_ENABLE);
523 
524 	/* Halt DMA */
525 	writel(0, base + CIO2_REG_CDMAC0(CIO2_DMA_CHAN));
526 	do {
527 		if (readl(base + CIO2_REG_CDMAC0(CIO2_DMA_CHAN)) &
528 		    CIO2_CDMAC0_DMA_HALTED)
529 			break;
530 		usleep_range(1000, 2000);
531 	} while (--maxloops);
532 	if (!maxloops)
533 		dev_err(&cio2->pci_dev->dev,
534 			"DMA %i can not be halted\n", CIO2_DMA_CHAN);
535 
536 	for (i = 0; i < CIO2_NUM_PORTS; i++) {
537 		writel(readl(base + CIO2_REG_PXM_FRF_CFG(i)) |
538 		       CIO2_PXM_FRF_CFG_ABORT, base + CIO2_REG_PXM_FRF_CFG(i));
539 		writel(readl(base + CIO2_REG_PBM_FOPN_ABORT) |
540 		       CIO2_PBM_FOPN_ABORT(i), base + CIO2_REG_PBM_FOPN_ABORT);
541 	}
542 }
543 
544 static void cio2_buffer_done(struct cio2_device *cio2, unsigned int dma_chan)
545 {
546 	struct device *dev = &cio2->pci_dev->dev;
547 	struct cio2_queue *q = cio2->cur_queue;
548 	int buffers_found = 0;
549 	u64 ns = ktime_get_ns();
550 
551 	if (dma_chan >= CIO2_QUEUES) {
552 		dev_err(dev, "bad DMA channel %i\n", dma_chan);
553 		return;
554 	}
555 
556 	/* Find out which buffer(s) are ready */
557 	do {
558 		struct cio2_fbpt_entry *const entry =
559 			&q->fbpt[q->bufs_first * CIO2_MAX_LOPS];
560 		struct cio2_buffer *b;
561 
562 		if (entry->first_entry.ctrl & CIO2_FBPT_CTRL_VALID)
563 			break;
564 
565 		b = q->bufs[q->bufs_first];
566 		if (b) {
567 			unsigned int bytes = entry[1].second_entry.num_of_bytes;
568 
569 			q->bufs[q->bufs_first] = NULL;
570 			atomic_dec(&q->bufs_queued);
571 			dev_dbg(&cio2->pci_dev->dev,
572 				"buffer %i done\n", b->vbb.vb2_buf.index);
573 
574 			b->vbb.vb2_buf.timestamp = ns;
575 			b->vbb.field = V4L2_FIELD_NONE;
576 			b->vbb.sequence = atomic_read(&q->frame_sequence);
577 			if (b->vbb.vb2_buf.planes[0].length != bytes)
578 				dev_warn(dev, "buffer length is %d received %d\n",
579 					 b->vbb.vb2_buf.planes[0].length,
580 					 bytes);
581 			vb2_buffer_done(&b->vbb.vb2_buf, VB2_BUF_STATE_DONE);
582 		}
583 		atomic_inc(&q->frame_sequence);
584 		cio2_fbpt_entry_init_dummy(cio2, entry);
585 		q->bufs_first = (q->bufs_first + 1) % CIO2_MAX_BUFFERS;
586 		buffers_found++;
587 	} while (1);
588 
589 	if (buffers_found == 0)
590 		dev_warn(&cio2->pci_dev->dev,
591 			 "no ready buffers found on DMA channel %u\n",
592 			 dma_chan);
593 }
594 
595 static void cio2_queue_event_sof(struct cio2_device *cio2, struct cio2_queue *q)
596 {
597 	/*
598 	 * For the user space camera control algorithms it is essential
599 	 * to know when the reception of a frame has begun. That's often
600 	 * the best timing information to get from the hardware.
601 	 */
602 	struct v4l2_event event = {
603 		.type = V4L2_EVENT_FRAME_SYNC,
604 		.u.frame_sync.frame_sequence = atomic_read(&q->frame_sequence),
605 	};
606 
607 	v4l2_event_queue(q->subdev.devnode, &event);
608 }
609 
610 static const char *const cio2_irq_errs[] = {
611 	"single packet header error corrected",
612 	"multiple packet header errors detected",
613 	"payload checksum (CRC) error",
614 	"fifo overflow",
615 	"reserved short packet data type detected",
616 	"reserved long packet data type detected",
617 	"incomplete long packet detected",
618 	"frame sync error",
619 	"line sync error",
620 	"DPHY start of transmission error",
621 	"DPHY synchronization error",
622 	"escape mode error",
623 	"escape mode trigger event",
624 	"escape mode ultra-low power state for data lane(s)",
625 	"escape mode ultra-low power state exit for clock lane",
626 	"inter-frame short packet discarded",
627 	"inter-frame long packet discarded",
628 	"non-matching Long Packet stalled",
629 };
630 
631 static const char *const cio2_port_errs[] = {
632 	"ECC recoverable",
633 	"DPHY not recoverable",
634 	"ECC not recoverable",
635 	"CRC error",
636 	"INTERFRAMEDATA",
637 	"PKT2SHORT",
638 	"PKT2LONG",
639 };
640 
641 static void cio2_irq_handle_once(struct cio2_device *cio2, u32 int_status)
642 {
643 	void __iomem *const base = cio2->base;
644 	struct device *dev = &cio2->pci_dev->dev;
645 
646 	if (int_status & CIO2_INT_IOOE) {
647 		/*
648 		 * Interrupt on Output Error:
649 		 * 1) SRAM is full and FS received, or
650 		 * 2) An invalid bit detected by DMA.
651 		 */
652 		u32 oe_status, oe_clear;
653 
654 		oe_clear = readl(base + CIO2_REG_INT_STS_EXT_OE);
655 		oe_status = oe_clear;
656 
657 		if (oe_status & CIO2_INT_EXT_OE_DMAOE_MASK) {
658 			dev_err(dev, "DMA output error: 0x%x\n",
659 				(oe_status & CIO2_INT_EXT_OE_DMAOE_MASK)
660 				>> CIO2_INT_EXT_OE_DMAOE_SHIFT);
661 			oe_status &= ~CIO2_INT_EXT_OE_DMAOE_MASK;
662 		}
663 		if (oe_status & CIO2_INT_EXT_OE_OES_MASK) {
664 			dev_err(dev, "DMA output error on CSI2 buses: 0x%x\n",
665 				(oe_status & CIO2_INT_EXT_OE_OES_MASK)
666 				>> CIO2_INT_EXT_OE_OES_SHIFT);
667 			oe_status &= ~CIO2_INT_EXT_OE_OES_MASK;
668 		}
669 		writel(oe_clear, base + CIO2_REG_INT_STS_EXT_OE);
670 		if (oe_status)
671 			dev_warn(dev, "unknown interrupt 0x%x on OE\n",
672 				 oe_status);
673 		int_status &= ~CIO2_INT_IOOE;
674 	}
675 
676 	if (int_status & CIO2_INT_IOC_MASK) {
677 		/* DMA IO done -- frame ready */
678 		u32 clr = 0;
679 		unsigned int d;
680 
681 		for (d = 0; d < CIO2_NUM_DMA_CHAN; d++)
682 			if (int_status & CIO2_INT_IOC(d)) {
683 				clr |= CIO2_INT_IOC(d);
684 				cio2_buffer_done(cio2, d);
685 			}
686 		int_status &= ~clr;
687 	}
688 
689 	if (int_status & CIO2_INT_IOS_IOLN_MASK) {
690 		/* DMA IO starts or reached specified line */
691 		u32 clr = 0;
692 		unsigned int d;
693 
694 		for (d = 0; d < CIO2_NUM_DMA_CHAN; d++)
695 			if (int_status & CIO2_INT_IOS_IOLN(d)) {
696 				clr |= CIO2_INT_IOS_IOLN(d);
697 				if (d == CIO2_DMA_CHAN)
698 					cio2_queue_event_sof(cio2,
699 							     cio2->cur_queue);
700 			}
701 		int_status &= ~clr;
702 	}
703 
704 	if (int_status & (CIO2_INT_IOIE | CIO2_INT_IOIRQ)) {
705 		/* CSI2 receiver (error) interrupt */
706 		u32 ie_status, ie_clear;
707 		unsigned int port;
708 
709 		ie_clear = readl(base + CIO2_REG_INT_STS_EXT_IE);
710 		ie_status = ie_clear;
711 
712 		for (port = 0; port < CIO2_NUM_PORTS; port++) {
713 			u32 port_status = (ie_status >> (port * 8)) & 0xff;
714 			u32 err_mask = BIT_MASK(ARRAY_SIZE(cio2_port_errs)) - 1;
715 			void __iomem *const csi_rx_base =
716 						base + CIO2_REG_PIPE_BASE(port);
717 			unsigned int i;
718 
719 			while (port_status & err_mask) {
720 				i = ffs(port_status) - 1;
721 				dev_err(dev, "port %i error %s\n",
722 					port, cio2_port_errs[i]);
723 				ie_status &= ~BIT(port * 8 + i);
724 				port_status &= ~BIT(i);
725 			}
726 
727 			if (ie_status & CIO2_INT_EXT_IE_IRQ(port)) {
728 				u32 csi2_status, csi2_clear;
729 
730 				csi2_status = readl(csi_rx_base +
731 						CIO2_REG_IRQCTRL_STATUS);
732 				csi2_clear = csi2_status;
733 				err_mask =
734 					BIT_MASK(ARRAY_SIZE(cio2_irq_errs)) - 1;
735 
736 				while (csi2_status & err_mask) {
737 					i = ffs(csi2_status) - 1;
738 					dev_err(dev,
739 						"CSI-2 receiver port %i: %s\n",
740 							port, cio2_irq_errs[i]);
741 					csi2_status &= ~BIT(i);
742 				}
743 
744 				writel(csi2_clear,
745 				       csi_rx_base + CIO2_REG_IRQCTRL_CLEAR);
746 				if (csi2_status)
747 					dev_warn(dev,
748 						 "unknown CSI2 error 0x%x on port %i\n",
749 						 csi2_status, port);
750 
751 				ie_status &= ~CIO2_INT_EXT_IE_IRQ(port);
752 			}
753 		}
754 
755 		writel(ie_clear, base + CIO2_REG_INT_STS_EXT_IE);
756 		if (ie_status)
757 			dev_warn(dev, "unknown interrupt 0x%x on IE\n",
758 				 ie_status);
759 
760 		int_status &= ~(CIO2_INT_IOIE | CIO2_INT_IOIRQ);
761 	}
762 
763 	if (int_status)
764 		dev_warn(dev, "unknown interrupt 0x%x on INT\n", int_status);
765 }
766 
767 static irqreturn_t cio2_irq(int irq, void *cio2_ptr)
768 {
769 	struct cio2_device *cio2 = cio2_ptr;
770 	void __iomem *const base = cio2->base;
771 	struct device *dev = &cio2->pci_dev->dev;
772 	u32 int_status;
773 
774 	int_status = readl(base + CIO2_REG_INT_STS);
775 	dev_dbg(dev, "isr enter - interrupt status 0x%x\n", int_status);
776 	if (!int_status)
777 		return IRQ_NONE;
778 
779 	do {
780 		writel(int_status, base + CIO2_REG_INT_STS);
781 		cio2_irq_handle_once(cio2, int_status);
782 		int_status = readl(base + CIO2_REG_INT_STS);
783 		if (int_status)
784 			dev_dbg(dev, "pending status 0x%x\n", int_status);
785 	} while (int_status);
786 
787 	return IRQ_HANDLED;
788 }
789 
790 /**************** Videobuf2 interface ****************/
791 
792 static void cio2_vb2_return_all_buffers(struct cio2_queue *q,
793 					enum vb2_buffer_state state)
794 {
795 	unsigned int i;
796 
797 	for (i = 0; i < CIO2_MAX_BUFFERS; i++) {
798 		if (q->bufs[i]) {
799 			atomic_dec(&q->bufs_queued);
800 			vb2_buffer_done(&q->bufs[i]->vbb.vb2_buf,
801 					state);
802 		}
803 	}
804 }
805 
806 static int cio2_vb2_queue_setup(struct vb2_queue *vq,
807 				unsigned int *num_buffers,
808 				unsigned int *num_planes,
809 				unsigned int sizes[],
810 				struct device *alloc_devs[])
811 {
812 	struct cio2_device *cio2 = vb2_get_drv_priv(vq);
813 	struct cio2_queue *q = vb2q_to_cio2_queue(vq);
814 	unsigned int i;
815 
816 	*num_planes = q->format.num_planes;
817 
818 	for (i = 0; i < *num_planes; ++i) {
819 		sizes[i] = q->format.plane_fmt[i].sizeimage;
820 		alloc_devs[i] = &cio2->pci_dev->dev;
821 	}
822 
823 	*num_buffers = clamp_val(*num_buffers, 1, CIO2_MAX_BUFFERS);
824 
825 	/* Initialize buffer queue */
826 	for (i = 0; i < CIO2_MAX_BUFFERS; i++) {
827 		q->bufs[i] = NULL;
828 		cio2_fbpt_entry_init_dummy(cio2, &q->fbpt[i * CIO2_MAX_LOPS]);
829 	}
830 	atomic_set(&q->bufs_queued, 0);
831 	q->bufs_first = 0;
832 	q->bufs_next = 0;
833 
834 	return 0;
835 }
836 
837 /* Called after each buffer is allocated */
838 static int cio2_vb2_buf_init(struct vb2_buffer *vb)
839 {
840 	struct cio2_device *cio2 = vb2_get_drv_priv(vb->vb2_queue);
841 	struct device *dev = &cio2->pci_dev->dev;
842 	struct cio2_buffer *b =
843 		container_of(vb, struct cio2_buffer, vbb.vb2_buf);
844 	static const unsigned int entries_per_page =
845 		CIO2_PAGE_SIZE / sizeof(u32);
846 	unsigned int pages = DIV_ROUND_UP(vb->planes[0].length, CIO2_PAGE_SIZE);
847 	unsigned int lops = DIV_ROUND_UP(pages + 1, entries_per_page);
848 	struct sg_table *sg;
849 	struct sg_dma_page_iter sg_iter;
850 	int i, j;
851 
852 	if (lops <= 0 || lops > CIO2_MAX_LOPS) {
853 		dev_err(dev, "%s: bad buffer size (%i)\n", __func__,
854 			vb->planes[0].length);
855 		return -ENOSPC;		/* Should never happen */
856 	}
857 
858 	memset(b->lop, 0, sizeof(b->lop));
859 	/* Allocate LOP table */
860 	for (i = 0; i < lops; i++) {
861 		b->lop[i] = dma_alloc_coherent(dev, CIO2_PAGE_SIZE,
862 					       &b->lop_bus_addr[i], GFP_KERNEL);
863 		if (!b->lop[i])
864 			goto fail;
865 	}
866 
867 	/* Fill LOP */
868 	sg = vb2_dma_sg_plane_desc(vb, 0);
869 	if (!sg)
870 		return -ENOMEM;
871 
872 	if (sg->nents && sg->sgl)
873 		b->offset = sg->sgl->offset;
874 
875 	i = j = 0;
876 	for_each_sg_dma_page (sg->sgl, &sg_iter, sg->nents, 0) {
877 		if (!pages--)
878 			break;
879 		b->lop[i][j] = sg_page_iter_dma_address(&sg_iter) >> PAGE_SHIFT;
880 		j++;
881 		if (j == entries_per_page) {
882 			i++;
883 			j = 0;
884 		}
885 	}
886 
887 	b->lop[i][j] = cio2->dummy_page_bus_addr >> PAGE_SHIFT;
888 	return 0;
889 fail:
890 	for (i--; i >= 0; i--)
891 		dma_free_coherent(dev, CIO2_PAGE_SIZE,
892 				  b->lop[i], b->lop_bus_addr[i]);
893 	return -ENOMEM;
894 }
895 
896 /* Transfer buffer ownership to cio2 */
897 static void cio2_vb2_buf_queue(struct vb2_buffer *vb)
898 {
899 	struct cio2_device *cio2 = vb2_get_drv_priv(vb->vb2_queue);
900 	struct cio2_queue *q =
901 		container_of(vb->vb2_queue, struct cio2_queue, vbq);
902 	struct cio2_buffer *b =
903 		container_of(vb, struct cio2_buffer, vbb.vb2_buf);
904 	struct cio2_fbpt_entry *entry;
905 	unsigned long flags;
906 	unsigned int i, j, next = q->bufs_next;
907 	int bufs_queued = atomic_inc_return(&q->bufs_queued);
908 	u32 fbpt_rp;
909 
910 	dev_dbg(&cio2->pci_dev->dev, "queue buffer %d\n", vb->index);
911 
912 	/*
913 	 * This code queues the buffer to the CIO2 DMA engine, which starts
914 	 * running once streaming has started. It is possible that this code
915 	 * gets pre-empted due to increased CPU load. Upon this, the driver
916 	 * does not get an opportunity to queue new buffers to the CIO2 DMA
917 	 * engine. When the DMA engine encounters an FBPT entry without the
918 	 * VALID bit set, the DMA engine halts, which requires a restart of
919 	 * the DMA engine and sensor, to continue streaming.
920 	 * This is not desired and is highly unlikely given that there are
921 	 * 32 FBPT entries that the DMA engine needs to process, to run into
922 	 * an FBPT entry, without the VALID bit set. We try to mitigate this
923 	 * by disabling interrupts for the duration of this queueing.
924 	 */
925 	local_irq_save(flags);
926 
927 	fbpt_rp = (readl(cio2->base + CIO2_REG_CDMARI(CIO2_DMA_CHAN))
928 		   >> CIO2_CDMARI_FBPT_RP_SHIFT)
929 		   & CIO2_CDMARI_FBPT_RP_MASK;
930 
931 	/*
932 	 * fbpt_rp is the fbpt entry that the dma is currently working
933 	 * on, but since it could jump to next entry at any time,
934 	 * assume that we might already be there.
935 	 */
936 	fbpt_rp = (fbpt_rp + 1) % CIO2_MAX_BUFFERS;
937 
938 	if (bufs_queued <= 1 || fbpt_rp == next)
939 		/* Buffers were drained */
940 		next = (fbpt_rp + 1) % CIO2_MAX_BUFFERS;
941 
942 	for (i = 0; i < CIO2_MAX_BUFFERS; i++) {
943 		/*
944 		 * We have allocated CIO2_MAX_BUFFERS circularly for the
945 		 * hw, the user has requested N buffer queue. The driver
946 		 * ensures N <= CIO2_MAX_BUFFERS and guarantees that whenever
947 		 * user queues a buffer, there necessarily is a free buffer.
948 		 */
949 		if (!q->bufs[next]) {
950 			q->bufs[next] = b;
951 			entry = &q->fbpt[next * CIO2_MAX_LOPS];
952 			cio2_fbpt_entry_init_buf(cio2, b, entry);
953 			local_irq_restore(flags);
954 			q->bufs_next = (next + 1) % CIO2_MAX_BUFFERS;
955 			for (j = 0; j < vb->num_planes; j++)
956 				vb2_set_plane_payload(vb, j,
957 					q->format.plane_fmt[j].sizeimage);
958 			return;
959 		}
960 
961 		dev_dbg(&cio2->pci_dev->dev, "entry %i was full!\n", next);
962 		next = (next + 1) % CIO2_MAX_BUFFERS;
963 	}
964 
965 	local_irq_restore(flags);
966 	dev_err(&cio2->pci_dev->dev, "error: all cio2 entries were full!\n");
967 	atomic_dec(&q->bufs_queued);
968 	vb2_buffer_done(vb, VB2_BUF_STATE_ERROR);
969 }
970 
971 /* Called when each buffer is freed */
972 static void cio2_vb2_buf_cleanup(struct vb2_buffer *vb)
973 {
974 	struct cio2_device *cio2 = vb2_get_drv_priv(vb->vb2_queue);
975 	struct cio2_buffer *b =
976 		container_of(vb, struct cio2_buffer, vbb.vb2_buf);
977 	unsigned int i;
978 
979 	/* Free LOP table */
980 	for (i = 0; i < CIO2_MAX_LOPS; i++) {
981 		if (b->lop[i])
982 			dma_free_coherent(&cio2->pci_dev->dev, CIO2_PAGE_SIZE,
983 					  b->lop[i], b->lop_bus_addr[i]);
984 	}
985 }
986 
987 static int cio2_vb2_start_streaming(struct vb2_queue *vq, unsigned int count)
988 {
989 	struct cio2_queue *q = vb2q_to_cio2_queue(vq);
990 	struct cio2_device *cio2 = vb2_get_drv_priv(vq);
991 	int r;
992 
993 	cio2->cur_queue = q;
994 	atomic_set(&q->frame_sequence, 0);
995 
996 	r = pm_runtime_get_sync(&cio2->pci_dev->dev);
997 	if (r < 0) {
998 		dev_info(&cio2->pci_dev->dev, "failed to set power %d\n", r);
999 		pm_runtime_put_noidle(&cio2->pci_dev->dev);
1000 		return r;
1001 	}
1002 
1003 	r = media_pipeline_start(&q->vdev.entity, &q->pipe);
1004 	if (r)
1005 		goto fail_pipeline;
1006 
1007 	r = cio2_hw_init(cio2, q);
1008 	if (r)
1009 		goto fail_hw;
1010 
1011 	/* Start streaming on sensor */
1012 	r = v4l2_subdev_call(q->sensor, video, s_stream, 1);
1013 	if (r)
1014 		goto fail_csi2_subdev;
1015 
1016 	cio2->streaming = true;
1017 
1018 	return 0;
1019 
1020 fail_csi2_subdev:
1021 	cio2_hw_exit(cio2, q);
1022 fail_hw:
1023 	media_pipeline_stop(&q->vdev.entity);
1024 fail_pipeline:
1025 	dev_dbg(&cio2->pci_dev->dev, "failed to start streaming (%d)\n", r);
1026 	cio2_vb2_return_all_buffers(q, VB2_BUF_STATE_QUEUED);
1027 	pm_runtime_put(&cio2->pci_dev->dev);
1028 
1029 	return r;
1030 }
1031 
1032 static void cio2_vb2_stop_streaming(struct vb2_queue *vq)
1033 {
1034 	struct cio2_queue *q = vb2q_to_cio2_queue(vq);
1035 	struct cio2_device *cio2 = vb2_get_drv_priv(vq);
1036 
1037 	if (v4l2_subdev_call(q->sensor, video, s_stream, 0))
1038 		dev_err(&cio2->pci_dev->dev,
1039 			"failed to stop sensor streaming\n");
1040 
1041 	cio2_hw_exit(cio2, q);
1042 	synchronize_irq(cio2->pci_dev->irq);
1043 	cio2_vb2_return_all_buffers(q, VB2_BUF_STATE_ERROR);
1044 	media_pipeline_stop(&q->vdev.entity);
1045 	pm_runtime_put(&cio2->pci_dev->dev);
1046 	cio2->streaming = false;
1047 }
1048 
1049 static const struct vb2_ops cio2_vb2_ops = {
1050 	.buf_init = cio2_vb2_buf_init,
1051 	.buf_queue = cio2_vb2_buf_queue,
1052 	.buf_cleanup = cio2_vb2_buf_cleanup,
1053 	.queue_setup = cio2_vb2_queue_setup,
1054 	.start_streaming = cio2_vb2_start_streaming,
1055 	.stop_streaming = cio2_vb2_stop_streaming,
1056 	.wait_prepare = vb2_ops_wait_prepare,
1057 	.wait_finish = vb2_ops_wait_finish,
1058 };
1059 
1060 /**************** V4L2 interface ****************/
1061 
1062 static int cio2_v4l2_querycap(struct file *file, void *fh,
1063 			      struct v4l2_capability *cap)
1064 {
1065 	struct cio2_device *cio2 = video_drvdata(file);
1066 
1067 	strscpy(cap->driver, CIO2_NAME, sizeof(cap->driver));
1068 	strscpy(cap->card, CIO2_DEVICE_NAME, sizeof(cap->card));
1069 	snprintf(cap->bus_info, sizeof(cap->bus_info),
1070 		 "PCI:%s", pci_name(cio2->pci_dev));
1071 
1072 	return 0;
1073 }
1074 
1075 static int cio2_v4l2_enum_fmt(struct file *file, void *fh,
1076 			      struct v4l2_fmtdesc *f)
1077 {
1078 	if (f->index >= ARRAY_SIZE(formats))
1079 		return -EINVAL;
1080 
1081 	f->pixelformat = formats[f->index].fourcc;
1082 
1083 	return 0;
1084 }
1085 
1086 /* The format is validated in cio2_video_link_validate() */
1087 static int cio2_v4l2_g_fmt(struct file *file, void *fh, struct v4l2_format *f)
1088 {
1089 	struct cio2_queue *q = file_to_cio2_queue(file);
1090 
1091 	f->fmt.pix_mp = q->format;
1092 
1093 	return 0;
1094 }
1095 
1096 static int cio2_v4l2_try_fmt(struct file *file, void *fh, struct v4l2_format *f)
1097 {
1098 	const struct ipu3_cio2_fmt *fmt;
1099 	struct v4l2_pix_format_mplane *mpix = &f->fmt.pix_mp;
1100 
1101 	fmt = cio2_find_format(&mpix->pixelformat, NULL);
1102 	if (!fmt)
1103 		fmt = &formats[0];
1104 
1105 	/* Only supports up to 4224x3136 */
1106 	if (mpix->width > CIO2_IMAGE_MAX_WIDTH)
1107 		mpix->width = CIO2_IMAGE_MAX_WIDTH;
1108 	if (mpix->height > CIO2_IMAGE_MAX_LENGTH)
1109 		mpix->height = CIO2_IMAGE_MAX_LENGTH;
1110 
1111 	mpix->num_planes = 1;
1112 	mpix->pixelformat = fmt->fourcc;
1113 	mpix->colorspace = V4L2_COLORSPACE_RAW;
1114 	mpix->field = V4L2_FIELD_NONE;
1115 	memset(mpix->reserved, 0, sizeof(mpix->reserved));
1116 	mpix->plane_fmt[0].bytesperline = cio2_bytesperline(mpix->width);
1117 	mpix->plane_fmt[0].sizeimage = mpix->plane_fmt[0].bytesperline *
1118 							mpix->height;
1119 	memset(mpix->plane_fmt[0].reserved, 0,
1120 	       sizeof(mpix->plane_fmt[0].reserved));
1121 
1122 	/* use default */
1123 	mpix->ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
1124 	mpix->quantization = V4L2_QUANTIZATION_DEFAULT;
1125 	mpix->xfer_func = V4L2_XFER_FUNC_DEFAULT;
1126 
1127 	return 0;
1128 }
1129 
1130 static int cio2_v4l2_s_fmt(struct file *file, void *fh, struct v4l2_format *f)
1131 {
1132 	struct cio2_queue *q = file_to_cio2_queue(file);
1133 
1134 	cio2_v4l2_try_fmt(file, fh, f);
1135 	q->format = f->fmt.pix_mp;
1136 
1137 	return 0;
1138 }
1139 
1140 static int
1141 cio2_video_enum_input(struct file *file, void *fh, struct v4l2_input *input)
1142 {
1143 	if (input->index > 0)
1144 		return -EINVAL;
1145 
1146 	strscpy(input->name, "camera", sizeof(input->name));
1147 	input->type = V4L2_INPUT_TYPE_CAMERA;
1148 
1149 	return 0;
1150 }
1151 
1152 static int
1153 cio2_video_g_input(struct file *file, void *fh, unsigned int *input)
1154 {
1155 	*input = 0;
1156 
1157 	return 0;
1158 }
1159 
1160 static int
1161 cio2_video_s_input(struct file *file, void *fh, unsigned int input)
1162 {
1163 	return input == 0 ? 0 : -EINVAL;
1164 }
1165 
1166 static const struct v4l2_file_operations cio2_v4l2_fops = {
1167 	.owner = THIS_MODULE,
1168 	.unlocked_ioctl = video_ioctl2,
1169 	.open = v4l2_fh_open,
1170 	.release = vb2_fop_release,
1171 	.poll = vb2_fop_poll,
1172 	.mmap = vb2_fop_mmap,
1173 };
1174 
1175 static const struct v4l2_ioctl_ops cio2_v4l2_ioctl_ops = {
1176 	.vidioc_querycap = cio2_v4l2_querycap,
1177 	.vidioc_enum_fmt_vid_cap = cio2_v4l2_enum_fmt,
1178 	.vidioc_g_fmt_vid_cap_mplane = cio2_v4l2_g_fmt,
1179 	.vidioc_s_fmt_vid_cap_mplane = cio2_v4l2_s_fmt,
1180 	.vidioc_try_fmt_vid_cap_mplane = cio2_v4l2_try_fmt,
1181 	.vidioc_reqbufs = vb2_ioctl_reqbufs,
1182 	.vidioc_create_bufs = vb2_ioctl_create_bufs,
1183 	.vidioc_prepare_buf = vb2_ioctl_prepare_buf,
1184 	.vidioc_querybuf = vb2_ioctl_querybuf,
1185 	.vidioc_qbuf = vb2_ioctl_qbuf,
1186 	.vidioc_dqbuf = vb2_ioctl_dqbuf,
1187 	.vidioc_streamon = vb2_ioctl_streamon,
1188 	.vidioc_streamoff = vb2_ioctl_streamoff,
1189 	.vidioc_expbuf = vb2_ioctl_expbuf,
1190 	.vidioc_enum_input = cio2_video_enum_input,
1191 	.vidioc_g_input	= cio2_video_g_input,
1192 	.vidioc_s_input	= cio2_video_s_input,
1193 };
1194 
1195 static int cio2_subdev_subscribe_event(struct v4l2_subdev *sd,
1196 				       struct v4l2_fh *fh,
1197 				       struct v4l2_event_subscription *sub)
1198 {
1199 	if (sub->type != V4L2_EVENT_FRAME_SYNC)
1200 		return -EINVAL;
1201 
1202 	/* Line number. For now only zero accepted. */
1203 	if (sub->id != 0)
1204 		return -EINVAL;
1205 
1206 	return v4l2_event_subscribe(fh, sub, 0, NULL);
1207 }
1208 
1209 static int cio2_subdev_open(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh)
1210 {
1211 	struct v4l2_mbus_framefmt *format;
1212 	const struct v4l2_mbus_framefmt fmt_default = {
1213 		.width = 1936,
1214 		.height = 1096,
1215 		.code = formats[0].mbus_code,
1216 		.field = V4L2_FIELD_NONE,
1217 		.colorspace = V4L2_COLORSPACE_RAW,
1218 		.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT,
1219 		.quantization = V4L2_QUANTIZATION_DEFAULT,
1220 		.xfer_func = V4L2_XFER_FUNC_DEFAULT,
1221 	};
1222 
1223 	/* Initialize try_fmt */
1224 	format = v4l2_subdev_get_try_format(sd, fh->pad, CIO2_PAD_SINK);
1225 	*format = fmt_default;
1226 
1227 	/* same as sink */
1228 	format = v4l2_subdev_get_try_format(sd, fh->pad, CIO2_PAD_SOURCE);
1229 	*format = fmt_default;
1230 
1231 	return 0;
1232 }
1233 
1234 /*
1235  * cio2_subdev_get_fmt - Handle get format by pads subdev method
1236  * @sd : pointer to v4l2 subdev structure
1237  * @cfg: V4L2 subdev pad config
1238  * @fmt: pointer to v4l2 subdev format structure
1239  * return -EINVAL or zero on success
1240  */
1241 static int cio2_subdev_get_fmt(struct v4l2_subdev *sd,
1242 			       struct v4l2_subdev_pad_config *cfg,
1243 			       struct v4l2_subdev_format *fmt)
1244 {
1245 	struct cio2_queue *q = container_of(sd, struct cio2_queue, subdev);
1246 	struct v4l2_subdev_format format;
1247 	int ret;
1248 
1249 	if (fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
1250 		fmt->format = *v4l2_subdev_get_try_format(sd, cfg, fmt->pad);
1251 		return 0;
1252 	}
1253 
1254 	if (fmt->pad == CIO2_PAD_SINK) {
1255 		format.which = V4L2_SUBDEV_FORMAT_ACTIVE;
1256 		ret = v4l2_subdev_call(sd, pad, get_fmt, NULL,
1257 				       &format);
1258 
1259 		if (ret)
1260 			return ret;
1261 		/* update colorspace etc */
1262 		q->subdev_fmt.colorspace = format.format.colorspace;
1263 		q->subdev_fmt.ycbcr_enc = format.format.ycbcr_enc;
1264 		q->subdev_fmt.quantization = format.format.quantization;
1265 		q->subdev_fmt.xfer_func = format.format.xfer_func;
1266 	}
1267 
1268 	fmt->format = q->subdev_fmt;
1269 
1270 	return 0;
1271 }
1272 
1273 /*
1274  * cio2_subdev_set_fmt - Handle set format by pads subdev method
1275  * @sd : pointer to v4l2 subdev structure
1276  * @cfg: V4L2 subdev pad config
1277  * @fmt: pointer to v4l2 subdev format structure
1278  * return -EINVAL or zero on success
1279  */
1280 static int cio2_subdev_set_fmt(struct v4l2_subdev *sd,
1281 			       struct v4l2_subdev_pad_config *cfg,
1282 			       struct v4l2_subdev_format *fmt)
1283 {
1284 	struct cio2_queue *q = container_of(sd, struct cio2_queue, subdev);
1285 
1286 	/*
1287 	 * Only allow setting sink pad format;
1288 	 * source always propagates from sink
1289 	 */
1290 	if (fmt->pad == CIO2_PAD_SOURCE)
1291 		return cio2_subdev_get_fmt(sd, cfg, fmt);
1292 
1293 	if (fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
1294 		*v4l2_subdev_get_try_format(sd, cfg, fmt->pad) = fmt->format;
1295 	} else {
1296 		/* It's the sink, allow changing frame size */
1297 		q->subdev_fmt.width = fmt->format.width;
1298 		q->subdev_fmt.height = fmt->format.height;
1299 		q->subdev_fmt.code = fmt->format.code;
1300 		fmt->format = q->subdev_fmt;
1301 	}
1302 
1303 	return 0;
1304 }
1305 
1306 static int cio2_subdev_enum_mbus_code(struct v4l2_subdev *sd,
1307 				      struct v4l2_subdev_pad_config *cfg,
1308 				      struct v4l2_subdev_mbus_code_enum *code)
1309 {
1310 	if (code->index >= ARRAY_SIZE(formats))
1311 		return -EINVAL;
1312 
1313 	code->code = formats[code->index].mbus_code;
1314 	return 0;
1315 }
1316 
1317 static int cio2_subdev_link_validate_get_format(struct media_pad *pad,
1318 						struct v4l2_subdev_format *fmt)
1319 {
1320 	if (is_media_entity_v4l2_subdev(pad->entity)) {
1321 		struct v4l2_subdev *sd =
1322 			media_entity_to_v4l2_subdev(pad->entity);
1323 
1324 		fmt->which = V4L2_SUBDEV_FORMAT_ACTIVE;
1325 		fmt->pad = pad->index;
1326 		return v4l2_subdev_call(sd, pad, get_fmt, NULL, fmt);
1327 	}
1328 
1329 	return -EINVAL;
1330 }
1331 
1332 static int cio2_video_link_validate(struct media_link *link)
1333 {
1334 	struct video_device *vd = container_of(link->sink->entity,
1335 						struct video_device, entity);
1336 	struct cio2_queue *q = container_of(vd, struct cio2_queue, vdev);
1337 	struct cio2_device *cio2 = video_get_drvdata(vd);
1338 	struct v4l2_subdev_format source_fmt;
1339 	int ret;
1340 
1341 	if (!media_entity_remote_pad(link->sink->entity->pads)) {
1342 		dev_info(&cio2->pci_dev->dev,
1343 			 "video node %s pad not connected\n", vd->name);
1344 		return -ENOTCONN;
1345 	}
1346 
1347 	ret = cio2_subdev_link_validate_get_format(link->source, &source_fmt);
1348 	if (ret < 0)
1349 		return 0;
1350 
1351 	if (source_fmt.format.width != q->format.width ||
1352 	    source_fmt.format.height != q->format.height) {
1353 		dev_err(&cio2->pci_dev->dev,
1354 			"Wrong width or height %ux%u (%ux%u expected)\n",
1355 			q->format.width, q->format.height,
1356 			source_fmt.format.width, source_fmt.format.height);
1357 		return -EINVAL;
1358 	}
1359 
1360 	if (!cio2_find_format(&q->format.pixelformat, &source_fmt.format.code))
1361 		return -EINVAL;
1362 
1363 	return 0;
1364 }
1365 
1366 static const struct v4l2_subdev_core_ops cio2_subdev_core_ops = {
1367 	.subscribe_event = cio2_subdev_subscribe_event,
1368 	.unsubscribe_event = v4l2_event_subdev_unsubscribe,
1369 };
1370 
1371 static const struct v4l2_subdev_internal_ops cio2_subdev_internal_ops = {
1372 	.open = cio2_subdev_open,
1373 };
1374 
1375 static const struct v4l2_subdev_pad_ops cio2_subdev_pad_ops = {
1376 	.link_validate = v4l2_subdev_link_validate_default,
1377 	.get_fmt = cio2_subdev_get_fmt,
1378 	.set_fmt = cio2_subdev_set_fmt,
1379 	.enum_mbus_code = cio2_subdev_enum_mbus_code,
1380 };
1381 
1382 static const struct v4l2_subdev_ops cio2_subdev_ops = {
1383 	.core = &cio2_subdev_core_ops,
1384 	.pad = &cio2_subdev_pad_ops,
1385 };
1386 
1387 /******* V4L2 sub-device asynchronous registration callbacks***********/
1388 
1389 struct sensor_async_subdev {
1390 	struct v4l2_async_subdev asd;
1391 	struct csi2_bus_info csi2;
1392 };
1393 
1394 /* The .bound() notifier callback when a match is found */
1395 static int cio2_notifier_bound(struct v4l2_async_notifier *notifier,
1396 			       struct v4l2_subdev *sd,
1397 			       struct v4l2_async_subdev *asd)
1398 {
1399 	struct cio2_device *cio2 = container_of(notifier,
1400 					struct cio2_device, notifier);
1401 	struct sensor_async_subdev *s_asd = container_of(asd,
1402 					struct sensor_async_subdev, asd);
1403 	struct cio2_queue *q;
1404 
1405 	if (cio2->queue[s_asd->csi2.port].sensor)
1406 		return -EBUSY;
1407 
1408 	q = &cio2->queue[s_asd->csi2.port];
1409 
1410 	q->csi2 = s_asd->csi2;
1411 	q->sensor = sd;
1412 	q->csi_rx_base = cio2->base + CIO2_REG_PIPE_BASE(q->csi2.port);
1413 
1414 	return 0;
1415 }
1416 
1417 /* The .unbind callback */
1418 static void cio2_notifier_unbind(struct v4l2_async_notifier *notifier,
1419 				 struct v4l2_subdev *sd,
1420 				 struct v4l2_async_subdev *asd)
1421 {
1422 	struct cio2_device *cio2 = container_of(notifier,
1423 						struct cio2_device, notifier);
1424 	struct sensor_async_subdev *s_asd = container_of(asd,
1425 					struct sensor_async_subdev, asd);
1426 
1427 	cio2->queue[s_asd->csi2.port].sensor = NULL;
1428 }
1429 
1430 /* .complete() is called after all subdevices have been located */
1431 static int cio2_notifier_complete(struct v4l2_async_notifier *notifier)
1432 {
1433 	struct cio2_device *cio2 = container_of(notifier, struct cio2_device,
1434 						notifier);
1435 	struct sensor_async_subdev *s_asd;
1436 	struct v4l2_async_subdev *asd;
1437 	struct cio2_queue *q;
1438 	unsigned int pad;
1439 	int ret;
1440 
1441 	list_for_each_entry(asd, &cio2->notifier.asd_list, asd_list) {
1442 		s_asd = container_of(asd, struct sensor_async_subdev, asd);
1443 		q = &cio2->queue[s_asd->csi2.port];
1444 
1445 		for (pad = 0; pad < q->sensor->entity.num_pads; pad++)
1446 			if (q->sensor->entity.pads[pad].flags &
1447 						MEDIA_PAD_FL_SOURCE)
1448 				break;
1449 
1450 		if (pad == q->sensor->entity.num_pads) {
1451 			dev_err(&cio2->pci_dev->dev,
1452 				"failed to find src pad for %s\n",
1453 				q->sensor->name);
1454 			return -ENXIO;
1455 		}
1456 
1457 		ret = media_create_pad_link(
1458 				&q->sensor->entity, pad,
1459 				&q->subdev.entity, CIO2_PAD_SINK,
1460 				0);
1461 		if (ret) {
1462 			dev_err(&cio2->pci_dev->dev,
1463 				"failed to create link for %s\n",
1464 				q->sensor->name);
1465 			return ret;
1466 		}
1467 	}
1468 
1469 	return v4l2_device_register_subdev_nodes(&cio2->v4l2_dev);
1470 }
1471 
1472 static const struct v4l2_async_notifier_operations cio2_async_ops = {
1473 	.bound = cio2_notifier_bound,
1474 	.unbind = cio2_notifier_unbind,
1475 	.complete = cio2_notifier_complete,
1476 };
1477 
1478 static int cio2_fwnode_parse(struct device *dev,
1479 			     struct v4l2_fwnode_endpoint *vep,
1480 			     struct v4l2_async_subdev *asd)
1481 {
1482 	struct sensor_async_subdev *s_asd =
1483 			container_of(asd, struct sensor_async_subdev, asd);
1484 
1485 	if (vep->bus_type != V4L2_MBUS_CSI2_DPHY) {
1486 		dev_err(dev, "Only CSI2 bus type is currently supported\n");
1487 		return -EINVAL;
1488 	}
1489 
1490 	s_asd->csi2.port = vep->base.port;
1491 	s_asd->csi2.lanes = vep->bus.mipi_csi2.num_data_lanes;
1492 
1493 	return 0;
1494 }
1495 
1496 static int cio2_notifier_init(struct cio2_device *cio2)
1497 {
1498 	int ret;
1499 
1500 	v4l2_async_notifier_init(&cio2->notifier);
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 (list_empty(&cio2->notifier.asd_list))
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 	subdev->entity.function = MEDIA_ENT_F_VID_IF_BRIDGE;
1605 	v4l2_set_subdevdata(subdev, cio2);
1606 	r = v4l2_device_register_subdev(&cio2->v4l2_dev, subdev);
1607 	if (r) {
1608 		dev_err(&cio2->pci_dev->dev,
1609 			"failed initialize subdev (%d)\n", r);
1610 		goto fail_subdev;
1611 	}
1612 
1613 	/* Initialize vbq */
1614 	vbq->type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
1615 	vbq->io_modes = VB2_USERPTR | VB2_MMAP | VB2_DMABUF;
1616 	vbq->ops = &cio2_vb2_ops;
1617 	vbq->mem_ops = &vb2_dma_sg_memops;
1618 	vbq->buf_struct_size = sizeof(struct cio2_buffer);
1619 	vbq->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_MONOTONIC;
1620 	vbq->min_buffers_needed = 1;
1621 	vbq->drv_priv = cio2;
1622 	vbq->lock = &q->lock;
1623 	r = vb2_queue_init(vbq);
1624 	if (r) {
1625 		dev_err(&cio2->pci_dev->dev,
1626 			"failed to initialize videobuf2 queue (%d)\n", r);
1627 		goto fail_vbq;
1628 	}
1629 
1630 	/* Initialize vdev */
1631 	snprintf(vdev->name, sizeof(vdev->name),
1632 		 "%s %td", CIO2_NAME, q - cio2->queue);
1633 	vdev->release = video_device_release_empty;
1634 	vdev->fops = &cio2_v4l2_fops;
1635 	vdev->ioctl_ops = &cio2_v4l2_ioctl_ops;
1636 	vdev->lock = &cio2->lock;
1637 	vdev->v4l2_dev = &cio2->v4l2_dev;
1638 	vdev->queue = &q->vbq;
1639 	vdev->device_caps = V4L2_CAP_VIDEO_CAPTURE_MPLANE | V4L2_CAP_STREAMING;
1640 	video_set_drvdata(vdev, cio2);
1641 	r = video_register_device(vdev, VFL_TYPE_GRABBER, -1);
1642 	if (r) {
1643 		dev_err(&cio2->pci_dev->dev,
1644 			"failed to register video device (%d)\n", r);
1645 		goto fail_vdev;
1646 	}
1647 
1648 	/* Create link from CIO2 subdev to output node */
1649 	r = media_create_pad_link(
1650 		&subdev->entity, CIO2_PAD_SOURCE, &vdev->entity, 0,
1651 		MEDIA_LNK_FL_ENABLED | MEDIA_LNK_FL_IMMUTABLE);
1652 	if (r)
1653 		goto fail_link;
1654 
1655 	return 0;
1656 
1657 fail_link:
1658 	video_unregister_device(&q->vdev);
1659 fail_vdev:
1660 	vb2_queue_release(vbq);
1661 fail_vbq:
1662 	v4l2_device_unregister_subdev(subdev);
1663 fail_subdev:
1664 	media_entity_cleanup(&vdev->entity);
1665 fail_vdev_media_entity:
1666 	media_entity_cleanup(&subdev->entity);
1667 fail_subdev_media_entity:
1668 	cio2_fbpt_exit(q, &cio2->pci_dev->dev);
1669 fail_fbpt:
1670 	mutex_destroy(&q->lock);
1671 
1672 	return r;
1673 }
1674 
1675 static void cio2_queue_exit(struct cio2_device *cio2, struct cio2_queue *q)
1676 {
1677 	video_unregister_device(&q->vdev);
1678 	media_entity_cleanup(&q->vdev.entity);
1679 	vb2_queue_release(&q->vbq);
1680 	v4l2_device_unregister_subdev(&q->subdev);
1681 	media_entity_cleanup(&q->subdev.entity);
1682 	cio2_fbpt_exit(q, &cio2->pci_dev->dev);
1683 	mutex_destroy(&q->lock);
1684 }
1685 
1686 static int cio2_queues_init(struct cio2_device *cio2)
1687 {
1688 	int i, r;
1689 
1690 	for (i = 0; i < CIO2_QUEUES; i++) {
1691 		r = cio2_queue_init(cio2, &cio2->queue[i]);
1692 		if (r)
1693 			break;
1694 	}
1695 
1696 	if (i == CIO2_QUEUES)
1697 		return 0;
1698 
1699 	for (i--; i >= 0; i--)
1700 		cio2_queue_exit(cio2, &cio2->queue[i]);
1701 
1702 	return r;
1703 }
1704 
1705 static void cio2_queues_exit(struct cio2_device *cio2)
1706 {
1707 	unsigned int i;
1708 
1709 	for (i = 0; i < CIO2_QUEUES; i++)
1710 		cio2_queue_exit(cio2, &cio2->queue[i]);
1711 }
1712 
1713 /**************** PCI interface ****************/
1714 
1715 static int cio2_pci_config_setup(struct pci_dev *dev)
1716 {
1717 	u16 pci_command;
1718 	int r = pci_enable_msi(dev);
1719 
1720 	if (r) {
1721 		dev_err(&dev->dev, "failed to enable MSI (%d)\n", r);
1722 		return r;
1723 	}
1724 
1725 	pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1726 	pci_command |= PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
1727 		PCI_COMMAND_INTX_DISABLE;
1728 	pci_write_config_word(dev, PCI_COMMAND, pci_command);
1729 
1730 	return 0;
1731 }
1732 
1733 static int cio2_pci_probe(struct pci_dev *pci_dev,
1734 			  const struct pci_device_id *id)
1735 {
1736 	struct cio2_device *cio2;
1737 	void __iomem *const *iomap;
1738 	int r;
1739 
1740 	cio2 = devm_kzalloc(&pci_dev->dev, sizeof(*cio2), GFP_KERNEL);
1741 	if (!cio2)
1742 		return -ENOMEM;
1743 	cio2->pci_dev = pci_dev;
1744 
1745 	r = pcim_enable_device(pci_dev);
1746 	if (r) {
1747 		dev_err(&pci_dev->dev, "failed to enable device (%d)\n", r);
1748 		return r;
1749 	}
1750 
1751 	dev_info(&pci_dev->dev, "device 0x%x (rev: 0x%x)\n",
1752 		 pci_dev->device, pci_dev->revision);
1753 
1754 	r = pcim_iomap_regions(pci_dev, 1 << CIO2_PCI_BAR, pci_name(pci_dev));
1755 	if (r) {
1756 		dev_err(&pci_dev->dev, "failed to remap I/O memory (%d)\n", r);
1757 		return -ENODEV;
1758 	}
1759 
1760 	iomap = pcim_iomap_table(pci_dev);
1761 	if (!iomap) {
1762 		dev_err(&pci_dev->dev, "failed to iomap table\n");
1763 		return -ENODEV;
1764 	}
1765 
1766 	cio2->base = iomap[CIO2_PCI_BAR];
1767 
1768 	pci_set_drvdata(pci_dev, cio2);
1769 
1770 	pci_set_master(pci_dev);
1771 
1772 	r = pci_set_dma_mask(pci_dev, CIO2_DMA_MASK);
1773 	if (r) {
1774 		dev_err(&pci_dev->dev, "failed to set DMA mask (%d)\n", r);
1775 		return -ENODEV;
1776 	}
1777 
1778 	r = cio2_pci_config_setup(pci_dev);
1779 	if (r)
1780 		return -ENODEV;
1781 
1782 	r = cio2_fbpt_init_dummy(cio2);
1783 	if (r)
1784 		return r;
1785 
1786 	mutex_init(&cio2->lock);
1787 
1788 	cio2->media_dev.dev = &cio2->pci_dev->dev;
1789 	strscpy(cio2->media_dev.model, CIO2_DEVICE_NAME,
1790 		sizeof(cio2->media_dev.model));
1791 	snprintf(cio2->media_dev.bus_info, sizeof(cio2->media_dev.bus_info),
1792 		 "PCI:%s", pci_name(cio2->pci_dev));
1793 	cio2->media_dev.hw_revision = 0;
1794 
1795 	media_device_init(&cio2->media_dev);
1796 	r = media_device_register(&cio2->media_dev);
1797 	if (r < 0)
1798 		goto fail_mutex_destroy;
1799 
1800 	cio2->v4l2_dev.mdev = &cio2->media_dev;
1801 	r = v4l2_device_register(&pci_dev->dev, &cio2->v4l2_dev);
1802 	if (r) {
1803 		dev_err(&pci_dev->dev,
1804 			"failed to register V4L2 device (%d)\n", r);
1805 		goto fail_media_device_unregister;
1806 	}
1807 
1808 	r = cio2_queues_init(cio2);
1809 	if (r)
1810 		goto fail_v4l2_device_unregister;
1811 
1812 	/* Register notifier for subdevices we care */
1813 	r = cio2_notifier_init(cio2);
1814 	/* Proceed without sensors connected to allow the device to suspend. */
1815 	if (r && r != -ENODEV)
1816 		goto fail_cio2_queue_exit;
1817 
1818 	r = devm_request_irq(&pci_dev->dev, pci_dev->irq, cio2_irq,
1819 			     IRQF_SHARED, CIO2_NAME, cio2);
1820 	if (r) {
1821 		dev_err(&pci_dev->dev, "failed to request IRQ (%d)\n", r);
1822 		goto fail;
1823 	}
1824 
1825 	pm_runtime_put_noidle(&pci_dev->dev);
1826 	pm_runtime_allow(&pci_dev->dev);
1827 
1828 	return 0;
1829 
1830 fail:
1831 	cio2_notifier_exit(cio2);
1832 fail_cio2_queue_exit:
1833 	cio2_queues_exit(cio2);
1834 fail_v4l2_device_unregister:
1835 	v4l2_device_unregister(&cio2->v4l2_dev);
1836 fail_media_device_unregister:
1837 	media_device_unregister(&cio2->media_dev);
1838 	media_device_cleanup(&cio2->media_dev);
1839 fail_mutex_destroy:
1840 	mutex_destroy(&cio2->lock);
1841 	cio2_fbpt_exit_dummy(cio2);
1842 
1843 	return r;
1844 }
1845 
1846 static void cio2_pci_remove(struct pci_dev *pci_dev)
1847 {
1848 	struct cio2_device *cio2 = pci_get_drvdata(pci_dev);
1849 
1850 	media_device_unregister(&cio2->media_dev);
1851 	cio2_notifier_exit(cio2);
1852 	cio2_queues_exit(cio2);
1853 	cio2_fbpt_exit_dummy(cio2);
1854 	v4l2_device_unregister(&cio2->v4l2_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");
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