xref: /linux/drivers/misc/bcm-vk/bcm_vk_dev.c (revision ff428d052b3b6fb22242d17c213c4898e5136323)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright 2018-2020 Broadcom.
4  */
5 
6 #include <linux/delay.h>
7 #include <linux/dma-mapping.h>
8 #include <linux/firmware.h>
9 #include <linux/fs.h>
10 #include <linux/idr.h>
11 #include <linux/kref.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/pci.h>
15 #include <linux/pci_regs.h>
16 #include <uapi/linux/misc/bcm_vk.h>
17 
18 #include "bcm_vk.h"
19 
20 #define PCI_DEVICE_ID_VALKYRIE	0x5e87
21 #define PCI_DEVICE_ID_VIPER	0x5e88
22 
23 static DEFINE_IDA(bcm_vk_ida);
24 
25 enum soc_idx {
26 	VALKYRIE_A0 = 0,
27 	VALKYRIE_B0,
28 	VIPER,
29 	VK_IDX_INVALID
30 };
31 
32 enum img_idx {
33 	IMG_PRI = 0,
34 	IMG_SEC,
35 	IMG_PER_TYPE_MAX
36 };
37 
38 struct load_image_entry {
39 	const u32 image_type;
40 	const char *image_name[IMG_PER_TYPE_MAX];
41 };
42 
43 #define NUM_BOOT_STAGES 2
44 /* default firmware images names */
45 static const struct load_image_entry image_tab[][NUM_BOOT_STAGES] = {
46 	[VALKYRIE_A0] = {
47 		{VK_IMAGE_TYPE_BOOT1, {"vk_a0-boot1.bin", "vk-boot1.bin"}},
48 		{VK_IMAGE_TYPE_BOOT2, {"vk_a0-boot2.bin", "vk-boot2.bin"}}
49 	},
50 	[VALKYRIE_B0] = {
51 		{VK_IMAGE_TYPE_BOOT1, {"vk_b0-boot1.bin", "vk-boot1.bin"}},
52 		{VK_IMAGE_TYPE_BOOT2, {"vk_b0-boot2.bin", "vk-boot2.bin"}}
53 	},
54 
55 	[VIPER] = {
56 		{VK_IMAGE_TYPE_BOOT1, {"vp-boot1.bin", ""}},
57 		{VK_IMAGE_TYPE_BOOT2, {"vp-boot2.bin", ""}}
58 	},
59 };
60 
61 /* Location of memory base addresses of interest in BAR1 */
62 /* Load Boot1 to start of ITCM */
63 #define BAR1_CODEPUSH_BASE_BOOT1	0x100000
64 
65 /* Allow minimum 1s for Load Image timeout responses */
66 #define LOAD_IMAGE_TIMEOUT_MS		(1 * MSEC_PER_SEC)
67 
68 /* Image startup timeouts */
69 #define BOOT1_STARTUP_TIMEOUT_MS	(5 * MSEC_PER_SEC)
70 #define BOOT2_STARTUP_TIMEOUT_MS	(10 * MSEC_PER_SEC)
71 
72 /* 1ms wait for checking the transfer complete status */
73 #define TXFR_COMPLETE_TIMEOUT_MS	1
74 
75 /* MSIX usages */
76 #define VK_MSIX_MSGQ_MAX		3
77 #define VK_MSIX_NOTF_MAX		1
78 #define VK_MSIX_TTY_MAX			BCM_VK_NUM_TTY
79 #define VK_MSIX_IRQ_MAX			(VK_MSIX_MSGQ_MAX + VK_MSIX_NOTF_MAX + \
80 					 VK_MSIX_TTY_MAX)
81 #define VK_MSIX_IRQ_MIN_REQ             (VK_MSIX_MSGQ_MAX + VK_MSIX_NOTF_MAX)
82 
83 /* Number of bits set in DMA mask*/
84 #define BCM_VK_DMA_BITS			64
85 
86 /* Ucode boot wait time */
87 #define BCM_VK_UCODE_BOOT_US            (100 * USEC_PER_MSEC)
88 /* 50% margin */
89 #define BCM_VK_UCODE_BOOT_MAX_US        ((BCM_VK_UCODE_BOOT_US * 3) >> 1)
90 
91 /* deinit time for the card os after receiving doorbell */
92 #define BCM_VK_DEINIT_TIME_MS		(2 * MSEC_PER_SEC)
93 
94 /*
95  * module parameters
96  */
97 static bool auto_load = true;
98 module_param(auto_load, bool, 0444);
99 MODULE_PARM_DESC(auto_load,
100 		 "Load images automatically at PCIe probe time.\n");
101 static uint nr_scratch_pages = VK_BAR1_SCRATCH_DEF_NR_PAGES;
102 module_param(nr_scratch_pages, uint, 0444);
103 MODULE_PARM_DESC(nr_scratch_pages,
104 		 "Number of pre allocated DMAable coherent pages.\n");
105 
106 static int bcm_vk_intf_ver_chk(struct bcm_vk *vk)
107 {
108 	struct device *dev = &vk->pdev->dev;
109 	u32 reg;
110 	u16 major, minor;
111 	int ret = 0;
112 
113 	/* read interface register */
114 	reg = vkread32(vk, BAR_0, BAR_INTF_VER);
115 	major = (reg >> BAR_INTF_VER_MAJOR_SHIFT) & BAR_INTF_VER_MASK;
116 	minor = reg & BAR_INTF_VER_MASK;
117 
118 	/*
119 	 * if major number is 0, it is pre-release and it would be allowed
120 	 * to continue, else, check versions accordingly
121 	 */
122 	if (!major) {
123 		dev_warn(dev, "Pre-release major.minor=%d.%d - drv %d.%d\n",
124 			 major, minor, SEMANTIC_MAJOR, SEMANTIC_MINOR);
125 	} else if (major != SEMANTIC_MAJOR) {
126 		dev_err(dev,
127 			"Intf major.minor=%d.%d rejected - drv %d.%d\n",
128 			major, minor, SEMANTIC_MAJOR, SEMANTIC_MINOR);
129 		ret = -EPFNOSUPPORT;
130 	} else {
131 		dev_dbg(dev,
132 			"Intf major.minor=%d.%d passed - drv %d.%d\n",
133 			major, minor, SEMANTIC_MAJOR, SEMANTIC_MINOR);
134 	}
135 	return ret;
136 }
137 
138 static inline int bcm_vk_wait(struct bcm_vk *vk, enum pci_barno bar,
139 			      u64 offset, u32 mask, u32 value,
140 			      unsigned long timeout_ms)
141 {
142 	struct device *dev = &vk->pdev->dev;
143 	unsigned long start_time;
144 	unsigned long timeout;
145 	u32 rd_val, boot_status;
146 
147 	start_time = jiffies;
148 	timeout = start_time + msecs_to_jiffies(timeout_ms);
149 
150 	do {
151 		rd_val = vkread32(vk, bar, offset);
152 		dev_dbg(dev, "BAR%d Offset=0x%llx: 0x%x\n",
153 			bar, offset, rd_val);
154 
155 		/* check for any boot err condition */
156 		boot_status = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
157 		if (boot_status & BOOT_ERR_MASK) {
158 			dev_err(dev, "Boot Err 0x%x, progress 0x%x after %d ms\n",
159 				(boot_status & BOOT_ERR_MASK) >> BOOT_ERR_SHIFT,
160 				boot_status & BOOT_PROG_MASK,
161 				jiffies_to_msecs(jiffies - start_time));
162 			return -EFAULT;
163 		}
164 
165 		if (time_after(jiffies, timeout))
166 			return -ETIMEDOUT;
167 
168 		cpu_relax();
169 		cond_resched();
170 	} while ((rd_val & mask) != value);
171 
172 	return 0;
173 }
174 
175 static void bcm_vk_get_card_info(struct bcm_vk *vk)
176 {
177 	struct device *dev = &vk->pdev->dev;
178 	u32 offset;
179 	int i;
180 	u8 *dst;
181 	struct bcm_vk_card_info *info = &vk->card_info;
182 
183 	/* first read the offset from spare register */
184 	offset = vkread32(vk, BAR_0, BAR_CARD_STATIC_INFO);
185 	offset &= (pci_resource_len(vk->pdev, BAR_2 * 2) - 1);
186 
187 	/* based on the offset, read info to internal card info structure */
188 	dst = (u8 *)info;
189 	for (i = 0; i < sizeof(*info); i++)
190 		*dst++ = vkread8(vk, BAR_2, offset++);
191 
192 #define CARD_INFO_LOG_FMT "version   : %x\n" \
193 			  "os_tag    : %s\n" \
194 			  "cmpt_tag  : %s\n" \
195 			  "cpu_freq  : %d MHz\n" \
196 			  "cpu_scale : %d full, %d lowest\n" \
197 			  "ddr_freq  : %d MHz\n" \
198 			  "ddr_size  : %d MB\n" \
199 			  "video_freq: %d MHz\n"
200 	dev_dbg(dev, CARD_INFO_LOG_FMT, info->version, info->os_tag,
201 		info->cmpt_tag, info->cpu_freq_mhz, info->cpu_scale[0],
202 		info->cpu_scale[MAX_OPP - 1], info->ddr_freq_mhz,
203 		info->ddr_size_MB, info->video_core_freq_mhz);
204 
205 	/*
206 	 * get the peer log pointer, only need the offset, and get record
207 	 * of the log buffer information which would be used for checking
208 	 * before dump, in case the BAR2 memory has been corrupted.
209 	 */
210 	vk->peerlog_off = offset;
211 	memcpy_fromio(&vk->peerlog_info, vk->bar[BAR_2] + vk->peerlog_off,
212 		      sizeof(vk->peerlog_info));
213 
214 	/*
215 	 * Do a range checking and if out of bound, the record will be zeroed
216 	 * which guarantees that nothing would be dumped.  In other words,
217 	 * peer dump is disabled.
218 	 */
219 	if ((vk->peerlog_info.buf_size > BCM_VK_PEER_LOG_BUF_MAX) ||
220 	    (vk->peerlog_info.mask != (vk->peerlog_info.buf_size - 1)) ||
221 	    (vk->peerlog_info.rd_idx > vk->peerlog_info.mask) ||
222 	    (vk->peerlog_info.wr_idx > vk->peerlog_info.mask)) {
223 		dev_err(dev, "Peer log disabled - range error: Size 0x%x(0x%x), [Rd Wr] = [%d %d]\n",
224 			vk->peerlog_info.buf_size,
225 			vk->peerlog_info.mask,
226 			vk->peerlog_info.rd_idx,
227 			vk->peerlog_info.wr_idx);
228 		memset(&vk->peerlog_info, 0, sizeof(vk->peerlog_info));
229 	} else {
230 		dev_dbg(dev, "Peer log: Size 0x%x(0x%x), [Rd Wr] = [%d %d]\n",
231 			vk->peerlog_info.buf_size,
232 			vk->peerlog_info.mask,
233 			vk->peerlog_info.rd_idx,
234 			vk->peerlog_info.wr_idx);
235 	}
236 }
237 
238 static void bcm_vk_get_proc_mon_info(struct bcm_vk *vk)
239 {
240 	struct device *dev = &vk->pdev->dev;
241 	struct bcm_vk_proc_mon_info *mon = &vk->proc_mon_info;
242 	u32 num, entry_size, offset, buf_size;
243 	u8 *dst;
244 
245 	/* calculate offset which is based on peerlog offset */
246 	buf_size = vkread32(vk, BAR_2,
247 			    vk->peerlog_off
248 			    + offsetof(struct bcm_vk_peer_log, buf_size));
249 	offset = vk->peerlog_off + sizeof(struct bcm_vk_peer_log)
250 		 + buf_size;
251 
252 	/* first read the num and entry size */
253 	num = vkread32(vk, BAR_2, offset);
254 	entry_size = vkread32(vk, BAR_2, offset + sizeof(num));
255 
256 	/* check for max allowed */
257 	if (num > BCM_VK_PROC_MON_MAX) {
258 		dev_err(dev, "Processing monitoring entry %d exceeds max %d\n",
259 			num, BCM_VK_PROC_MON_MAX);
260 		return;
261 	}
262 	mon->num = num;
263 	mon->entry_size = entry_size;
264 
265 	vk->proc_mon_off = offset;
266 
267 	/* read it once that will capture those static info */
268 	dst = (u8 *)&mon->entries[0];
269 	offset += sizeof(num) + sizeof(entry_size);
270 	memcpy_fromio(dst, vk->bar[BAR_2] + offset, num * entry_size);
271 }
272 
273 static int bcm_vk_sync_card_info(struct bcm_vk *vk)
274 {
275 	u32 rdy_marker = vkread32(vk, BAR_1, VK_BAR1_MSGQ_DEF_RDY);
276 
277 	/* check for marker, but allow diags mode to skip sync */
278 	if (!bcm_vk_msgq_marker_valid(vk))
279 		return (rdy_marker == VK_BAR1_DIAG_RDY_MARKER ? 0 : -EINVAL);
280 
281 	/*
282 	 * Write down scratch addr which is used for DMA. For
283 	 * signed part, BAR1 is accessible only after boot2 has come
284 	 * up
285 	 */
286 	if (vk->tdma_addr) {
287 		vkwrite32(vk, (u64)vk->tdma_addr >> 32, BAR_1,
288 			  VK_BAR1_SCRATCH_OFF_HI);
289 		vkwrite32(vk, (u32)vk->tdma_addr, BAR_1,
290 			  VK_BAR1_SCRATCH_OFF_LO);
291 		vkwrite32(vk, nr_scratch_pages * PAGE_SIZE, BAR_1,
292 			  VK_BAR1_SCRATCH_SZ_ADDR);
293 	}
294 
295 	/* get static card info, only need to read once */
296 	bcm_vk_get_card_info(vk);
297 
298 	/* get the proc mon info once */
299 	bcm_vk_get_proc_mon_info(vk);
300 
301 	return 0;
302 }
303 
304 static void bcm_vk_buf_notify(struct bcm_vk *vk, void *bufp,
305 			      dma_addr_t host_buf_addr, u32 buf_size)
306 {
307 	/* update the dma address to the card */
308 	vkwrite32(vk, (u64)host_buf_addr >> 32, BAR_1,
309 		  VK_BAR1_DMA_BUF_OFF_HI);
310 	vkwrite32(vk, (u32)host_buf_addr, BAR_1,
311 		  VK_BAR1_DMA_BUF_OFF_LO);
312 	vkwrite32(vk, buf_size, BAR_1, VK_BAR1_DMA_BUF_SZ);
313 }
314 
315 static int bcm_vk_load_image_by_type(struct bcm_vk *vk, u32 load_type,
316 				     const char *filename)
317 {
318 	struct device *dev = &vk->pdev->dev;
319 	const struct firmware *fw = NULL;
320 	void *bufp = NULL;
321 	size_t max_buf, offset;
322 	int ret;
323 	u64 offset_codepush;
324 	u32 codepush;
325 	u32 value;
326 	dma_addr_t boot_dma_addr;
327 	bool is_stdalone;
328 
329 	if (load_type == VK_IMAGE_TYPE_BOOT1) {
330 		/*
331 		 * After POR, enable VK soft BOOTSRC so bootrom do not clear
332 		 * the pushed image (the TCM memories).
333 		 */
334 		value = vkread32(vk, BAR_0, BAR_BOOTSRC_SELECT);
335 		value |= BOOTSRC_SOFT_ENABLE;
336 		vkwrite32(vk, value, BAR_0, BAR_BOOTSRC_SELECT);
337 
338 		codepush = CODEPUSH_BOOTSTART + CODEPUSH_BOOT1_ENTRY;
339 		offset_codepush = BAR_CODEPUSH_SBL;
340 
341 		/* Write a 1 to request SRAM open bit */
342 		vkwrite32(vk, CODEPUSH_BOOTSTART, BAR_0, offset_codepush);
343 
344 		/* Wait for VK to respond */
345 		ret = bcm_vk_wait(vk, BAR_0, BAR_BOOT_STATUS, SRAM_OPEN,
346 				  SRAM_OPEN, LOAD_IMAGE_TIMEOUT_MS);
347 		if (ret < 0) {
348 			dev_err(dev, "boot1 wait SRAM err - ret(%d)\n", ret);
349 			goto err_buf_out;
350 		}
351 
352 		max_buf = SZ_256K;
353 		bufp = dma_alloc_coherent(dev,
354 					  max_buf,
355 					  &boot_dma_addr, GFP_KERNEL);
356 		if (!bufp) {
357 			dev_err(dev, "Error allocating 0x%zx\n", max_buf);
358 			ret = -ENOMEM;
359 			goto err_buf_out;
360 		}
361 	} else if (load_type == VK_IMAGE_TYPE_BOOT2) {
362 		codepush = CODEPUSH_BOOT2_ENTRY;
363 		offset_codepush = BAR_CODEPUSH_SBI;
364 
365 		/* Wait for VK to respond */
366 		ret = bcm_vk_wait(vk, BAR_0, BAR_BOOT_STATUS, DDR_OPEN,
367 				  DDR_OPEN, LOAD_IMAGE_TIMEOUT_MS);
368 		if (ret < 0) {
369 			dev_err(dev, "boot2 wait DDR open error - ret(%d)\n",
370 				ret);
371 			goto err_buf_out;
372 		}
373 
374 		max_buf = SZ_4M;
375 		bufp = dma_alloc_coherent(dev,
376 					  max_buf,
377 					  &boot_dma_addr, GFP_KERNEL);
378 		if (!bufp) {
379 			dev_err(dev, "Error allocating 0x%zx\n", max_buf);
380 			ret = -ENOMEM;
381 			goto err_buf_out;
382 		}
383 
384 		bcm_vk_buf_notify(vk, bufp, boot_dma_addr, max_buf);
385 	} else {
386 		dev_err(dev, "Error invalid image type 0x%x\n", load_type);
387 		ret = -EINVAL;
388 		goto err_buf_out;
389 	}
390 
391 	offset = 0;
392 	ret = request_partial_firmware_into_buf(&fw, filename, dev,
393 						bufp, max_buf, offset);
394 	if (ret) {
395 		dev_err(dev, "Error %d requesting firmware file: %s\n",
396 			ret, filename);
397 		goto err_firmware_out;
398 	}
399 	dev_dbg(dev, "size=0x%zx\n", fw->size);
400 	if (load_type == VK_IMAGE_TYPE_BOOT1)
401 		memcpy_toio(vk->bar[BAR_1] + BAR1_CODEPUSH_BASE_BOOT1,
402 			    bufp,
403 			    fw->size);
404 
405 	dev_dbg(dev, "Signaling 0x%x to 0x%llx\n", codepush, offset_codepush);
406 	vkwrite32(vk, codepush, BAR_0, offset_codepush);
407 
408 	if (load_type == VK_IMAGE_TYPE_BOOT1) {
409 		u32 boot_status;
410 
411 		/* wait until done */
412 		ret = bcm_vk_wait(vk, BAR_0, BAR_BOOT_STATUS,
413 				  BOOT1_RUNNING,
414 				  BOOT1_RUNNING,
415 				  BOOT1_STARTUP_TIMEOUT_MS);
416 
417 		boot_status = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
418 		is_stdalone = !BCM_VK_INTF_IS_DOWN(boot_status) &&
419 			      (boot_status & BOOT_STDALONE_RUNNING);
420 		if (ret && !is_stdalone) {
421 			dev_err(dev,
422 				"Timeout %ld ms waiting for boot1 to come up - ret(%d)\n",
423 				BOOT1_STARTUP_TIMEOUT_MS, ret);
424 			goto err_firmware_out;
425 		} else if (is_stdalone) {
426 			u32 reg;
427 
428 			reg = vkread32(vk, BAR_0, BAR_BOOT1_STDALONE_PROGRESS);
429 			if ((reg & BOOT1_STDALONE_PROGRESS_MASK) ==
430 				     BOOT1_STDALONE_SUCCESS) {
431 				dev_info(dev, "Boot1 standalone success\n");
432 				ret = 0;
433 			} else {
434 				dev_err(dev, "Timeout %ld ms - Boot1 standalone failure\n",
435 					BOOT1_STARTUP_TIMEOUT_MS);
436 				ret = -EINVAL;
437 				goto err_firmware_out;
438 			}
439 		}
440 	} else if (load_type == VK_IMAGE_TYPE_BOOT2) {
441 		unsigned long timeout;
442 
443 		timeout = jiffies + msecs_to_jiffies(LOAD_IMAGE_TIMEOUT_MS);
444 
445 		/* To send more data to VK than max_buf allowed at a time */
446 		do {
447 			/*
448 			 * Check for ack from card. when Ack is received,
449 			 * it means all the data is received by card.
450 			 * Exit the loop after ack is received.
451 			 */
452 			ret = bcm_vk_wait(vk, BAR_0, BAR_BOOT_STATUS,
453 					  FW_LOADER_ACK_RCVD_ALL_DATA,
454 					  FW_LOADER_ACK_RCVD_ALL_DATA,
455 					  TXFR_COMPLETE_TIMEOUT_MS);
456 			if (ret == 0) {
457 				dev_dbg(dev, "Exit boot2 download\n");
458 				break;
459 			} else if (ret == -EFAULT) {
460 				dev_err(dev, "Error detected during ACK waiting");
461 				goto err_firmware_out;
462 			}
463 
464 			/* exit the loop, if there is no response from card */
465 			if (time_after(jiffies, timeout)) {
466 				dev_err(dev, "Error. No reply from card\n");
467 				ret = -ETIMEDOUT;
468 				goto err_firmware_out;
469 			}
470 
471 			/* Wait for VK to open BAR space to copy new data */
472 			ret = bcm_vk_wait(vk, BAR_0, offset_codepush,
473 					  codepush, 0,
474 					  TXFR_COMPLETE_TIMEOUT_MS);
475 			if (ret == 0) {
476 				offset += max_buf;
477 				ret = request_partial_firmware_into_buf
478 						(&fw,
479 						 filename,
480 						 dev, bufp,
481 						 max_buf,
482 						 offset);
483 				if (ret) {
484 					dev_err(dev,
485 						"Error %d requesting firmware file: %s offset: 0x%zx\n",
486 						ret, filename, offset);
487 					goto err_firmware_out;
488 				}
489 				dev_dbg(dev, "size=0x%zx\n", fw->size);
490 				dev_dbg(dev, "Signaling 0x%x to 0x%llx\n",
491 					codepush, offset_codepush);
492 				vkwrite32(vk, codepush, BAR_0, offset_codepush);
493 				/* reload timeout after every codepush */
494 				timeout = jiffies +
495 				    msecs_to_jiffies(LOAD_IMAGE_TIMEOUT_MS);
496 			} else if (ret == -EFAULT) {
497 				dev_err(dev, "Error detected waiting for transfer\n");
498 				goto err_firmware_out;
499 			}
500 		} while (1);
501 
502 		/* wait for fw status bits to indicate app ready */
503 		ret = bcm_vk_wait(vk, BAR_0, VK_BAR_FWSTS,
504 				  VK_FWSTS_READY,
505 				  VK_FWSTS_READY,
506 				  BOOT2_STARTUP_TIMEOUT_MS);
507 		if (ret < 0) {
508 			dev_err(dev, "Boot2 not ready - ret(%d)\n", ret);
509 			goto err_firmware_out;
510 		}
511 
512 		is_stdalone = vkread32(vk, BAR_0, BAR_BOOT_STATUS) &
513 			      BOOT_STDALONE_RUNNING;
514 		if (!is_stdalone) {
515 			ret = bcm_vk_intf_ver_chk(vk);
516 			if (ret) {
517 				dev_err(dev, "failure in intf version check\n");
518 				goto err_firmware_out;
519 			}
520 
521 			/* sync & channel other info */
522 			ret = bcm_vk_sync_card_info(vk);
523 			if (ret) {
524 				dev_err(dev, "Syncing Card Info failure\n");
525 				goto err_firmware_out;
526 			}
527 		}
528 	}
529 
530 err_firmware_out:
531 	release_firmware(fw);
532 
533 err_buf_out:
534 	if (bufp)
535 		dma_free_coherent(dev, max_buf, bufp, boot_dma_addr);
536 
537 	return ret;
538 }
539 
540 static u32 bcm_vk_next_boot_image(struct bcm_vk *vk)
541 {
542 	u32 boot_status;
543 	u32 fw_status;
544 	u32 load_type = 0;  /* default for unknown */
545 
546 	boot_status = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
547 	fw_status = vkread32(vk, BAR_0, VK_BAR_FWSTS);
548 
549 	if (!BCM_VK_INTF_IS_DOWN(boot_status) && (boot_status & SRAM_OPEN))
550 		load_type = VK_IMAGE_TYPE_BOOT1;
551 	else if (boot_status == BOOT1_RUNNING)
552 		load_type = VK_IMAGE_TYPE_BOOT2;
553 
554 	/* Log status so that we know different stages */
555 	dev_info(&vk->pdev->dev,
556 		 "boot-status value for next image: 0x%x : fw-status 0x%x\n",
557 		 boot_status, fw_status);
558 
559 	return load_type;
560 }
561 
562 static enum soc_idx get_soc_idx(struct bcm_vk *vk)
563 {
564 	struct pci_dev *pdev = vk->pdev;
565 	enum soc_idx idx = VK_IDX_INVALID;
566 	u32 rev;
567 	static enum soc_idx const vk_soc_tab[] = { VALKYRIE_A0, VALKYRIE_B0 };
568 
569 	switch (pdev->device) {
570 	case PCI_DEVICE_ID_VALKYRIE:
571 		/* get the chip id to decide sub-class */
572 		rev = MAJOR_SOC_REV(vkread32(vk, BAR_0, BAR_CHIP_ID));
573 		if (rev < ARRAY_SIZE(vk_soc_tab)) {
574 			idx = vk_soc_tab[rev];
575 		} else {
576 			/* Default to A0 firmware for all other chip revs */
577 			idx = VALKYRIE_A0;
578 			dev_warn(&pdev->dev,
579 				 "Rev %d not in image lookup table, default to idx=%d\n",
580 				 rev, idx);
581 		}
582 		break;
583 
584 	case PCI_DEVICE_ID_VIPER:
585 		idx = VIPER;
586 		break;
587 
588 	default:
589 		dev_err(&pdev->dev, "no images for 0x%x\n", pdev->device);
590 	}
591 	return idx;
592 }
593 
594 static const char *get_load_fw_name(struct bcm_vk *vk,
595 				    const struct load_image_entry *entry)
596 {
597 	const struct firmware *fw;
598 	struct device *dev = &vk->pdev->dev;
599 	int ret;
600 	unsigned long dummy;
601 	int i;
602 
603 	for (i = 0; i < IMG_PER_TYPE_MAX; i++) {
604 		fw = NULL;
605 		ret = request_partial_firmware_into_buf(&fw,
606 							entry->image_name[i],
607 							dev, &dummy,
608 							sizeof(dummy),
609 							0);
610 		release_firmware(fw);
611 		if (!ret)
612 			return entry->image_name[i];
613 	}
614 	return NULL;
615 }
616 
617 int bcm_vk_auto_load_all_images(struct bcm_vk *vk)
618 {
619 	int i, ret = -1;
620 	enum soc_idx idx;
621 	struct device *dev = &vk->pdev->dev;
622 	u32 curr_type;
623 	const char *curr_name;
624 
625 	idx = get_soc_idx(vk);
626 	if (idx == VK_IDX_INVALID)
627 		goto auto_load_all_exit;
628 
629 	/* log a message to know the relative loading order */
630 	dev_dbg(dev, "Load All for device %d\n", vk->devid);
631 
632 	for (i = 0; i < NUM_BOOT_STAGES; i++) {
633 		curr_type = image_tab[idx][i].image_type;
634 		if (bcm_vk_next_boot_image(vk) == curr_type) {
635 			curr_name = get_load_fw_name(vk, &image_tab[idx][i]);
636 			if (!curr_name) {
637 				dev_err(dev, "No suitable firmware exists for type %d",
638 					curr_type);
639 				ret = -ENOENT;
640 				goto auto_load_all_exit;
641 			}
642 			ret = bcm_vk_load_image_by_type(vk, curr_type,
643 							curr_name);
644 			dev_info(dev, "Auto load %s, ret %d\n",
645 				 curr_name, ret);
646 
647 			if (ret) {
648 				dev_err(dev, "Error loading default %s\n",
649 					curr_name);
650 				goto auto_load_all_exit;
651 			}
652 		}
653 	}
654 
655 auto_load_all_exit:
656 	return ret;
657 }
658 
659 static int bcm_vk_trigger_autoload(struct bcm_vk *vk)
660 {
661 	if (test_and_set_bit(BCM_VK_WQ_DWNLD_PEND, vk->wq_offload) != 0)
662 		return -EPERM;
663 
664 	set_bit(BCM_VK_WQ_DWNLD_AUTO, vk->wq_offload);
665 	queue_work(vk->wq_thread, &vk->wq_work);
666 
667 	return 0;
668 }
669 
670 /*
671  * deferred work queue for auto download.
672  */
673 static void bcm_vk_wq_handler(struct work_struct *work)
674 {
675 	struct bcm_vk *vk = container_of(work, struct bcm_vk, wq_work);
676 
677 	if (test_bit(BCM_VK_WQ_DWNLD_AUTO, vk->wq_offload)) {
678 		bcm_vk_auto_load_all_images(vk);
679 
680 		/*
681 		 * at the end of operation, clear AUTO bit and pending
682 		 * bit
683 		 */
684 		clear_bit(BCM_VK_WQ_DWNLD_AUTO, vk->wq_offload);
685 		clear_bit(BCM_VK_WQ_DWNLD_PEND, vk->wq_offload);
686 	}
687 }
688 
689 static long bcm_vk_load_image(struct bcm_vk *vk,
690 			      const struct vk_image __user *arg)
691 {
692 	struct device *dev = &vk->pdev->dev;
693 	const char *image_name;
694 	struct vk_image image;
695 	u32 next_loadable;
696 	enum soc_idx idx;
697 	int image_idx;
698 	int ret = -EPERM;
699 
700 	if (copy_from_user(&image, arg, sizeof(image)))
701 		return -EACCES;
702 
703 	if ((image.type != VK_IMAGE_TYPE_BOOT1) &&
704 	    (image.type != VK_IMAGE_TYPE_BOOT2)) {
705 		dev_err(dev, "invalid image.type %u\n", image.type);
706 		return ret;
707 	}
708 
709 	next_loadable = bcm_vk_next_boot_image(vk);
710 	if (next_loadable != image.type) {
711 		dev_err(dev, "Next expected image %u, Loading %u\n",
712 			next_loadable, image.type);
713 		return ret;
714 	}
715 
716 	/*
717 	 * if something is pending download already.  This could only happen
718 	 * for now when the driver is being loaded, or if someone has issued
719 	 * another download command in another shell.
720 	 */
721 	if (test_and_set_bit(BCM_VK_WQ_DWNLD_PEND, vk->wq_offload) != 0) {
722 		dev_err(dev, "Download operation already pending.\n");
723 		return ret;
724 	}
725 
726 	image_name = image.filename;
727 	if (image_name[0] == '\0') {
728 		/* Use default image name if NULL */
729 		idx = get_soc_idx(vk);
730 		if (idx == VK_IDX_INVALID)
731 			goto err_idx;
732 
733 		/* Image idx starts with boot1 */
734 		image_idx = image.type - VK_IMAGE_TYPE_BOOT1;
735 		image_name = get_load_fw_name(vk, &image_tab[idx][image_idx]);
736 		if (!image_name) {
737 			dev_err(dev, "No suitable image found for type %d",
738 				image.type);
739 			ret = -ENOENT;
740 			goto err_idx;
741 		}
742 	} else {
743 		/* Ensure filename is NULL terminated */
744 		image.filename[sizeof(image.filename) - 1] = '\0';
745 	}
746 	ret = bcm_vk_load_image_by_type(vk, image.type, image_name);
747 	dev_info(dev, "Load %s, ret %d\n", image_name, ret);
748 err_idx:
749 	clear_bit(BCM_VK_WQ_DWNLD_PEND, vk->wq_offload);
750 
751 	return ret;
752 }
753 
754 static void bcm_to_v_reset_doorbell(struct bcm_vk *vk, u32 db_val)
755 {
756 	vkwrite32(vk, db_val, BAR_0, VK_BAR0_RESET_DB_BASE);
757 }
758 
759 static int bcm_vk_trigger_reset(struct bcm_vk *vk)
760 {
761 	u32 i;
762 	u32 value, boot_status;
763 	static const u32 bar0_reg_clr_list[] = { BAR_OS_UPTIME,
764 						 BAR_INTF_VER,
765 						 BAR_CARD_VOLTAGE,
766 						 BAR_CARD_TEMPERATURE,
767 						 BAR_CARD_PWR_AND_THRE };
768 
769 	/* make tag '\0' terminated */
770 	vkwrite32(vk, 0, BAR_1, VK_BAR1_BOOT1_VER_TAG);
771 
772 	for (i = 0; i < VK_BAR1_DAUTH_MAX; i++) {
773 		vkwrite32(vk, 0, BAR_1, VK_BAR1_DAUTH_STORE_ADDR(i));
774 		vkwrite32(vk, 0, BAR_1, VK_BAR1_DAUTH_VALID_ADDR(i));
775 	}
776 	for (i = 0; i < VK_BAR1_SOTP_REVID_MAX; i++)
777 		vkwrite32(vk, 0, BAR_1, VK_BAR1_SOTP_REVID_ADDR(i));
778 
779 	/*
780 	 * When boot request fails, the CODE_PUSH_OFFSET stays persistent.
781 	 * Allowing us to debug the failure. When we call reset,
782 	 * we should clear CODE_PUSH_OFFSET so ROM does not execute
783 	 * boot again (and fails again) and instead waits for a new
784 	 * codepush.  And, if previous boot has encountered error, need
785 	 * to clear the entry values
786 	 */
787 	boot_status = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
788 	if (boot_status & BOOT_ERR_MASK) {
789 		dev_info(&vk->pdev->dev,
790 			 "Card in boot error 0x%x, clear CODEPUSH val\n",
791 			 boot_status);
792 		value = 0;
793 	} else {
794 		value = vkread32(vk, BAR_0, BAR_CODEPUSH_SBL);
795 		value &= CODEPUSH_MASK;
796 	}
797 	vkwrite32(vk, value, BAR_0, BAR_CODEPUSH_SBL);
798 
799 	/* reset fw_status with proper reason, and press db */
800 	vkwrite32(vk, VK_FWSTS_RESET_MBOX_DB, BAR_0, VK_BAR_FWSTS);
801 	bcm_to_v_reset_doorbell(vk, VK_BAR0_RESET_DB_SOFT);
802 
803 	/* clear other necessary registers records */
804 	for (i = 0; i < ARRAY_SIZE(bar0_reg_clr_list); i++)
805 		vkwrite32(vk, 0, BAR_0, bar0_reg_clr_list[i]);
806 
807 	return 0;
808 }
809 
810 static long bcm_vk_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
811 {
812 	long ret = -EINVAL;
813 	struct bcm_vk_ctx *ctx = file->private_data;
814 	struct bcm_vk *vk = container_of(ctx->miscdev, struct bcm_vk, miscdev);
815 	void __user *argp = (void __user *)arg;
816 
817 	dev_dbg(&vk->pdev->dev,
818 		"ioctl, cmd=0x%02x, arg=0x%02lx\n",
819 		cmd, arg);
820 
821 	mutex_lock(&vk->mutex);
822 
823 	switch (cmd) {
824 	case VK_IOCTL_LOAD_IMAGE:
825 		ret = bcm_vk_load_image(vk, argp);
826 		break;
827 
828 	default:
829 		break;
830 	}
831 
832 	mutex_unlock(&vk->mutex);
833 
834 	return ret;
835 }
836 
837 static const struct file_operations bcm_vk_fops = {
838 	.owner = THIS_MODULE,
839 	.open = bcm_vk_open,
840 	.release = bcm_vk_release,
841 	.unlocked_ioctl = bcm_vk_ioctl,
842 };
843 
844 static int bcm_vk_on_panic(struct notifier_block *nb,
845 			   unsigned long e, void *p)
846 {
847 	struct bcm_vk *vk = container_of(nb, struct bcm_vk, panic_nb);
848 
849 	bcm_to_v_reset_doorbell(vk, VK_BAR0_RESET_DB_HARD);
850 
851 	return 0;
852 }
853 
854 static int bcm_vk_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
855 {
856 	int err;
857 	int i;
858 	int id;
859 	int irq;
860 	char name[20];
861 	struct bcm_vk *vk;
862 	struct device *dev = &pdev->dev;
863 	struct miscdevice *misc_device;
864 	u32 boot_status;
865 
866 	/* allocate vk structure which is tied to kref for freeing */
867 	vk = kzalloc(sizeof(*vk), GFP_KERNEL);
868 	if (!vk)
869 		return -ENOMEM;
870 
871 	kref_init(&vk->kref);
872 	mutex_init(&vk->mutex);
873 
874 	err = pci_enable_device(pdev);
875 	if (err) {
876 		dev_err(dev, "Cannot enable PCI device\n");
877 		goto err_free_exit;
878 	}
879 	vk->pdev = pci_dev_get(pdev);
880 
881 	err = pci_request_regions(pdev, DRV_MODULE_NAME);
882 	if (err) {
883 		dev_err(dev, "Cannot obtain PCI resources\n");
884 		goto err_disable_pdev;
885 	}
886 
887 	/* make sure DMA is good */
888 	err = dma_set_mask_and_coherent(&pdev->dev,
889 					DMA_BIT_MASK(BCM_VK_DMA_BITS));
890 	if (err) {
891 		dev_err(dev, "failed to set DMA mask\n");
892 		goto err_disable_pdev;
893 	}
894 
895 	/* The tdma is a scratch area for some DMA testings. */
896 	if (nr_scratch_pages) {
897 		vk->tdma_vaddr = dma_alloc_coherent
898 					(dev,
899 					 nr_scratch_pages * PAGE_SIZE,
900 					 &vk->tdma_addr, GFP_KERNEL);
901 		if (!vk->tdma_vaddr) {
902 			err = -ENOMEM;
903 			goto err_disable_pdev;
904 		}
905 	}
906 
907 	pci_set_master(pdev);
908 	pci_set_drvdata(pdev, vk);
909 
910 	irq = pci_alloc_irq_vectors(pdev,
911 				    1,
912 				    VK_MSIX_IRQ_MAX,
913 				    PCI_IRQ_MSI | PCI_IRQ_MSIX);
914 
915 	if (irq < VK_MSIX_IRQ_MIN_REQ) {
916 		dev_err(dev, "failed to get min %d MSIX interrupts, irq(%d)\n",
917 			VK_MSIX_IRQ_MIN_REQ, irq);
918 		err = (irq >= 0) ? -EINVAL : irq;
919 		goto err_disable_pdev;
920 	}
921 
922 	if (irq != VK_MSIX_IRQ_MAX)
923 		dev_warn(dev, "Number of IRQs %d allocated - requested(%d).\n",
924 			 irq, VK_MSIX_IRQ_MAX);
925 
926 	for (i = 0; i < MAX_BAR; i++) {
927 		/* multiple by 2 for 64 bit BAR mapping */
928 		vk->bar[i] = pci_ioremap_bar(pdev, i * 2);
929 		if (!vk->bar[i]) {
930 			dev_err(dev, "failed to remap BAR%d\n", i);
931 			goto err_iounmap;
932 		}
933 	}
934 
935 	id = ida_simple_get(&bcm_vk_ida, 0, 0, GFP_KERNEL);
936 	if (id < 0) {
937 		err = id;
938 		dev_err(dev, "unable to get id\n");
939 		goto err_iounmap;
940 	}
941 
942 	vk->devid = id;
943 	snprintf(name, sizeof(name), DRV_MODULE_NAME ".%d", id);
944 	misc_device = &vk->miscdev;
945 	misc_device->minor = MISC_DYNAMIC_MINOR;
946 	misc_device->name = kstrdup(name, GFP_KERNEL);
947 	if (!misc_device->name) {
948 		err = -ENOMEM;
949 		goto err_ida_remove;
950 	}
951 	misc_device->fops = &bcm_vk_fops,
952 
953 	err = misc_register(misc_device);
954 	if (err) {
955 		dev_err(dev, "failed to register device\n");
956 		goto err_kfree_name;
957 	}
958 
959 	INIT_WORK(&vk->wq_work, bcm_vk_wq_handler);
960 
961 	/* create dedicated workqueue */
962 	vk->wq_thread = create_singlethread_workqueue(name);
963 	if (!vk->wq_thread) {
964 		dev_err(dev, "Fail to create workqueue thread\n");
965 		err = -ENOMEM;
966 		goto err_misc_deregister;
967 	}
968 
969 	/* sync other info */
970 	bcm_vk_sync_card_info(vk);
971 
972 	/* register for panic notifier */
973 	vk->panic_nb.notifier_call = bcm_vk_on_panic;
974 	err = atomic_notifier_chain_register(&panic_notifier_list,
975 					     &vk->panic_nb);
976 	if (err) {
977 		dev_err(dev, "Fail to register panic notifier\n");
978 		goto err_destroy_workqueue;
979 	}
980 
981 	/*
982 	 * lets trigger an auto download.  We don't want to do it serially here
983 	 * because at probing time, it is not supposed to block for a long time.
984 	 */
985 	boot_status = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
986 	if (auto_load) {
987 		if ((boot_status & BOOT_STATE_MASK) == BROM_RUNNING) {
988 			if (bcm_vk_trigger_autoload(vk))
989 				goto err_unregister_panic_notifier;
990 		} else {
991 			dev_err(dev,
992 				"Auto-load skipped - BROM not in proper state (0x%x)\n",
993 				boot_status);
994 		}
995 	}
996 
997 	dev_dbg(dev, "BCM-VK:%u created\n", id);
998 
999 	return 0;
1000 
1001 err_unregister_panic_notifier:
1002 	atomic_notifier_chain_unregister(&panic_notifier_list,
1003 					 &vk->panic_nb);
1004 
1005 err_destroy_workqueue:
1006 	destroy_workqueue(vk->wq_thread);
1007 
1008 err_misc_deregister:
1009 	misc_deregister(misc_device);
1010 
1011 err_kfree_name:
1012 	kfree(misc_device->name);
1013 	misc_device->name = NULL;
1014 
1015 err_ida_remove:
1016 	ida_simple_remove(&bcm_vk_ida, id);
1017 
1018 err_iounmap:
1019 	for (i = 0; i < MAX_BAR; i++) {
1020 		if (vk->bar[i])
1021 			pci_iounmap(pdev, vk->bar[i]);
1022 	}
1023 	pci_release_regions(pdev);
1024 
1025 err_disable_pdev:
1026 	if (vk->tdma_vaddr)
1027 		dma_free_coherent(&pdev->dev, nr_scratch_pages * PAGE_SIZE,
1028 				  vk->tdma_vaddr, vk->tdma_addr);
1029 
1030 	pci_free_irq_vectors(pdev);
1031 	pci_disable_device(pdev);
1032 	pci_dev_put(pdev);
1033 
1034 err_free_exit:
1035 	kfree(vk);
1036 
1037 	return err;
1038 }
1039 
1040 void bcm_vk_release_data(struct kref *kref)
1041 {
1042 	struct bcm_vk *vk = container_of(kref, struct bcm_vk, kref);
1043 	struct pci_dev *pdev = vk->pdev;
1044 
1045 	dev_dbg(&pdev->dev, "BCM-VK:%d release data 0x%p\n", vk->devid, vk);
1046 	pci_dev_put(pdev);
1047 	kfree(vk);
1048 }
1049 
1050 static void bcm_vk_remove(struct pci_dev *pdev)
1051 {
1052 	int i;
1053 	struct bcm_vk *vk = pci_get_drvdata(pdev);
1054 	struct miscdevice *misc_device = &vk->miscdev;
1055 
1056 	/*
1057 	 * Trigger a reset to card and wait enough time for UCODE to rerun,
1058 	 * which re-initialize the card into its default state.
1059 	 * This ensures when driver is re-enumerated it will start from
1060 	 * a completely clean state.
1061 	 */
1062 	bcm_vk_trigger_reset(vk);
1063 	usleep_range(BCM_VK_UCODE_BOOT_US, BCM_VK_UCODE_BOOT_MAX_US);
1064 
1065 	/* unregister panic notifier */
1066 	atomic_notifier_chain_unregister(&panic_notifier_list,
1067 					 &vk->panic_nb);
1068 
1069 	if (vk->tdma_vaddr)
1070 		dma_free_coherent(&pdev->dev, nr_scratch_pages * PAGE_SIZE,
1071 				  vk->tdma_vaddr, vk->tdma_addr);
1072 
1073 	/* remove if name is set which means misc dev registered */
1074 	if (misc_device->name) {
1075 		misc_deregister(misc_device);
1076 		kfree(misc_device->name);
1077 		ida_simple_remove(&bcm_vk_ida, vk->devid);
1078 	}
1079 
1080 	cancel_work_sync(&vk->wq_work);
1081 	destroy_workqueue(vk->wq_thread);
1082 
1083 	for (i = 0; i < MAX_BAR; i++) {
1084 		if (vk->bar[i])
1085 			pci_iounmap(pdev, vk->bar[i]);
1086 	}
1087 
1088 	dev_dbg(&pdev->dev, "BCM-VK:%d released\n", vk->devid);
1089 
1090 	pci_release_regions(pdev);
1091 	pci_free_irq_vectors(pdev);
1092 	pci_disable_device(pdev);
1093 
1094 	kref_put(&vk->kref, bcm_vk_release_data);
1095 }
1096 
1097 static void bcm_vk_shutdown(struct pci_dev *pdev)
1098 {
1099 	struct bcm_vk *vk = pci_get_drvdata(pdev);
1100 	u32 reg, boot_stat;
1101 
1102 	reg = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
1103 	boot_stat = reg & BOOT_STATE_MASK;
1104 
1105 	if (boot_stat == BOOT1_RUNNING) {
1106 		/* simply trigger a reset interrupt to park it */
1107 		bcm_vk_trigger_reset(vk);
1108 	} else if (boot_stat == BROM_NOT_RUN) {
1109 		int err;
1110 		u16 lnksta;
1111 
1112 		/*
1113 		 * The boot status only reflects boot condition since last reset
1114 		 * As ucode will run only once to configure pcie, if multiple
1115 		 * resets happen, we lost track if ucode has run or not.
1116 		 * Here, read the current link speed and use that to
1117 		 * sync up the bootstatus properly so that on reboot-back-up,
1118 		 * it has the proper state to start with autoload
1119 		 */
1120 		err = pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnksta);
1121 		if (!err &&
1122 		    (lnksta & PCI_EXP_LNKSTA_CLS) != PCI_EXP_LNKSTA_CLS_2_5GB) {
1123 			reg |= BROM_STATUS_COMPLETE;
1124 			vkwrite32(vk, reg, BAR_0, BAR_BOOT_STATUS);
1125 		}
1126 	}
1127 }
1128 
1129 static const struct pci_device_id bcm_vk_ids[] = {
1130 	{ PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, PCI_DEVICE_ID_VALKYRIE), },
1131 	{ PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, PCI_DEVICE_ID_VIPER), },
1132 	{ }
1133 };
1134 MODULE_DEVICE_TABLE(pci, bcm_vk_ids);
1135 
1136 static struct pci_driver pci_driver = {
1137 	.name     = DRV_MODULE_NAME,
1138 	.id_table = bcm_vk_ids,
1139 	.probe    = bcm_vk_probe,
1140 	.remove   = bcm_vk_remove,
1141 	.shutdown = bcm_vk_shutdown,
1142 };
1143 module_pci_driver(pci_driver);
1144 
1145 MODULE_DESCRIPTION("Broadcom VK Host Driver");
1146 MODULE_AUTHOR("Scott Branden <scott.branden@broadcom.com>");
1147 MODULE_LICENSE("GPL v2");
1148 MODULE_VERSION("1.0");
1149