xref: /linux/drivers/misc/bcm-vk/bcm_vk_dev.c (revision af22527e82d12f9d0b5afb39f25926a91d5fa7e7)
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/module.h>
12 #include <linux/pci.h>
13 #include <linux/pci_regs.h>
14 #include <uapi/linux/misc/bcm_vk.h>
15 
16 #include "bcm_vk.h"
17 
18 #define PCI_DEVICE_ID_VALKYRIE	0x5e87
19 #define PCI_DEVICE_ID_VIPER	0x5e88
20 
21 static DEFINE_IDA(bcm_vk_ida);
22 
23 enum soc_idx {
24 	VALKYRIE_A0 = 0,
25 	VALKYRIE_B0,
26 	VIPER,
27 	VK_IDX_INVALID
28 };
29 
30 enum img_idx {
31 	IMG_PRI = 0,
32 	IMG_SEC,
33 	IMG_PER_TYPE_MAX
34 };
35 
36 struct load_image_entry {
37 	const u32 image_type;
38 	const char *image_name[IMG_PER_TYPE_MAX];
39 };
40 
41 #define NUM_BOOT_STAGES 2
42 /* default firmware images names */
43 static const struct load_image_entry image_tab[][NUM_BOOT_STAGES] = {
44 	[VALKYRIE_A0] = {
45 		{VK_IMAGE_TYPE_BOOT1, {"vk_a0-boot1.bin", "vk-boot1.bin"}},
46 		{VK_IMAGE_TYPE_BOOT2, {"vk_a0-boot2.bin", "vk-boot2.bin"}}
47 	},
48 	[VALKYRIE_B0] = {
49 		{VK_IMAGE_TYPE_BOOT1, {"vk_b0-boot1.bin", "vk-boot1.bin"}},
50 		{VK_IMAGE_TYPE_BOOT2, {"vk_b0-boot2.bin", "vk-boot2.bin"}}
51 	},
52 
53 	[VIPER] = {
54 		{VK_IMAGE_TYPE_BOOT1, {"vp-boot1.bin", ""}},
55 		{VK_IMAGE_TYPE_BOOT2, {"vp-boot2.bin", ""}}
56 	},
57 };
58 
59 /* Location of memory base addresses of interest in BAR1 */
60 /* Load Boot1 to start of ITCM */
61 #define BAR1_CODEPUSH_BASE_BOOT1	0x100000
62 
63 /* Allow minimum 1s for Load Image timeout responses */
64 #define LOAD_IMAGE_TIMEOUT_MS		(1 * MSEC_PER_SEC)
65 
66 /* Image startup timeouts */
67 #define BOOT1_STARTUP_TIMEOUT_MS	(5 * MSEC_PER_SEC)
68 #define BOOT2_STARTUP_TIMEOUT_MS	(10 * MSEC_PER_SEC)
69 
70 /* 1ms wait for checking the transfer complete status */
71 #define TXFR_COMPLETE_TIMEOUT_MS	1
72 
73 /* MSIX usages */
74 #define VK_MSIX_MSGQ_MAX		3
75 #define VK_MSIX_NOTF_MAX		1
76 #define VK_MSIX_TTY_MAX			BCM_VK_NUM_TTY
77 #define VK_MSIX_IRQ_MAX			(VK_MSIX_MSGQ_MAX + VK_MSIX_NOTF_MAX + \
78 					 VK_MSIX_TTY_MAX)
79 #define VK_MSIX_IRQ_MIN_REQ             (VK_MSIX_MSGQ_MAX + VK_MSIX_NOTF_MAX)
80 
81 /* Number of bits set in DMA mask*/
82 #define BCM_VK_DMA_BITS			64
83 
84 /* Ucode boot wait time */
85 #define BCM_VK_UCODE_BOOT_US            (100 * USEC_PER_MSEC)
86 /* 50% margin */
87 #define BCM_VK_UCODE_BOOT_MAX_US        ((BCM_VK_UCODE_BOOT_US * 3) >> 1)
88 
89 /* deinit time for the card os after receiving doorbell */
90 #define BCM_VK_DEINIT_TIME_MS		(2 * MSEC_PER_SEC)
91 
92 /*
93  * module parameters
94  */
95 static bool auto_load = true;
96 module_param(auto_load, bool, 0444);
97 MODULE_PARM_DESC(auto_load,
98 		 "Load images automatically at PCIe probe time.\n");
99 static uint nr_scratch_pages = VK_BAR1_SCRATCH_DEF_NR_PAGES;
100 module_param(nr_scratch_pages, uint, 0444);
101 MODULE_PARM_DESC(nr_scratch_pages,
102 		 "Number of pre allocated DMAable coherent pages.\n");
103 
104 static int bcm_vk_intf_ver_chk(struct bcm_vk *vk)
105 {
106 	struct device *dev = &vk->pdev->dev;
107 	u32 reg;
108 	u16 major, minor;
109 	int ret = 0;
110 
111 	/* read interface register */
112 	reg = vkread32(vk, BAR_0, BAR_INTF_VER);
113 	major = (reg >> BAR_INTF_VER_MAJOR_SHIFT) & BAR_INTF_VER_MASK;
114 	minor = reg & BAR_INTF_VER_MASK;
115 
116 	/*
117 	 * if major number is 0, it is pre-release and it would be allowed
118 	 * to continue, else, check versions accordingly
119 	 */
120 	if (!major) {
121 		dev_warn(dev, "Pre-release major.minor=%d.%d - drv %d.%d\n",
122 			 major, minor, SEMANTIC_MAJOR, SEMANTIC_MINOR);
123 	} else if (major != SEMANTIC_MAJOR) {
124 		dev_err(dev,
125 			"Intf major.minor=%d.%d rejected - drv %d.%d\n",
126 			major, minor, SEMANTIC_MAJOR, SEMANTIC_MINOR);
127 		ret = -EPFNOSUPPORT;
128 	} else {
129 		dev_dbg(dev,
130 			"Intf major.minor=%d.%d passed - drv %d.%d\n",
131 			major, minor, SEMANTIC_MAJOR, SEMANTIC_MINOR);
132 	}
133 	return ret;
134 }
135 
136 static inline int bcm_vk_wait(struct bcm_vk *vk, enum pci_barno bar,
137 			      u64 offset, u32 mask, u32 value,
138 			      unsigned long timeout_ms)
139 {
140 	struct device *dev = &vk->pdev->dev;
141 	unsigned long start_time;
142 	unsigned long timeout;
143 	u32 rd_val, boot_status;
144 
145 	start_time = jiffies;
146 	timeout = start_time + msecs_to_jiffies(timeout_ms);
147 
148 	do {
149 		rd_val = vkread32(vk, bar, offset);
150 		dev_dbg(dev, "BAR%d Offset=0x%llx: 0x%x\n",
151 			bar, offset, rd_val);
152 
153 		/* check for any boot err condition */
154 		boot_status = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
155 		if (boot_status & BOOT_ERR_MASK) {
156 			dev_err(dev, "Boot Err 0x%x, progress 0x%x after %d ms\n",
157 				(boot_status & BOOT_ERR_MASK) >> BOOT_ERR_SHIFT,
158 				boot_status & BOOT_PROG_MASK,
159 				jiffies_to_msecs(jiffies - start_time));
160 			return -EFAULT;
161 		}
162 
163 		if (time_after(jiffies, timeout))
164 			return -ETIMEDOUT;
165 
166 		cpu_relax();
167 		cond_resched();
168 	} while ((rd_val & mask) != value);
169 
170 	return 0;
171 }
172 
173 static int bcm_vk_sync_card_info(struct bcm_vk *vk)
174 {
175 	u32 rdy_marker = vkread32(vk, BAR_1, VK_BAR1_MSGQ_DEF_RDY);
176 
177 	/* check for marker, but allow diags mode to skip sync */
178 	if (!bcm_vk_msgq_marker_valid(vk))
179 		return (rdy_marker == VK_BAR1_DIAG_RDY_MARKER ? 0 : -EINVAL);
180 
181 	/*
182 	 * Write down scratch addr which is used for DMA. For
183 	 * signed part, BAR1 is accessible only after boot2 has come
184 	 * up
185 	 */
186 	if (vk->tdma_addr) {
187 		vkwrite32(vk, (u64)vk->tdma_addr >> 32, BAR_1,
188 			  VK_BAR1_SCRATCH_OFF_HI);
189 		vkwrite32(vk, (u32)vk->tdma_addr, BAR_1,
190 			  VK_BAR1_SCRATCH_OFF_LO);
191 		vkwrite32(vk, nr_scratch_pages * PAGE_SIZE, BAR_1,
192 			  VK_BAR1_SCRATCH_SZ_ADDR);
193 	}
194 	return 0;
195 }
196 
197 static void bcm_vk_buf_notify(struct bcm_vk *vk, void *bufp,
198 			      dma_addr_t host_buf_addr, u32 buf_size)
199 {
200 	/* update the dma address to the card */
201 	vkwrite32(vk, (u64)host_buf_addr >> 32, BAR_1,
202 		  VK_BAR1_DMA_BUF_OFF_HI);
203 	vkwrite32(vk, (u32)host_buf_addr, BAR_1,
204 		  VK_BAR1_DMA_BUF_OFF_LO);
205 	vkwrite32(vk, buf_size, BAR_1, VK_BAR1_DMA_BUF_SZ);
206 }
207 
208 static int bcm_vk_load_image_by_type(struct bcm_vk *vk, u32 load_type,
209 				     const char *filename)
210 {
211 	struct device *dev = &vk->pdev->dev;
212 	const struct firmware *fw = NULL;
213 	void *bufp = NULL;
214 	size_t max_buf, offset;
215 	int ret;
216 	u64 offset_codepush;
217 	u32 codepush;
218 	u32 value;
219 	dma_addr_t boot_dma_addr;
220 	bool is_stdalone;
221 
222 	if (load_type == VK_IMAGE_TYPE_BOOT1) {
223 		/*
224 		 * After POR, enable VK soft BOOTSRC so bootrom do not clear
225 		 * the pushed image (the TCM memories).
226 		 */
227 		value = vkread32(vk, BAR_0, BAR_BOOTSRC_SELECT);
228 		value |= BOOTSRC_SOFT_ENABLE;
229 		vkwrite32(vk, value, BAR_0, BAR_BOOTSRC_SELECT);
230 
231 		codepush = CODEPUSH_BOOTSTART + CODEPUSH_BOOT1_ENTRY;
232 		offset_codepush = BAR_CODEPUSH_SBL;
233 
234 		/* Write a 1 to request SRAM open bit */
235 		vkwrite32(vk, CODEPUSH_BOOTSTART, BAR_0, offset_codepush);
236 
237 		/* Wait for VK to respond */
238 		ret = bcm_vk_wait(vk, BAR_0, BAR_BOOT_STATUS, SRAM_OPEN,
239 				  SRAM_OPEN, LOAD_IMAGE_TIMEOUT_MS);
240 		if (ret < 0) {
241 			dev_err(dev, "boot1 wait SRAM err - ret(%d)\n", ret);
242 			goto err_buf_out;
243 		}
244 
245 		max_buf = SZ_256K;
246 		bufp = dma_alloc_coherent(dev,
247 					  max_buf,
248 					  &boot_dma_addr, GFP_KERNEL);
249 		if (!bufp) {
250 			dev_err(dev, "Error allocating 0x%zx\n", max_buf);
251 			ret = -ENOMEM;
252 			goto err_buf_out;
253 		}
254 	} else if (load_type == VK_IMAGE_TYPE_BOOT2) {
255 		codepush = CODEPUSH_BOOT2_ENTRY;
256 		offset_codepush = BAR_CODEPUSH_SBI;
257 
258 		/* Wait for VK to respond */
259 		ret = bcm_vk_wait(vk, BAR_0, BAR_BOOT_STATUS, DDR_OPEN,
260 				  DDR_OPEN, LOAD_IMAGE_TIMEOUT_MS);
261 		if (ret < 0) {
262 			dev_err(dev, "boot2 wait DDR open error - ret(%d)\n",
263 				ret);
264 			goto err_buf_out;
265 		}
266 
267 		max_buf = SZ_4M;
268 		bufp = dma_alloc_coherent(dev,
269 					  max_buf,
270 					  &boot_dma_addr, GFP_KERNEL);
271 		if (!bufp) {
272 			dev_err(dev, "Error allocating 0x%zx\n", max_buf);
273 			ret = -ENOMEM;
274 			goto err_buf_out;
275 		}
276 
277 		bcm_vk_buf_notify(vk, bufp, boot_dma_addr, max_buf);
278 	} else {
279 		dev_err(dev, "Error invalid image type 0x%x\n", load_type);
280 		ret = -EINVAL;
281 		goto err_buf_out;
282 	}
283 
284 	offset = 0;
285 	ret = request_partial_firmware_into_buf(&fw, filename, dev,
286 						bufp, max_buf, offset);
287 	if (ret) {
288 		dev_err(dev, "Error %d requesting firmware file: %s\n",
289 			ret, filename);
290 		goto err_firmware_out;
291 	}
292 	dev_dbg(dev, "size=0x%zx\n", fw->size);
293 	if (load_type == VK_IMAGE_TYPE_BOOT1)
294 		memcpy_toio(vk->bar[BAR_1] + BAR1_CODEPUSH_BASE_BOOT1,
295 			    bufp,
296 			    fw->size);
297 
298 	dev_dbg(dev, "Signaling 0x%x to 0x%llx\n", codepush, offset_codepush);
299 	vkwrite32(vk, codepush, BAR_0, offset_codepush);
300 
301 	if (load_type == VK_IMAGE_TYPE_BOOT1) {
302 		u32 boot_status;
303 
304 		/* wait until done */
305 		ret = bcm_vk_wait(vk, BAR_0, BAR_BOOT_STATUS,
306 				  BOOT1_RUNNING,
307 				  BOOT1_RUNNING,
308 				  BOOT1_STARTUP_TIMEOUT_MS);
309 
310 		boot_status = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
311 		is_stdalone = !BCM_VK_INTF_IS_DOWN(boot_status) &&
312 			      (boot_status & BOOT_STDALONE_RUNNING);
313 		if (ret && !is_stdalone) {
314 			dev_err(dev,
315 				"Timeout %ld ms waiting for boot1 to come up - ret(%d)\n",
316 				BOOT1_STARTUP_TIMEOUT_MS, ret);
317 			goto err_firmware_out;
318 		} else if (is_stdalone) {
319 			u32 reg;
320 
321 			reg = vkread32(vk, BAR_0, BAR_BOOT1_STDALONE_PROGRESS);
322 			if ((reg & BOOT1_STDALONE_PROGRESS_MASK) ==
323 				     BOOT1_STDALONE_SUCCESS) {
324 				dev_info(dev, "Boot1 standalone success\n");
325 				ret = 0;
326 			} else {
327 				dev_err(dev, "Timeout %ld ms - Boot1 standalone failure\n",
328 					BOOT1_STARTUP_TIMEOUT_MS);
329 				ret = -EINVAL;
330 				goto err_firmware_out;
331 			}
332 		}
333 	} else if (load_type == VK_IMAGE_TYPE_BOOT2) {
334 		unsigned long timeout;
335 
336 		timeout = jiffies + msecs_to_jiffies(LOAD_IMAGE_TIMEOUT_MS);
337 
338 		/* To send more data to VK than max_buf allowed at a time */
339 		do {
340 			/*
341 			 * Check for ack from card. when Ack is received,
342 			 * it means all the data is received by card.
343 			 * Exit the loop after ack is received.
344 			 */
345 			ret = bcm_vk_wait(vk, BAR_0, BAR_BOOT_STATUS,
346 					  FW_LOADER_ACK_RCVD_ALL_DATA,
347 					  FW_LOADER_ACK_RCVD_ALL_DATA,
348 					  TXFR_COMPLETE_TIMEOUT_MS);
349 			if (ret == 0) {
350 				dev_dbg(dev, "Exit boot2 download\n");
351 				break;
352 			} else if (ret == -EFAULT) {
353 				dev_err(dev, "Error detected during ACK waiting");
354 				goto err_firmware_out;
355 			}
356 
357 			/* exit the loop, if there is no response from card */
358 			if (time_after(jiffies, timeout)) {
359 				dev_err(dev, "Error. No reply from card\n");
360 				ret = -ETIMEDOUT;
361 				goto err_firmware_out;
362 			}
363 
364 			/* Wait for VK to open BAR space to copy new data */
365 			ret = bcm_vk_wait(vk, BAR_0, offset_codepush,
366 					  codepush, 0,
367 					  TXFR_COMPLETE_TIMEOUT_MS);
368 			if (ret == 0) {
369 				offset += max_buf;
370 				ret = request_partial_firmware_into_buf
371 						(&fw,
372 						 filename,
373 						 dev, bufp,
374 						 max_buf,
375 						 offset);
376 				if (ret) {
377 					dev_err(dev,
378 						"Error %d requesting firmware file: %s offset: 0x%zx\n",
379 						ret, filename, offset);
380 					goto err_firmware_out;
381 				}
382 				dev_dbg(dev, "size=0x%zx\n", fw->size);
383 				dev_dbg(dev, "Signaling 0x%x to 0x%llx\n",
384 					codepush, offset_codepush);
385 				vkwrite32(vk, codepush, BAR_0, offset_codepush);
386 				/* reload timeout after every codepush */
387 				timeout = jiffies +
388 				    msecs_to_jiffies(LOAD_IMAGE_TIMEOUT_MS);
389 			} else if (ret == -EFAULT) {
390 				dev_err(dev, "Error detected waiting for transfer\n");
391 				goto err_firmware_out;
392 			}
393 		} while (1);
394 
395 		/* wait for fw status bits to indicate app ready */
396 		ret = bcm_vk_wait(vk, BAR_0, VK_BAR_FWSTS,
397 				  VK_FWSTS_READY,
398 				  VK_FWSTS_READY,
399 				  BOOT2_STARTUP_TIMEOUT_MS);
400 		if (ret < 0) {
401 			dev_err(dev, "Boot2 not ready - ret(%d)\n", ret);
402 			goto err_firmware_out;
403 		}
404 
405 		is_stdalone = vkread32(vk, BAR_0, BAR_BOOT_STATUS) &
406 			      BOOT_STDALONE_RUNNING;
407 		if (!is_stdalone) {
408 			ret = bcm_vk_intf_ver_chk(vk);
409 			if (ret) {
410 				dev_err(dev, "failure in intf version check\n");
411 				goto err_firmware_out;
412 			}
413 
414 			/* sync & channel other info */
415 			ret = bcm_vk_sync_card_info(vk);
416 			if (ret) {
417 				dev_err(dev, "Syncing Card Info failure\n");
418 				goto err_firmware_out;
419 			}
420 		}
421 	}
422 
423 err_firmware_out:
424 	release_firmware(fw);
425 
426 err_buf_out:
427 	if (bufp)
428 		dma_free_coherent(dev, max_buf, bufp, boot_dma_addr);
429 
430 	return ret;
431 }
432 
433 static u32 bcm_vk_next_boot_image(struct bcm_vk *vk)
434 {
435 	u32 boot_status;
436 	u32 fw_status;
437 	u32 load_type = 0;  /* default for unknown */
438 
439 	boot_status = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
440 	fw_status = vkread32(vk, BAR_0, VK_BAR_FWSTS);
441 
442 	if (!BCM_VK_INTF_IS_DOWN(boot_status) && (boot_status & SRAM_OPEN))
443 		load_type = VK_IMAGE_TYPE_BOOT1;
444 	else if (boot_status == BOOT1_RUNNING)
445 		load_type = VK_IMAGE_TYPE_BOOT2;
446 
447 	/* Log status so that we know different stages */
448 	dev_info(&vk->pdev->dev,
449 		 "boot-status value for next image: 0x%x : fw-status 0x%x\n",
450 		 boot_status, fw_status);
451 
452 	return load_type;
453 }
454 
455 static enum soc_idx get_soc_idx(struct bcm_vk *vk)
456 {
457 	struct pci_dev *pdev = vk->pdev;
458 	enum soc_idx idx = VK_IDX_INVALID;
459 	u32 rev;
460 	static enum soc_idx const vk_soc_tab[] = { VALKYRIE_A0, VALKYRIE_B0 };
461 
462 	switch (pdev->device) {
463 	case PCI_DEVICE_ID_VALKYRIE:
464 		/* get the chip id to decide sub-class */
465 		rev = MAJOR_SOC_REV(vkread32(vk, BAR_0, BAR_CHIP_ID));
466 		if (rev < ARRAY_SIZE(vk_soc_tab)) {
467 			idx = vk_soc_tab[rev];
468 		} else {
469 			/* Default to A0 firmware for all other chip revs */
470 			idx = VALKYRIE_A0;
471 			dev_warn(&pdev->dev,
472 				 "Rev %d not in image lookup table, default to idx=%d\n",
473 				 rev, idx);
474 		}
475 		break;
476 
477 	case PCI_DEVICE_ID_VIPER:
478 		idx = VIPER;
479 		break;
480 
481 	default:
482 		dev_err(&pdev->dev, "no images for 0x%x\n", pdev->device);
483 	}
484 	return idx;
485 }
486 
487 static const char *get_load_fw_name(struct bcm_vk *vk,
488 				    const struct load_image_entry *entry)
489 {
490 	const struct firmware *fw;
491 	struct device *dev = &vk->pdev->dev;
492 	int ret;
493 	unsigned long dummy;
494 	int i;
495 
496 	for (i = 0; i < IMG_PER_TYPE_MAX; i++) {
497 		fw = NULL;
498 		ret = request_partial_firmware_into_buf(&fw,
499 							entry->image_name[i],
500 							dev, &dummy,
501 							sizeof(dummy),
502 							0);
503 		release_firmware(fw);
504 		if (!ret)
505 			return entry->image_name[i];
506 	}
507 	return NULL;
508 }
509 
510 int bcm_vk_auto_load_all_images(struct bcm_vk *vk)
511 {
512 	int i, ret = -1;
513 	enum soc_idx idx;
514 	struct device *dev = &vk->pdev->dev;
515 	u32 curr_type;
516 	const char *curr_name;
517 
518 	idx = get_soc_idx(vk);
519 	if (idx == VK_IDX_INVALID)
520 		goto auto_load_all_exit;
521 
522 	/* log a message to know the relative loading order */
523 	dev_dbg(dev, "Load All for device %d\n", vk->devid);
524 
525 	for (i = 0; i < NUM_BOOT_STAGES; i++) {
526 		curr_type = image_tab[idx][i].image_type;
527 		if (bcm_vk_next_boot_image(vk) == curr_type) {
528 			curr_name = get_load_fw_name(vk, &image_tab[idx][i]);
529 			if (!curr_name) {
530 				dev_err(dev, "No suitable firmware exists for type %d",
531 					curr_type);
532 				ret = -ENOENT;
533 				goto auto_load_all_exit;
534 			}
535 			ret = bcm_vk_load_image_by_type(vk, curr_type,
536 							curr_name);
537 			dev_info(dev, "Auto load %s, ret %d\n",
538 				 curr_name, ret);
539 
540 			if (ret) {
541 				dev_err(dev, "Error loading default %s\n",
542 					curr_name);
543 				goto auto_load_all_exit;
544 			}
545 		}
546 	}
547 
548 auto_load_all_exit:
549 	return ret;
550 }
551 
552 static int bcm_vk_trigger_autoload(struct bcm_vk *vk)
553 {
554 	if (test_and_set_bit(BCM_VK_WQ_DWNLD_PEND, vk->wq_offload) != 0)
555 		return -EPERM;
556 
557 	set_bit(BCM_VK_WQ_DWNLD_AUTO, vk->wq_offload);
558 	queue_work(vk->wq_thread, &vk->wq_work);
559 
560 	return 0;
561 }
562 
563 /*
564  * deferred work queue for auto download.
565  */
566 static void bcm_vk_wq_handler(struct work_struct *work)
567 {
568 	struct bcm_vk *vk = container_of(work, struct bcm_vk, wq_work);
569 
570 	if (test_bit(BCM_VK_WQ_DWNLD_AUTO, vk->wq_offload)) {
571 		bcm_vk_auto_load_all_images(vk);
572 
573 		/*
574 		 * at the end of operation, clear AUTO bit and pending
575 		 * bit
576 		 */
577 		clear_bit(BCM_VK_WQ_DWNLD_AUTO, vk->wq_offload);
578 		clear_bit(BCM_VK_WQ_DWNLD_PEND, vk->wq_offload);
579 	}
580 }
581 
582 static void bcm_to_v_reset_doorbell(struct bcm_vk *vk, u32 db_val)
583 {
584 	vkwrite32(vk, db_val, BAR_0, VK_BAR0_RESET_DB_BASE);
585 }
586 
587 static int bcm_vk_trigger_reset(struct bcm_vk *vk)
588 {
589 	u32 i;
590 	u32 value, boot_status;
591 	static const u32 bar0_reg_clr_list[] = { BAR_OS_UPTIME,
592 						 BAR_INTF_VER,
593 						 BAR_CARD_VOLTAGE,
594 						 BAR_CARD_TEMPERATURE,
595 						 BAR_CARD_PWR_AND_THRE };
596 
597 	/* make tag '\0' terminated */
598 	vkwrite32(vk, 0, BAR_1, VK_BAR1_BOOT1_VER_TAG);
599 
600 	for (i = 0; i < VK_BAR1_DAUTH_MAX; i++) {
601 		vkwrite32(vk, 0, BAR_1, VK_BAR1_DAUTH_STORE_ADDR(i));
602 		vkwrite32(vk, 0, BAR_1, VK_BAR1_DAUTH_VALID_ADDR(i));
603 	}
604 	for (i = 0; i < VK_BAR1_SOTP_REVID_MAX; i++)
605 		vkwrite32(vk, 0, BAR_1, VK_BAR1_SOTP_REVID_ADDR(i));
606 
607 	/*
608 	 * When boot request fails, the CODE_PUSH_OFFSET stays persistent.
609 	 * Allowing us to debug the failure. When we call reset,
610 	 * we should clear CODE_PUSH_OFFSET so ROM does not execute
611 	 * boot again (and fails again) and instead waits for a new
612 	 * codepush.  And, if previous boot has encountered error, need
613 	 * to clear the entry values
614 	 */
615 	boot_status = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
616 	if (boot_status & BOOT_ERR_MASK) {
617 		dev_info(&vk->pdev->dev,
618 			 "Card in boot error 0x%x, clear CODEPUSH val\n",
619 			 boot_status);
620 		value = 0;
621 	} else {
622 		value = vkread32(vk, BAR_0, BAR_CODEPUSH_SBL);
623 		value &= CODEPUSH_MASK;
624 	}
625 	vkwrite32(vk, value, BAR_0, BAR_CODEPUSH_SBL);
626 
627 	/* reset fw_status with proper reason, and press db */
628 	vkwrite32(vk, VK_FWSTS_RESET_MBOX_DB, BAR_0, VK_BAR_FWSTS);
629 	bcm_to_v_reset_doorbell(vk, VK_BAR0_RESET_DB_SOFT);
630 
631 	/* clear other necessary registers records */
632 	for (i = 0; i < ARRAY_SIZE(bar0_reg_clr_list); i++)
633 		vkwrite32(vk, 0, BAR_0, bar0_reg_clr_list[i]);
634 
635 	return 0;
636 }
637 
638 static int bcm_vk_on_panic(struct notifier_block *nb,
639 			   unsigned long e, void *p)
640 {
641 	struct bcm_vk *vk = container_of(nb, struct bcm_vk, panic_nb);
642 
643 	bcm_to_v_reset_doorbell(vk, VK_BAR0_RESET_DB_HARD);
644 
645 	return 0;
646 }
647 
648 static int bcm_vk_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
649 {
650 	int err;
651 	int i;
652 	int id;
653 	int irq;
654 	char name[20];
655 	struct bcm_vk *vk;
656 	struct device *dev = &pdev->dev;
657 	struct miscdevice *misc_device;
658 	u32 boot_status;
659 
660 	vk = kzalloc(sizeof(*vk), GFP_KERNEL);
661 	if (!vk)
662 		return -ENOMEM;
663 
664 	err = pci_enable_device(pdev);
665 	if (err) {
666 		dev_err(dev, "Cannot enable PCI device\n");
667 		goto err_free_exit;
668 	}
669 	vk->pdev = pci_dev_get(pdev);
670 
671 	err = pci_request_regions(pdev, DRV_MODULE_NAME);
672 	if (err) {
673 		dev_err(dev, "Cannot obtain PCI resources\n");
674 		goto err_disable_pdev;
675 	}
676 
677 	/* make sure DMA is good */
678 	err = dma_set_mask_and_coherent(&pdev->dev,
679 					DMA_BIT_MASK(BCM_VK_DMA_BITS));
680 	if (err) {
681 		dev_err(dev, "failed to set DMA mask\n");
682 		goto err_disable_pdev;
683 	}
684 
685 	/* The tdma is a scratch area for some DMA testings. */
686 	if (nr_scratch_pages) {
687 		vk->tdma_vaddr = dma_alloc_coherent
688 					(dev,
689 					 nr_scratch_pages * PAGE_SIZE,
690 					 &vk->tdma_addr, GFP_KERNEL);
691 		if (!vk->tdma_vaddr) {
692 			err = -ENOMEM;
693 			goto err_disable_pdev;
694 		}
695 	}
696 
697 	pci_set_master(pdev);
698 	pci_set_drvdata(pdev, vk);
699 
700 	irq = pci_alloc_irq_vectors(pdev,
701 				    1,
702 				    VK_MSIX_IRQ_MAX,
703 				    PCI_IRQ_MSI | PCI_IRQ_MSIX);
704 
705 	if (irq < VK_MSIX_IRQ_MIN_REQ) {
706 		dev_err(dev, "failed to get min %d MSIX interrupts, irq(%d)\n",
707 			VK_MSIX_IRQ_MIN_REQ, irq);
708 		err = (irq >= 0) ? -EINVAL : irq;
709 		goto err_disable_pdev;
710 	}
711 
712 	if (irq != VK_MSIX_IRQ_MAX)
713 		dev_warn(dev, "Number of IRQs %d allocated - requested(%d).\n",
714 			 irq, VK_MSIX_IRQ_MAX);
715 
716 	for (i = 0; i < MAX_BAR; i++) {
717 		/* multiple by 2 for 64 bit BAR mapping */
718 		vk->bar[i] = pci_ioremap_bar(pdev, i * 2);
719 		if (!vk->bar[i]) {
720 			dev_err(dev, "failed to remap BAR%d\n", i);
721 			goto err_iounmap;
722 		}
723 	}
724 
725 	id = ida_simple_get(&bcm_vk_ida, 0, 0, GFP_KERNEL);
726 	if (id < 0) {
727 		err = id;
728 		dev_err(dev, "unable to get id\n");
729 		goto err_iounmap;
730 	}
731 
732 	vk->devid = id;
733 	snprintf(name, sizeof(name), DRV_MODULE_NAME ".%d", id);
734 	misc_device = &vk->miscdev;
735 	misc_device->minor = MISC_DYNAMIC_MINOR;
736 	misc_device->name = kstrdup(name, GFP_KERNEL);
737 	if (!misc_device->name) {
738 		err = -ENOMEM;
739 		goto err_ida_remove;
740 	}
741 
742 	err = misc_register(misc_device);
743 	if (err) {
744 		dev_err(dev, "failed to register device\n");
745 		goto err_kfree_name;
746 	}
747 
748 	INIT_WORK(&vk->wq_work, bcm_vk_wq_handler);
749 
750 	/* create dedicated workqueue */
751 	vk->wq_thread = create_singlethread_workqueue(name);
752 	if (!vk->wq_thread) {
753 		dev_err(dev, "Fail to create workqueue thread\n");
754 		err = -ENOMEM;
755 		goto err_misc_deregister;
756 	}
757 
758 	/* sync other info */
759 	bcm_vk_sync_card_info(vk);
760 
761 	/* register for panic notifier */
762 	vk->panic_nb.notifier_call = bcm_vk_on_panic;
763 	err = atomic_notifier_chain_register(&panic_notifier_list,
764 					     &vk->panic_nb);
765 	if (err) {
766 		dev_err(dev, "Fail to register panic notifier\n");
767 		goto err_destroy_workqueue;
768 	}
769 
770 	/*
771 	 * lets trigger an auto download.  We don't want to do it serially here
772 	 * because at probing time, it is not supposed to block for a long time.
773 	 */
774 	boot_status = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
775 	if (auto_load) {
776 		if ((boot_status & BOOT_STATE_MASK) == BROM_RUNNING) {
777 			if (bcm_vk_trigger_autoload(vk))
778 				goto err_unregister_panic_notifier;
779 		} else {
780 			dev_err(dev,
781 				"Auto-load skipped - BROM not in proper state (0x%x)\n",
782 				boot_status);
783 		}
784 	}
785 
786 	dev_dbg(dev, "BCM-VK:%u created\n", id);
787 
788 	return 0;
789 
790 err_unregister_panic_notifier:
791 	atomic_notifier_chain_unregister(&panic_notifier_list,
792 					 &vk->panic_nb);
793 
794 err_destroy_workqueue:
795 	destroy_workqueue(vk->wq_thread);
796 
797 err_misc_deregister:
798 	misc_deregister(misc_device);
799 
800 err_kfree_name:
801 	kfree(misc_device->name);
802 	misc_device->name = NULL;
803 
804 err_ida_remove:
805 	ida_simple_remove(&bcm_vk_ida, id);
806 
807 err_iounmap:
808 	for (i = 0; i < MAX_BAR; i++) {
809 		if (vk->bar[i])
810 			pci_iounmap(pdev, vk->bar[i]);
811 	}
812 	pci_release_regions(pdev);
813 
814 err_disable_pdev:
815 	if (vk->tdma_vaddr)
816 		dma_free_coherent(&pdev->dev, nr_scratch_pages * PAGE_SIZE,
817 				  vk->tdma_vaddr, vk->tdma_addr);
818 
819 	pci_free_irq_vectors(pdev);
820 	pci_disable_device(pdev);
821 	pci_dev_put(pdev);
822 
823 err_free_exit:
824 	kfree(vk);
825 
826 	return err;
827 }
828 
829 static void bcm_vk_remove(struct pci_dev *pdev)
830 {
831 	int i;
832 	struct bcm_vk *vk = pci_get_drvdata(pdev);
833 	struct miscdevice *misc_device = &vk->miscdev;
834 
835 	/*
836 	 * Trigger a reset to card and wait enough time for UCODE to rerun,
837 	 * which re-initialize the card into its default state.
838 	 * This ensures when driver is re-enumerated it will start from
839 	 * a completely clean state.
840 	 */
841 	bcm_vk_trigger_reset(vk);
842 	usleep_range(BCM_VK_UCODE_BOOT_US, BCM_VK_UCODE_BOOT_MAX_US);
843 
844 	/* unregister panic notifier */
845 	atomic_notifier_chain_unregister(&panic_notifier_list,
846 					 &vk->panic_nb);
847 
848 	if (vk->tdma_vaddr)
849 		dma_free_coherent(&pdev->dev, nr_scratch_pages * PAGE_SIZE,
850 				  vk->tdma_vaddr, vk->tdma_addr);
851 
852 	/* remove if name is set which means misc dev registered */
853 	if (misc_device->name) {
854 		misc_deregister(misc_device);
855 		kfree(misc_device->name);
856 		ida_simple_remove(&bcm_vk_ida, vk->devid);
857 	}
858 
859 	cancel_work_sync(&vk->wq_work);
860 	destroy_workqueue(vk->wq_thread);
861 
862 	for (i = 0; i < MAX_BAR; i++) {
863 		if (vk->bar[i])
864 			pci_iounmap(pdev, vk->bar[i]);
865 	}
866 
867 	dev_dbg(&pdev->dev, "BCM-VK:%d released\n", vk->devid);
868 
869 	pci_release_regions(pdev);
870 	pci_free_irq_vectors(pdev);
871 	pci_disable_device(pdev);
872 }
873 
874 static void bcm_vk_shutdown(struct pci_dev *pdev)
875 {
876 	struct bcm_vk *vk = pci_get_drvdata(pdev);
877 	u32 reg, boot_stat;
878 
879 	reg = vkread32(vk, BAR_0, BAR_BOOT_STATUS);
880 	boot_stat = reg & BOOT_STATE_MASK;
881 
882 	if (boot_stat == BOOT1_RUNNING) {
883 		/* simply trigger a reset interrupt to park it */
884 		bcm_vk_trigger_reset(vk);
885 	} else if (boot_stat == BROM_NOT_RUN) {
886 		int err;
887 		u16 lnksta;
888 
889 		/*
890 		 * The boot status only reflects boot condition since last reset
891 		 * As ucode will run only once to configure pcie, if multiple
892 		 * resets happen, we lost track if ucode has run or not.
893 		 * Here, read the current link speed and use that to
894 		 * sync up the bootstatus properly so that on reboot-back-up,
895 		 * it has the proper state to start with autoload
896 		 */
897 		err = pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnksta);
898 		if (!err &&
899 		    (lnksta & PCI_EXP_LNKSTA_CLS) != PCI_EXP_LNKSTA_CLS_2_5GB) {
900 			reg |= BROM_STATUS_COMPLETE;
901 			vkwrite32(vk, reg, BAR_0, BAR_BOOT_STATUS);
902 		}
903 	}
904 }
905 
906 static const struct pci_device_id bcm_vk_ids[] = {
907 	{ PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, PCI_DEVICE_ID_VALKYRIE), },
908 	{ PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, PCI_DEVICE_ID_VIPER), },
909 	{ }
910 };
911 MODULE_DEVICE_TABLE(pci, bcm_vk_ids);
912 
913 static struct pci_driver pci_driver = {
914 	.name     = DRV_MODULE_NAME,
915 	.id_table = bcm_vk_ids,
916 	.probe    = bcm_vk_probe,
917 	.remove   = bcm_vk_remove,
918 	.shutdown = bcm_vk_shutdown,
919 };
920 module_pci_driver(pci_driver);
921 
922 MODULE_DESCRIPTION("Broadcom VK Host Driver");
923 MODULE_AUTHOR("Scott Branden <scott.branden@broadcom.com>");
924 MODULE_LICENSE("GPL v2");
925 MODULE_VERSION("1.0");
926