1 /*
2  * Driver for the Micron P320 SSD
3  *   Copyright (C) 2011 Micron Technology, Inc.
4  *
5  * Portions of this code were derived from works subjected to the
6  * following copyright:
7  *    Copyright (C) 2009 Integrated Device Technology, Inc.
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License as published by
11  * the Free Software Foundation; either version 2 of the License, or
12  * (at your option) any later version.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  * GNU General Public License for more details.
18  *
19  */
20 
21 #include <linux/pci.h>
22 #include <linux/interrupt.h>
23 #include <linux/ata.h>
24 #include <linux/delay.h>
25 #include <linux/hdreg.h>
26 #include <linux/uaccess.h>
27 #include <linux/random.h>
28 #include <linux/smp.h>
29 #include <linux/compat.h>
30 #include <linux/fs.h>
31 #include <linux/module.h>
32 #include <linux/genhd.h>
33 #include <linux/blkdev.h>
34 #include <linux/blk-mq.h>
35 #include <linux/bio.h>
36 #include <linux/dma-mapping.h>
37 #include <linux/idr.h>
38 #include <linux/kthread.h>
39 #include <../drivers/ata/ahci.h>
40 #include <linux/export.h>
41 #include <linux/debugfs.h>
42 #include <linux/prefetch.h>
43 #include "mtip32xx.h"
44 
45 #define HW_CMD_SLOT_SZ		(MTIP_MAX_COMMAND_SLOTS * 32)
46 
47 /* DMA region containing RX Fis, Identify, RLE10, and SMART buffers */
48 #define AHCI_RX_FIS_SZ          0x100
49 #define AHCI_RX_FIS_OFFSET      0x0
50 #define AHCI_IDFY_SZ            ATA_SECT_SIZE
51 #define AHCI_IDFY_OFFSET        0x400
52 #define AHCI_SECTBUF_SZ         ATA_SECT_SIZE
53 #define AHCI_SECTBUF_OFFSET     0x800
54 #define AHCI_SMARTBUF_SZ        ATA_SECT_SIZE
55 #define AHCI_SMARTBUF_OFFSET    0xC00
56 /* 0x100 + 0x200 + 0x200 + 0x200 is smaller than 4k but we pad it out */
57 #define BLOCK_DMA_ALLOC_SZ      4096
58 
59 /* DMA region containing command table (should be 8192 bytes) */
60 #define AHCI_CMD_SLOT_SZ        sizeof(struct mtip_cmd_hdr)
61 #define AHCI_CMD_TBL_SZ         (MTIP_MAX_COMMAND_SLOTS * AHCI_CMD_SLOT_SZ)
62 #define AHCI_CMD_TBL_OFFSET     0x0
63 
64 /* DMA region per command (contains header and SGL) */
65 #define AHCI_CMD_TBL_HDR_SZ     0x80
66 #define AHCI_CMD_TBL_HDR_OFFSET 0x0
67 #define AHCI_CMD_TBL_SGL_SZ     (MTIP_MAX_SG * sizeof(struct mtip_cmd_sg))
68 #define AHCI_CMD_TBL_SGL_OFFSET AHCI_CMD_TBL_HDR_SZ
69 #define CMD_DMA_ALLOC_SZ        (AHCI_CMD_TBL_SGL_SZ + AHCI_CMD_TBL_HDR_SZ)
70 
71 
72 #define HOST_CAP_NZDMA		(1 << 19)
73 #define HOST_HSORG		0xFC
74 #define HSORG_DISABLE_SLOTGRP_INTR (1<<24)
75 #define HSORG_DISABLE_SLOTGRP_PXIS (1<<16)
76 #define HSORG_HWREV		0xFF00
77 #define HSORG_STYLE		0x8
78 #define HSORG_SLOTGROUPS	0x7
79 
80 #define PORT_COMMAND_ISSUE	0x38
81 #define PORT_SDBV		0x7C
82 
83 #define PORT_OFFSET		0x100
84 #define PORT_MEM_SIZE		0x80
85 
86 #define PORT_IRQ_ERR \
87 	(PORT_IRQ_HBUS_ERR | PORT_IRQ_IF_ERR | PORT_IRQ_CONNECT | \
88 	 PORT_IRQ_PHYRDY | PORT_IRQ_UNK_FIS | PORT_IRQ_BAD_PMP | \
89 	 PORT_IRQ_TF_ERR | PORT_IRQ_HBUS_DATA_ERR | PORT_IRQ_IF_NONFATAL | \
90 	 PORT_IRQ_OVERFLOW)
91 #define PORT_IRQ_LEGACY \
92 	(PORT_IRQ_PIOS_FIS | PORT_IRQ_D2H_REG_FIS)
93 #define PORT_IRQ_HANDLED \
94 	(PORT_IRQ_SDB_FIS | PORT_IRQ_LEGACY | \
95 	 PORT_IRQ_TF_ERR | PORT_IRQ_IF_ERR | \
96 	 PORT_IRQ_CONNECT | PORT_IRQ_PHYRDY)
97 #define DEF_PORT_IRQ \
98 	(PORT_IRQ_ERR | PORT_IRQ_LEGACY | PORT_IRQ_SDB_FIS)
99 
100 /* product numbers */
101 #define MTIP_PRODUCT_UNKNOWN	0x00
102 #define MTIP_PRODUCT_ASICFPGA	0x11
103 
104 /* Device instance number, incremented each time a device is probed. */
105 static int instance;
106 
107 static struct list_head online_list;
108 static struct list_head removing_list;
109 static spinlock_t dev_lock;
110 
111 /*
112  * Global variable used to hold the major block device number
113  * allocated in mtip_init().
114  */
115 static int mtip_major;
116 static struct dentry *dfs_parent;
117 static struct dentry *dfs_device_status;
118 
119 static u32 cpu_use[NR_CPUS];
120 
121 static DEFINE_SPINLOCK(rssd_index_lock);
122 static DEFINE_IDA(rssd_index_ida);
123 
124 static int mtip_block_initialize(struct driver_data *dd);
125 
126 #ifdef CONFIG_COMPAT
127 struct mtip_compat_ide_task_request_s {
128 	__u8		io_ports[8];
129 	__u8		hob_ports[8];
130 	ide_reg_valid_t	out_flags;
131 	ide_reg_valid_t	in_flags;
132 	int		data_phase;
133 	int		req_cmd;
134 	compat_ulong_t	out_size;
135 	compat_ulong_t	in_size;
136 };
137 #endif
138 
139 /*
140  * This function check_for_surprise_removal is called
141  * while card is removed from the system and it will
142  * read the vendor id from the configration space
143  *
144  * @pdev Pointer to the pci_dev structure.
145  *
146  * return value
147  *	 true if device removed, else false
148  */
149 static bool mtip_check_surprise_removal(struct pci_dev *pdev)
150 {
151 	u16 vendor_id = 0;
152 	struct driver_data *dd = pci_get_drvdata(pdev);
153 
154 	if (dd->sr)
155 		return true;
156 
157        /* Read the vendorID from the configuration space */
158 	pci_read_config_word(pdev, 0x00, &vendor_id);
159 	if (vendor_id == 0xFFFF) {
160 		dd->sr = true;
161 		if (dd->queue)
162 			blk_queue_flag_set(QUEUE_FLAG_DEAD, dd->queue);
163 		else
164 			dev_warn(&dd->pdev->dev,
165 				"%s: dd->queue is NULL\n", __func__);
166 		return true; /* device removed */
167 	}
168 
169 	return false; /* device present */
170 }
171 
172 /* we have to use runtime tag to setup command header */
173 static void mtip_init_cmd_header(struct request *rq)
174 {
175 	struct driver_data *dd = rq->q->queuedata;
176 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
177 
178 	/* Point the command headers at the command tables. */
179 	cmd->command_header = dd->port->command_list +
180 				(sizeof(struct mtip_cmd_hdr) * rq->tag);
181 	cmd->command_header_dma = dd->port->command_list_dma +
182 				(sizeof(struct mtip_cmd_hdr) * rq->tag);
183 
184 	if (test_bit(MTIP_PF_HOST_CAP_64, &dd->port->flags))
185 		cmd->command_header->ctbau = __force_bit2int cpu_to_le32((cmd->command_dma >> 16) >> 16);
186 
187 	cmd->command_header->ctba = __force_bit2int cpu_to_le32(cmd->command_dma & 0xFFFFFFFF);
188 }
189 
190 static struct mtip_cmd *mtip_get_int_command(struct driver_data *dd)
191 {
192 	struct request *rq;
193 
194 	if (mtip_check_surprise_removal(dd->pdev))
195 		return NULL;
196 
197 	rq = blk_mq_alloc_request(dd->queue, REQ_OP_DRV_IN, BLK_MQ_REQ_RESERVED);
198 	if (IS_ERR(rq))
199 		return NULL;
200 
201 	/* Internal cmd isn't submitted via .queue_rq */
202 	mtip_init_cmd_header(rq);
203 
204 	return blk_mq_rq_to_pdu(rq);
205 }
206 
207 static struct mtip_cmd *mtip_cmd_from_tag(struct driver_data *dd,
208 					  unsigned int tag)
209 {
210 	struct blk_mq_hw_ctx *hctx = dd->queue->queue_hw_ctx[0];
211 
212 	return blk_mq_rq_to_pdu(blk_mq_tag_to_rq(hctx->tags, tag));
213 }
214 
215 /*
216  * Reset the HBA (without sleeping)
217  *
218  * @dd Pointer to the driver data structure.
219  *
220  * return value
221  *	0	The reset was successful.
222  *	-1	The HBA Reset bit did not clear.
223  */
224 static int mtip_hba_reset(struct driver_data *dd)
225 {
226 	unsigned long timeout;
227 
228 	/* Set the reset bit */
229 	writel(HOST_RESET, dd->mmio + HOST_CTL);
230 
231 	/* Flush */
232 	readl(dd->mmio + HOST_CTL);
233 
234 	/*
235 	 * Spin for up to 10 seconds waiting for reset acknowledgement. Spec
236 	 * is 1 sec but in LUN failure conditions, up to 10 secs are required
237 	 */
238 	timeout = jiffies + msecs_to_jiffies(10000);
239 	do {
240 		mdelay(10);
241 		if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag))
242 			return -1;
243 
244 	} while ((readl(dd->mmio + HOST_CTL) & HOST_RESET)
245 		 && time_before(jiffies, timeout));
246 
247 	if (readl(dd->mmio + HOST_CTL) & HOST_RESET)
248 		return -1;
249 
250 	return 0;
251 }
252 
253 /*
254  * Issue a command to the hardware.
255  *
256  * Set the appropriate bit in the s_active and Command Issue hardware
257  * registers, causing hardware command processing to begin.
258  *
259  * @port Pointer to the port structure.
260  * @tag  The tag of the command to be issued.
261  *
262  * return value
263  *      None
264  */
265 static inline void mtip_issue_ncq_command(struct mtip_port *port, int tag)
266 {
267 	int group = tag >> 5;
268 
269 	/* guard SACT and CI registers */
270 	spin_lock(&port->cmd_issue_lock[group]);
271 	writel((1 << MTIP_TAG_BIT(tag)),
272 			port->s_active[MTIP_TAG_INDEX(tag)]);
273 	writel((1 << MTIP_TAG_BIT(tag)),
274 			port->cmd_issue[MTIP_TAG_INDEX(tag)]);
275 	spin_unlock(&port->cmd_issue_lock[group]);
276 }
277 
278 /*
279  * Enable/disable the reception of FIS
280  *
281  * @port   Pointer to the port data structure
282  * @enable 1 to enable, 0 to disable
283  *
284  * return value
285  *	Previous state: 1 enabled, 0 disabled
286  */
287 static int mtip_enable_fis(struct mtip_port *port, int enable)
288 {
289 	u32 tmp;
290 
291 	/* enable FIS reception */
292 	tmp = readl(port->mmio + PORT_CMD);
293 	if (enable)
294 		writel(tmp | PORT_CMD_FIS_RX, port->mmio + PORT_CMD);
295 	else
296 		writel(tmp & ~PORT_CMD_FIS_RX, port->mmio + PORT_CMD);
297 
298 	/* Flush */
299 	readl(port->mmio + PORT_CMD);
300 
301 	return (((tmp & PORT_CMD_FIS_RX) == PORT_CMD_FIS_RX));
302 }
303 
304 /*
305  * Enable/disable the DMA engine
306  *
307  * @port   Pointer to the port data structure
308  * @enable 1 to enable, 0 to disable
309  *
310  * return value
311  *	Previous state: 1 enabled, 0 disabled.
312  */
313 static int mtip_enable_engine(struct mtip_port *port, int enable)
314 {
315 	u32 tmp;
316 
317 	/* enable FIS reception */
318 	tmp = readl(port->mmio + PORT_CMD);
319 	if (enable)
320 		writel(tmp | PORT_CMD_START, port->mmio + PORT_CMD);
321 	else
322 		writel(tmp & ~PORT_CMD_START, port->mmio + PORT_CMD);
323 
324 	readl(port->mmio + PORT_CMD);
325 	return (((tmp & PORT_CMD_START) == PORT_CMD_START));
326 }
327 
328 /*
329  * Enables the port DMA engine and FIS reception.
330  *
331  * return value
332  *	None
333  */
334 static inline void mtip_start_port(struct mtip_port *port)
335 {
336 	/* Enable FIS reception */
337 	mtip_enable_fis(port, 1);
338 
339 	/* Enable the DMA engine */
340 	mtip_enable_engine(port, 1);
341 }
342 
343 /*
344  * Deinitialize a port by disabling port interrupts, the DMA engine,
345  * and FIS reception.
346  *
347  * @port Pointer to the port structure
348  *
349  * return value
350  *	None
351  */
352 static inline void mtip_deinit_port(struct mtip_port *port)
353 {
354 	/* Disable interrupts on this port */
355 	writel(0, port->mmio + PORT_IRQ_MASK);
356 
357 	/* Disable the DMA engine */
358 	mtip_enable_engine(port, 0);
359 
360 	/* Disable FIS reception */
361 	mtip_enable_fis(port, 0);
362 }
363 
364 /*
365  * Initialize a port.
366  *
367  * This function deinitializes the port by calling mtip_deinit_port() and
368  * then initializes it by setting the command header and RX FIS addresses,
369  * clearing the SError register and any pending port interrupts before
370  * re-enabling the default set of port interrupts.
371  *
372  * @port Pointer to the port structure.
373  *
374  * return value
375  *	None
376  */
377 static void mtip_init_port(struct mtip_port *port)
378 {
379 	int i;
380 	mtip_deinit_port(port);
381 
382 	/* Program the command list base and FIS base addresses */
383 	if (readl(port->dd->mmio + HOST_CAP) & HOST_CAP_64) {
384 		writel((port->command_list_dma >> 16) >> 16,
385 			 port->mmio + PORT_LST_ADDR_HI);
386 		writel((port->rxfis_dma >> 16) >> 16,
387 			 port->mmio + PORT_FIS_ADDR_HI);
388 		set_bit(MTIP_PF_HOST_CAP_64, &port->flags);
389 	}
390 
391 	writel(port->command_list_dma & 0xFFFFFFFF,
392 			port->mmio + PORT_LST_ADDR);
393 	writel(port->rxfis_dma & 0xFFFFFFFF, port->mmio + PORT_FIS_ADDR);
394 
395 	/* Clear SError */
396 	writel(readl(port->mmio + PORT_SCR_ERR), port->mmio + PORT_SCR_ERR);
397 
398 	/* reset the completed registers.*/
399 	for (i = 0; i < port->dd->slot_groups; i++)
400 		writel(0xFFFFFFFF, port->completed[i]);
401 
402 	/* Clear any pending interrupts for this port */
403 	writel(readl(port->mmio + PORT_IRQ_STAT), port->mmio + PORT_IRQ_STAT);
404 
405 	/* Clear any pending interrupts on the HBA. */
406 	writel(readl(port->dd->mmio + HOST_IRQ_STAT),
407 					port->dd->mmio + HOST_IRQ_STAT);
408 
409 	/* Enable port interrupts */
410 	writel(DEF_PORT_IRQ, port->mmio + PORT_IRQ_MASK);
411 }
412 
413 /*
414  * Restart a port
415  *
416  * @port Pointer to the port data structure.
417  *
418  * return value
419  *	None
420  */
421 static void mtip_restart_port(struct mtip_port *port)
422 {
423 	unsigned long timeout;
424 
425 	/* Disable the DMA engine */
426 	mtip_enable_engine(port, 0);
427 
428 	/* Chip quirk: wait up to 500ms for PxCMD.CR == 0 */
429 	timeout = jiffies + msecs_to_jiffies(500);
430 	while ((readl(port->mmio + PORT_CMD) & PORT_CMD_LIST_ON)
431 		 && time_before(jiffies, timeout))
432 		;
433 
434 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag))
435 		return;
436 
437 	/*
438 	 * Chip quirk: escalate to hba reset if
439 	 * PxCMD.CR not clear after 500 ms
440 	 */
441 	if (readl(port->mmio + PORT_CMD) & PORT_CMD_LIST_ON) {
442 		dev_warn(&port->dd->pdev->dev,
443 			"PxCMD.CR not clear, escalating reset\n");
444 
445 		if (mtip_hba_reset(port->dd))
446 			dev_err(&port->dd->pdev->dev,
447 				"HBA reset escalation failed.\n");
448 
449 		/* 30 ms delay before com reset to quiesce chip */
450 		mdelay(30);
451 	}
452 
453 	dev_warn(&port->dd->pdev->dev, "Issuing COM reset\n");
454 
455 	/* Set PxSCTL.DET */
456 	writel(readl(port->mmio + PORT_SCR_CTL) |
457 			 1, port->mmio + PORT_SCR_CTL);
458 	readl(port->mmio + PORT_SCR_CTL);
459 
460 	/* Wait 1 ms to quiesce chip function */
461 	timeout = jiffies + msecs_to_jiffies(1);
462 	while (time_before(jiffies, timeout))
463 		;
464 
465 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag))
466 		return;
467 
468 	/* Clear PxSCTL.DET */
469 	writel(readl(port->mmio + PORT_SCR_CTL) & ~1,
470 			 port->mmio + PORT_SCR_CTL);
471 	readl(port->mmio + PORT_SCR_CTL);
472 
473 	/* Wait 500 ms for bit 0 of PORT_SCR_STS to be set */
474 	timeout = jiffies + msecs_to_jiffies(500);
475 	while (((readl(port->mmio + PORT_SCR_STAT) & 0x01) == 0)
476 			 && time_before(jiffies, timeout))
477 		;
478 
479 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag))
480 		return;
481 
482 	if ((readl(port->mmio + PORT_SCR_STAT) & 0x01) == 0)
483 		dev_warn(&port->dd->pdev->dev,
484 			"COM reset failed\n");
485 
486 	mtip_init_port(port);
487 	mtip_start_port(port);
488 
489 }
490 
491 static int mtip_device_reset(struct driver_data *dd)
492 {
493 	int rv = 0;
494 
495 	if (mtip_check_surprise_removal(dd->pdev))
496 		return 0;
497 
498 	if (mtip_hba_reset(dd) < 0)
499 		rv = -EFAULT;
500 
501 	mdelay(1);
502 	mtip_init_port(dd->port);
503 	mtip_start_port(dd->port);
504 
505 	/* Enable interrupts on the HBA. */
506 	writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN,
507 					dd->mmio + HOST_CTL);
508 	return rv;
509 }
510 
511 /*
512  * Helper function for tag logging
513  */
514 static void print_tags(struct driver_data *dd,
515 			char *msg,
516 			unsigned long *tagbits,
517 			int cnt)
518 {
519 	unsigned char tagmap[128];
520 	int group, tagmap_len = 0;
521 
522 	memset(tagmap, 0, sizeof(tagmap));
523 	for (group = SLOTBITS_IN_LONGS; group > 0; group--)
524 		tagmap_len += sprintf(tagmap + tagmap_len, "%016lX ",
525 						tagbits[group-1]);
526 	dev_warn(&dd->pdev->dev,
527 			"%d command(s) %s: tagmap [%s]", cnt, msg, tagmap);
528 }
529 
530 static int mtip_read_log_page(struct mtip_port *port, u8 page, u16 *buffer,
531 				dma_addr_t buffer_dma, unsigned int sectors);
532 static int mtip_get_smart_attr(struct mtip_port *port, unsigned int id,
533 						struct smart_attr *attrib);
534 
535 static void mtip_complete_command(struct mtip_cmd *cmd, blk_status_t status)
536 {
537 	struct request *req = blk_mq_rq_from_pdu(cmd);
538 
539 	cmd->status = status;
540 	blk_mq_complete_request(req);
541 }
542 
543 /*
544  * Handle an error.
545  *
546  * @dd Pointer to the DRIVER_DATA structure.
547  *
548  * return value
549  *	None
550  */
551 static void mtip_handle_tfe(struct driver_data *dd)
552 {
553 	int group, tag, bit, reissue, rv;
554 	struct mtip_port *port;
555 	struct mtip_cmd  *cmd;
556 	u32 completed;
557 	struct host_to_dev_fis *fis;
558 	unsigned long tagaccum[SLOTBITS_IN_LONGS];
559 	unsigned int cmd_cnt = 0;
560 	unsigned char *buf;
561 	char *fail_reason = NULL;
562 	int fail_all_ncq_write = 0, fail_all_ncq_cmds = 0;
563 
564 	dev_warn(&dd->pdev->dev, "Taskfile error\n");
565 
566 	port = dd->port;
567 
568 	if (test_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags)) {
569 		cmd = mtip_cmd_from_tag(dd, MTIP_TAG_INTERNAL);
570 		dbg_printk(MTIP_DRV_NAME " TFE for the internal command\n");
571 		mtip_complete_command(cmd, BLK_STS_IOERR);
572 		return;
573 	}
574 
575 	/* clear the tag accumulator */
576 	memset(tagaccum, 0, SLOTBITS_IN_LONGS * sizeof(long));
577 
578 	/* Loop through all the groups */
579 	for (group = 0; group < dd->slot_groups; group++) {
580 		completed = readl(port->completed[group]);
581 
582 		dev_warn(&dd->pdev->dev, "g=%u, comp=%x\n", group, completed);
583 
584 		/* clear completed status register in the hardware.*/
585 		writel(completed, port->completed[group]);
586 
587 		/* Process successfully completed commands */
588 		for (bit = 0; bit < 32 && completed; bit++) {
589 			if (!(completed & (1<<bit)))
590 				continue;
591 			tag = (group << 5) + bit;
592 
593 			/* Skip the internal command slot */
594 			if (tag == MTIP_TAG_INTERNAL)
595 				continue;
596 
597 			cmd = mtip_cmd_from_tag(dd, tag);
598 			mtip_complete_command(cmd, 0);
599 			set_bit(tag, tagaccum);
600 			cmd_cnt++;
601 		}
602 	}
603 
604 	print_tags(dd, "completed (TFE)", tagaccum, cmd_cnt);
605 
606 	/* Restart the port */
607 	mdelay(20);
608 	mtip_restart_port(port);
609 
610 	/* Trying to determine the cause of the error */
611 	rv = mtip_read_log_page(dd->port, ATA_LOG_SATA_NCQ,
612 				dd->port->log_buf,
613 				dd->port->log_buf_dma, 1);
614 	if (rv) {
615 		dev_warn(&dd->pdev->dev,
616 			"Error in READ LOG EXT (10h) command\n");
617 		/* non-critical error, don't fail the load */
618 	} else {
619 		buf = (unsigned char *)dd->port->log_buf;
620 		if (buf[259] & 0x1) {
621 			dev_info(&dd->pdev->dev,
622 				"Write protect bit is set.\n");
623 			set_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag);
624 			fail_all_ncq_write = 1;
625 			fail_reason = "write protect";
626 		}
627 		if (buf[288] == 0xF7) {
628 			dev_info(&dd->pdev->dev,
629 				"Exceeded Tmax, drive in thermal shutdown.\n");
630 			set_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag);
631 			fail_all_ncq_cmds = 1;
632 			fail_reason = "thermal shutdown";
633 		}
634 		if (buf[288] == 0xBF) {
635 			set_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag);
636 			dev_info(&dd->pdev->dev,
637 				"Drive indicates rebuild has failed. Secure erase required.\n");
638 			fail_all_ncq_cmds = 1;
639 			fail_reason = "rebuild failed";
640 		}
641 	}
642 
643 	/* clear the tag accumulator */
644 	memset(tagaccum, 0, SLOTBITS_IN_LONGS * sizeof(long));
645 
646 	/* Loop through all the groups */
647 	for (group = 0; group < dd->slot_groups; group++) {
648 		for (bit = 0; bit < 32; bit++) {
649 			reissue = 1;
650 			tag = (group << 5) + bit;
651 			cmd = mtip_cmd_from_tag(dd, tag);
652 
653 			fis = (struct host_to_dev_fis *)cmd->command;
654 
655 			/* Should re-issue? */
656 			if (tag == MTIP_TAG_INTERNAL ||
657 			    fis->command == ATA_CMD_SET_FEATURES)
658 				reissue = 0;
659 			else {
660 				if (fail_all_ncq_cmds ||
661 					(fail_all_ncq_write &&
662 					fis->command == ATA_CMD_FPDMA_WRITE)) {
663 					dev_warn(&dd->pdev->dev,
664 					"  Fail: %s w/tag %d [%s].\n",
665 					fis->command == ATA_CMD_FPDMA_WRITE ?
666 						"write" : "read",
667 					tag,
668 					fail_reason != NULL ?
669 						fail_reason : "unknown");
670 					mtip_complete_command(cmd, BLK_STS_MEDIUM);
671 					continue;
672 				}
673 			}
674 
675 			/*
676 			 * First check if this command has
677 			 *  exceeded its retries.
678 			 */
679 			if (reissue && (cmd->retries-- > 0)) {
680 
681 				set_bit(tag, tagaccum);
682 
683 				/* Re-issue the command. */
684 				mtip_issue_ncq_command(port, tag);
685 
686 				continue;
687 			}
688 
689 			/* Retire a command that will not be reissued */
690 			dev_warn(&port->dd->pdev->dev,
691 				"retiring tag %d\n", tag);
692 
693 			mtip_complete_command(cmd, BLK_STS_IOERR);
694 		}
695 	}
696 	print_tags(dd, "reissued (TFE)", tagaccum, cmd_cnt);
697 }
698 
699 /*
700  * Handle a set device bits interrupt
701  */
702 static inline void mtip_workq_sdbfx(struct mtip_port *port, int group,
703 							u32 completed)
704 {
705 	struct driver_data *dd = port->dd;
706 	int tag, bit;
707 	struct mtip_cmd *command;
708 
709 	if (!completed) {
710 		WARN_ON_ONCE(!completed);
711 		return;
712 	}
713 	/* clear completed status register in the hardware.*/
714 	writel(completed, port->completed[group]);
715 
716 	/* Process completed commands. */
717 	for (bit = 0; (bit < 32) && completed; bit++) {
718 		if (completed & 0x01) {
719 			tag = (group << 5) | bit;
720 
721 			/* skip internal command slot. */
722 			if (unlikely(tag == MTIP_TAG_INTERNAL))
723 				continue;
724 
725 			command = mtip_cmd_from_tag(dd, tag);
726 			mtip_complete_command(command, 0);
727 		}
728 		completed >>= 1;
729 	}
730 
731 	/* If last, re-enable interrupts */
732 	if (atomic_dec_return(&dd->irq_workers_active) == 0)
733 		writel(0xffffffff, dd->mmio + HOST_IRQ_STAT);
734 }
735 
736 /*
737  * Process legacy pio and d2h interrupts
738  */
739 static inline void mtip_process_legacy(struct driver_data *dd, u32 port_stat)
740 {
741 	struct mtip_port *port = dd->port;
742 	struct mtip_cmd *cmd = mtip_cmd_from_tag(dd, MTIP_TAG_INTERNAL);
743 
744 	if (test_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags) && cmd) {
745 		int group = MTIP_TAG_INDEX(MTIP_TAG_INTERNAL);
746 		int status = readl(port->cmd_issue[group]);
747 
748 		if (!(status & (1 << MTIP_TAG_BIT(MTIP_TAG_INTERNAL))))
749 			mtip_complete_command(cmd, 0);
750 	}
751 }
752 
753 /*
754  * Demux and handle errors
755  */
756 static inline void mtip_process_errors(struct driver_data *dd, u32 port_stat)
757 {
758 	if (unlikely(port_stat & PORT_IRQ_CONNECT)) {
759 		dev_warn(&dd->pdev->dev,
760 			"Clearing PxSERR.DIAG.x\n");
761 		writel((1 << 26), dd->port->mmio + PORT_SCR_ERR);
762 	}
763 
764 	if (unlikely(port_stat & PORT_IRQ_PHYRDY)) {
765 		dev_warn(&dd->pdev->dev,
766 			"Clearing PxSERR.DIAG.n\n");
767 		writel((1 << 16), dd->port->mmio + PORT_SCR_ERR);
768 	}
769 
770 	if (unlikely(port_stat & ~PORT_IRQ_HANDLED)) {
771 		dev_warn(&dd->pdev->dev,
772 			"Port stat errors %x unhandled\n",
773 			(port_stat & ~PORT_IRQ_HANDLED));
774 		if (mtip_check_surprise_removal(dd->pdev))
775 			return;
776 	}
777 	if (likely(port_stat & (PORT_IRQ_TF_ERR | PORT_IRQ_IF_ERR))) {
778 		set_bit(MTIP_PF_EH_ACTIVE_BIT, &dd->port->flags);
779 		wake_up_interruptible(&dd->port->svc_wait);
780 	}
781 }
782 
783 static inline irqreturn_t mtip_handle_irq(struct driver_data *data)
784 {
785 	struct driver_data *dd = (struct driver_data *) data;
786 	struct mtip_port *port = dd->port;
787 	u32 hba_stat, port_stat;
788 	int rv = IRQ_NONE;
789 	int do_irq_enable = 1, i, workers;
790 	struct mtip_work *twork;
791 
792 	hba_stat = readl(dd->mmio + HOST_IRQ_STAT);
793 	if (hba_stat) {
794 		rv = IRQ_HANDLED;
795 
796 		/* Acknowledge the interrupt status on the port.*/
797 		port_stat = readl(port->mmio + PORT_IRQ_STAT);
798 		if (unlikely(port_stat == 0xFFFFFFFF)) {
799 			mtip_check_surprise_removal(dd->pdev);
800 			return IRQ_HANDLED;
801 		}
802 		writel(port_stat, port->mmio + PORT_IRQ_STAT);
803 
804 		/* Demux port status */
805 		if (likely(port_stat & PORT_IRQ_SDB_FIS)) {
806 			do_irq_enable = 0;
807 			WARN_ON_ONCE(atomic_read(&dd->irq_workers_active) != 0);
808 
809 			/* Start at 1: group zero is always local? */
810 			for (i = 0, workers = 0; i < MTIP_MAX_SLOT_GROUPS;
811 									i++) {
812 				twork = &dd->work[i];
813 				twork->completed = readl(port->completed[i]);
814 				if (twork->completed)
815 					workers++;
816 			}
817 
818 			atomic_set(&dd->irq_workers_active, workers);
819 			if (workers) {
820 				for (i = 1; i < MTIP_MAX_SLOT_GROUPS; i++) {
821 					twork = &dd->work[i];
822 					if (twork->completed)
823 						queue_work_on(
824 							twork->cpu_binding,
825 							dd->isr_workq,
826 							&twork->work);
827 				}
828 
829 				if (likely(dd->work[0].completed))
830 					mtip_workq_sdbfx(port, 0,
831 							dd->work[0].completed);
832 
833 			} else {
834 				/*
835 				 * Chip quirk: SDB interrupt but nothing
836 				 * to complete
837 				 */
838 				do_irq_enable = 1;
839 			}
840 		}
841 
842 		if (unlikely(port_stat & PORT_IRQ_ERR)) {
843 			if (unlikely(mtip_check_surprise_removal(dd->pdev))) {
844 				/* don't proceed further */
845 				return IRQ_HANDLED;
846 			}
847 			if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
848 							&dd->dd_flag))
849 				return rv;
850 
851 			mtip_process_errors(dd, port_stat & PORT_IRQ_ERR);
852 		}
853 
854 		if (unlikely(port_stat & PORT_IRQ_LEGACY))
855 			mtip_process_legacy(dd, port_stat & PORT_IRQ_LEGACY);
856 	}
857 
858 	/* acknowledge interrupt */
859 	if (unlikely(do_irq_enable))
860 		writel(hba_stat, dd->mmio + HOST_IRQ_STAT);
861 
862 	return rv;
863 }
864 
865 /*
866  * HBA interrupt subroutine.
867  *
868  * @irq		IRQ number.
869  * @instance	Pointer to the driver data structure.
870  *
871  * return value
872  *	IRQ_HANDLED	A HBA interrupt was pending and handled.
873  *	IRQ_NONE	This interrupt was not for the HBA.
874  */
875 static irqreturn_t mtip_irq_handler(int irq, void *instance)
876 {
877 	struct driver_data *dd = instance;
878 
879 	return mtip_handle_irq(dd);
880 }
881 
882 static void mtip_issue_non_ncq_command(struct mtip_port *port, int tag)
883 {
884 	writel(1 << MTIP_TAG_BIT(tag), port->cmd_issue[MTIP_TAG_INDEX(tag)]);
885 }
886 
887 static bool mtip_pause_ncq(struct mtip_port *port,
888 				struct host_to_dev_fis *fis)
889 {
890 	unsigned long task_file_data;
891 
892 	task_file_data = readl(port->mmio+PORT_TFDATA);
893 	if ((task_file_data & 1))
894 		return false;
895 
896 	if (fis->command == ATA_CMD_SEC_ERASE_PREP) {
897 		port->ic_pause_timer = jiffies;
898 		return true;
899 	} else if ((fis->command == ATA_CMD_DOWNLOAD_MICRO) &&
900 					(fis->features == 0x03)) {
901 		set_bit(MTIP_PF_DM_ACTIVE_BIT, &port->flags);
902 		port->ic_pause_timer = jiffies;
903 		return true;
904 	} else if ((fis->command == ATA_CMD_SEC_ERASE_UNIT) ||
905 		((fis->command == 0xFC) &&
906 			(fis->features == 0x27 || fis->features == 0x72 ||
907 			 fis->features == 0x62 || fis->features == 0x26))) {
908 		clear_bit(MTIP_DDF_SEC_LOCK_BIT, &port->dd->dd_flag);
909 		clear_bit(MTIP_DDF_REBUILD_FAILED_BIT, &port->dd->dd_flag);
910 		/* Com reset after secure erase or lowlevel format */
911 		mtip_restart_port(port);
912 		clear_bit(MTIP_PF_SE_ACTIVE_BIT, &port->flags);
913 		return false;
914 	}
915 
916 	return false;
917 }
918 
919 static bool mtip_commands_active(struct mtip_port *port)
920 {
921 	unsigned int active;
922 	unsigned int n;
923 
924 	/*
925 	 * Ignore s_active bit 0 of array element 0.
926 	 * This bit will always be set
927 	 */
928 	active = readl(port->s_active[0]) & 0xFFFFFFFE;
929 	for (n = 1; n < port->dd->slot_groups; n++)
930 		active |= readl(port->s_active[n]);
931 
932 	return active != 0;
933 }
934 
935 /*
936  * Wait for port to quiesce
937  *
938  * @port    Pointer to port data structure
939  * @timeout Max duration to wait (ms)
940  *
941  * return value
942  *	0	Success
943  *	-EBUSY  Commands still active
944  */
945 static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout)
946 {
947 	unsigned long to;
948 	bool active = true;
949 
950 	blk_mq_quiesce_queue(port->dd->queue);
951 
952 	to = jiffies + msecs_to_jiffies(timeout);
953 	do {
954 		if (test_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags) &&
955 			test_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags)) {
956 			msleep(20);
957 			continue; /* svc thd is actively issuing commands */
958 		}
959 
960 		msleep(100);
961 
962 		if (mtip_check_surprise_removal(port->dd->pdev))
963 			goto err_fault;
964 
965 		active = mtip_commands_active(port);
966 		if (!active)
967 			break;
968 	} while (time_before(jiffies, to));
969 
970 	blk_mq_unquiesce_queue(port->dd->queue);
971 	return active ? -EBUSY : 0;
972 err_fault:
973 	blk_mq_unquiesce_queue(port->dd->queue);
974 	return -EFAULT;
975 }
976 
977 struct mtip_int_cmd {
978 	int fis_len;
979 	dma_addr_t buffer;
980 	int buf_len;
981 	u32 opts;
982 };
983 
984 /*
985  * Execute an internal command and wait for the completion.
986  *
987  * @port    Pointer to the port data structure.
988  * @fis     Pointer to the FIS that describes the command.
989  * @fis_len  Length in WORDS of the FIS.
990  * @buffer  DMA accessible for command data.
991  * @buf_len  Length, in bytes, of the data buffer.
992  * @opts    Command header options, excluding the FIS length
993  *             and the number of PRD entries.
994  * @timeout Time in ms to wait for the command to complete.
995  *
996  * return value
997  *	0	 Command completed successfully.
998  *	-EFAULT  The buffer address is not correctly aligned.
999  *	-EBUSY   Internal command or other IO in progress.
1000  *	-EAGAIN  Time out waiting for command to complete.
1001  */
1002 static int mtip_exec_internal_command(struct mtip_port *port,
1003 					struct host_to_dev_fis *fis,
1004 					int fis_len,
1005 					dma_addr_t buffer,
1006 					int buf_len,
1007 					u32 opts,
1008 					unsigned long timeout)
1009 {
1010 	struct mtip_cmd *int_cmd;
1011 	struct driver_data *dd = port->dd;
1012 	struct request *rq;
1013 	struct mtip_int_cmd icmd = {
1014 		.fis_len = fis_len,
1015 		.buffer = buffer,
1016 		.buf_len = buf_len,
1017 		.opts = opts
1018 	};
1019 	int rv = 0;
1020 
1021 	/* Make sure the buffer is 8 byte aligned. This is asic specific. */
1022 	if (buffer & 0x00000007) {
1023 		dev_err(&dd->pdev->dev, "SG buffer is not 8 byte aligned\n");
1024 		return -EFAULT;
1025 	}
1026 
1027 	int_cmd = mtip_get_int_command(dd);
1028 	if (!int_cmd) {
1029 		dbg_printk(MTIP_DRV_NAME "Unable to allocate tag for PIO cmd\n");
1030 		return -EFAULT;
1031 	}
1032 	rq = blk_mq_rq_from_pdu(int_cmd);
1033 	rq->special = &icmd;
1034 
1035 	set_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags);
1036 
1037 	if (fis->command == ATA_CMD_SEC_ERASE_PREP)
1038 		set_bit(MTIP_PF_SE_ACTIVE_BIT, &port->flags);
1039 
1040 	clear_bit(MTIP_PF_DM_ACTIVE_BIT, &port->flags);
1041 
1042 	if (fis->command != ATA_CMD_STANDBYNOW1) {
1043 		/* wait for io to complete if non atomic */
1044 		if (mtip_quiesce_io(port, MTIP_QUIESCE_IO_TIMEOUT_MS) < 0) {
1045 			dev_warn(&dd->pdev->dev, "Failed to quiesce IO\n");
1046 			blk_mq_free_request(rq);
1047 			clear_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags);
1048 			wake_up_interruptible(&port->svc_wait);
1049 			return -EBUSY;
1050 		}
1051 	}
1052 
1053 	/* Copy the command to the command table */
1054 	memcpy(int_cmd->command, fis, fis_len*4);
1055 
1056 	rq->timeout = timeout;
1057 
1058 	/* insert request and run queue */
1059 	blk_execute_rq(rq->q, NULL, rq, true);
1060 
1061 	if (int_cmd->status) {
1062 		dev_err(&dd->pdev->dev, "Internal command [%02X] failed %d\n",
1063 				fis->command, int_cmd->status);
1064 		rv = -EIO;
1065 
1066 		if (mtip_check_surprise_removal(dd->pdev) ||
1067 			test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
1068 					&dd->dd_flag)) {
1069 			dev_err(&dd->pdev->dev,
1070 				"Internal command [%02X] wait returned due to SR\n",
1071 				fis->command);
1072 			rv = -ENXIO;
1073 			goto exec_ic_exit;
1074 		}
1075 		mtip_device_reset(dd); /* recover from timeout issue */
1076 		rv = -EAGAIN;
1077 		goto exec_ic_exit;
1078 	}
1079 
1080 	if (readl(port->cmd_issue[MTIP_TAG_INDEX(MTIP_TAG_INTERNAL)])
1081 			& (1 << MTIP_TAG_BIT(MTIP_TAG_INTERNAL))) {
1082 		rv = -ENXIO;
1083 		if (!test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)) {
1084 			mtip_device_reset(dd);
1085 			rv = -EAGAIN;
1086 		}
1087 	}
1088 exec_ic_exit:
1089 	/* Clear the allocated and active bits for the internal command. */
1090 	blk_mq_free_request(rq);
1091 	clear_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags);
1092 	if (rv >= 0 && mtip_pause_ncq(port, fis)) {
1093 		/* NCQ paused */
1094 		return rv;
1095 	}
1096 	wake_up_interruptible(&port->svc_wait);
1097 
1098 	return rv;
1099 }
1100 
1101 /*
1102  * Byte-swap ATA ID strings.
1103  *
1104  * ATA identify data contains strings in byte-swapped 16-bit words.
1105  * They must be swapped (on all architectures) to be usable as C strings.
1106  * This function swaps bytes in-place.
1107  *
1108  * @buf The buffer location of the string
1109  * @len The number of bytes to swap
1110  *
1111  * return value
1112  *	None
1113  */
1114 static inline void ata_swap_string(u16 *buf, unsigned int len)
1115 {
1116 	int i;
1117 	for (i = 0; i < (len/2); i++)
1118 		be16_to_cpus(&buf[i]);
1119 }
1120 
1121 static void mtip_set_timeout(struct driver_data *dd,
1122 					struct host_to_dev_fis *fis,
1123 					unsigned int *timeout, u8 erasemode)
1124 {
1125 	switch (fis->command) {
1126 	case ATA_CMD_DOWNLOAD_MICRO:
1127 		*timeout = 120000; /* 2 minutes */
1128 		break;
1129 	case ATA_CMD_SEC_ERASE_UNIT:
1130 	case 0xFC:
1131 		if (erasemode)
1132 			*timeout = ((*(dd->port->identify + 90) * 2) * 60000);
1133 		else
1134 			*timeout = ((*(dd->port->identify + 89) * 2) * 60000);
1135 		break;
1136 	case ATA_CMD_STANDBYNOW1:
1137 		*timeout = 120000;  /* 2 minutes */
1138 		break;
1139 	case 0xF7:
1140 	case 0xFA:
1141 		*timeout = 60000;  /* 60 seconds */
1142 		break;
1143 	case ATA_CMD_SMART:
1144 		*timeout = 15000;  /* 15 seconds */
1145 		break;
1146 	default:
1147 		*timeout = MTIP_IOCTL_CMD_TIMEOUT_MS;
1148 		break;
1149 	}
1150 }
1151 
1152 /*
1153  * Request the device identity information.
1154  *
1155  * If a user space buffer is not specified, i.e. is NULL, the
1156  * identify information is still read from the drive and placed
1157  * into the identify data buffer (@e port->identify) in the
1158  * port data structure.
1159  * When the identify buffer contains valid identify information @e
1160  * port->identify_valid is non-zero.
1161  *
1162  * @port	 Pointer to the port structure.
1163  * @user_buffer  A user space buffer where the identify data should be
1164  *                    copied.
1165  *
1166  * return value
1167  *	0	Command completed successfully.
1168  *	-EFAULT An error occurred while coping data to the user buffer.
1169  *	-1	Command failed.
1170  */
1171 static int mtip_get_identify(struct mtip_port *port, void __user *user_buffer)
1172 {
1173 	int rv = 0;
1174 	struct host_to_dev_fis fis;
1175 
1176 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag))
1177 		return -EFAULT;
1178 
1179 	/* Build the FIS. */
1180 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1181 	fis.type	= 0x27;
1182 	fis.opts	= 1 << 7;
1183 	fis.command	= ATA_CMD_ID_ATA;
1184 
1185 	/* Set the identify information as invalid. */
1186 	port->identify_valid = 0;
1187 
1188 	/* Clear the identify information. */
1189 	memset(port->identify, 0, sizeof(u16) * ATA_ID_WORDS);
1190 
1191 	/* Execute the command. */
1192 	if (mtip_exec_internal_command(port,
1193 				&fis,
1194 				5,
1195 				port->identify_dma,
1196 				sizeof(u16) * ATA_ID_WORDS,
1197 				0,
1198 				MTIP_INT_CMD_TIMEOUT_MS)
1199 				< 0) {
1200 		rv = -1;
1201 		goto out;
1202 	}
1203 
1204 	/*
1205 	 * Perform any necessary byte-swapping.  Yes, the kernel does in fact
1206 	 * perform field-sensitive swapping on the string fields.
1207 	 * See the kernel use of ata_id_string() for proof of this.
1208 	 */
1209 #ifdef __LITTLE_ENDIAN
1210 	ata_swap_string(port->identify + 27, 40);  /* model string*/
1211 	ata_swap_string(port->identify + 23, 8);   /* firmware string*/
1212 	ata_swap_string(port->identify + 10, 20);  /* serial# string*/
1213 #else
1214 	{
1215 		int i;
1216 		for (i = 0; i < ATA_ID_WORDS; i++)
1217 			port->identify[i] = le16_to_cpu(port->identify[i]);
1218 	}
1219 #endif
1220 
1221 	/* Check security locked state */
1222 	if (port->identify[128] & 0x4)
1223 		set_bit(MTIP_DDF_SEC_LOCK_BIT, &port->dd->dd_flag);
1224 	else
1225 		clear_bit(MTIP_DDF_SEC_LOCK_BIT, &port->dd->dd_flag);
1226 
1227 #ifdef MTIP_TRIM /* Disabling TRIM support temporarily */
1228 	/* Demux ID.DRAT & ID.RZAT to determine trim support */
1229 	if (port->identify[69] & (1 << 14) && port->identify[69] & (1 << 5))
1230 		port->dd->trim_supp = true;
1231 	else
1232 #endif
1233 		port->dd->trim_supp = false;
1234 
1235 	/* Set the identify buffer as valid. */
1236 	port->identify_valid = 1;
1237 
1238 	if (user_buffer) {
1239 		if (copy_to_user(
1240 			user_buffer,
1241 			port->identify,
1242 			ATA_ID_WORDS * sizeof(u16))) {
1243 			rv = -EFAULT;
1244 			goto out;
1245 		}
1246 	}
1247 
1248 out:
1249 	return rv;
1250 }
1251 
1252 /*
1253  * Issue a standby immediate command to the device.
1254  *
1255  * @port Pointer to the port structure.
1256  *
1257  * return value
1258  *	0	Command was executed successfully.
1259  *	-1	An error occurred while executing the command.
1260  */
1261 static int mtip_standby_immediate(struct mtip_port *port)
1262 {
1263 	int rv;
1264 	struct host_to_dev_fis	fis;
1265 	unsigned long start;
1266 	unsigned int timeout;
1267 
1268 	/* Build the FIS. */
1269 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1270 	fis.type	= 0x27;
1271 	fis.opts	= 1 << 7;
1272 	fis.command	= ATA_CMD_STANDBYNOW1;
1273 
1274 	mtip_set_timeout(port->dd, &fis, &timeout, 0);
1275 
1276 	start = jiffies;
1277 	rv = mtip_exec_internal_command(port,
1278 					&fis,
1279 					5,
1280 					0,
1281 					0,
1282 					0,
1283 					timeout);
1284 	dbg_printk(MTIP_DRV_NAME "Time taken to complete standby cmd: %d ms\n",
1285 			jiffies_to_msecs(jiffies - start));
1286 	if (rv)
1287 		dev_warn(&port->dd->pdev->dev,
1288 			"STANDBY IMMEDIATE command failed.\n");
1289 
1290 	return rv;
1291 }
1292 
1293 /*
1294  * Issue a READ LOG EXT command to the device.
1295  *
1296  * @port	pointer to the port structure.
1297  * @page	page number to fetch
1298  * @buffer	pointer to buffer
1299  * @buffer_dma	dma address corresponding to @buffer
1300  * @sectors	page length to fetch, in sectors
1301  *
1302  * return value
1303  *	@rv	return value from mtip_exec_internal_command()
1304  */
1305 static int mtip_read_log_page(struct mtip_port *port, u8 page, u16 *buffer,
1306 				dma_addr_t buffer_dma, unsigned int sectors)
1307 {
1308 	struct host_to_dev_fis fis;
1309 
1310 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1311 	fis.type	= 0x27;
1312 	fis.opts	= 1 << 7;
1313 	fis.command	= ATA_CMD_READ_LOG_EXT;
1314 	fis.sect_count	= sectors & 0xFF;
1315 	fis.sect_cnt_ex	= (sectors >> 8) & 0xFF;
1316 	fis.lba_low	= page;
1317 	fis.lba_mid	= 0;
1318 	fis.device	= ATA_DEVICE_OBS;
1319 
1320 	memset(buffer, 0, sectors * ATA_SECT_SIZE);
1321 
1322 	return mtip_exec_internal_command(port,
1323 					&fis,
1324 					5,
1325 					buffer_dma,
1326 					sectors * ATA_SECT_SIZE,
1327 					0,
1328 					MTIP_INT_CMD_TIMEOUT_MS);
1329 }
1330 
1331 /*
1332  * Issue a SMART READ DATA command to the device.
1333  *
1334  * @port	pointer to the port structure.
1335  * @buffer	pointer to buffer
1336  * @buffer_dma	dma address corresponding to @buffer
1337  *
1338  * return value
1339  *	@rv	return value from mtip_exec_internal_command()
1340  */
1341 static int mtip_get_smart_data(struct mtip_port *port, u8 *buffer,
1342 					dma_addr_t buffer_dma)
1343 {
1344 	struct host_to_dev_fis fis;
1345 
1346 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1347 	fis.type	= 0x27;
1348 	fis.opts	= 1 << 7;
1349 	fis.command	= ATA_CMD_SMART;
1350 	fis.features	= 0xD0;
1351 	fis.sect_count	= 1;
1352 	fis.lba_mid	= 0x4F;
1353 	fis.lba_hi	= 0xC2;
1354 	fis.device	= ATA_DEVICE_OBS;
1355 
1356 	return mtip_exec_internal_command(port,
1357 					&fis,
1358 					5,
1359 					buffer_dma,
1360 					ATA_SECT_SIZE,
1361 					0,
1362 					15000);
1363 }
1364 
1365 /*
1366  * Get the value of a smart attribute
1367  *
1368  * @port	pointer to the port structure
1369  * @id		attribute number
1370  * @attrib	pointer to return attrib information corresponding to @id
1371  *
1372  * return value
1373  *	-EINVAL	NULL buffer passed or unsupported attribute @id.
1374  *	-EPERM	Identify data not valid, SMART not supported or not enabled
1375  */
1376 static int mtip_get_smart_attr(struct mtip_port *port, unsigned int id,
1377 						struct smart_attr *attrib)
1378 {
1379 	int rv, i;
1380 	struct smart_attr *pattr;
1381 
1382 	if (!attrib)
1383 		return -EINVAL;
1384 
1385 	if (!port->identify_valid) {
1386 		dev_warn(&port->dd->pdev->dev, "IDENTIFY DATA not valid\n");
1387 		return -EPERM;
1388 	}
1389 	if (!(port->identify[82] & 0x1)) {
1390 		dev_warn(&port->dd->pdev->dev, "SMART not supported\n");
1391 		return -EPERM;
1392 	}
1393 	if (!(port->identify[85] & 0x1)) {
1394 		dev_warn(&port->dd->pdev->dev, "SMART not enabled\n");
1395 		return -EPERM;
1396 	}
1397 
1398 	memset(port->smart_buf, 0, ATA_SECT_SIZE);
1399 	rv = mtip_get_smart_data(port, port->smart_buf, port->smart_buf_dma);
1400 	if (rv) {
1401 		dev_warn(&port->dd->pdev->dev, "Failed to ge SMART data\n");
1402 		return rv;
1403 	}
1404 
1405 	pattr = (struct smart_attr *)(port->smart_buf + 2);
1406 	for (i = 0; i < 29; i++, pattr++)
1407 		if (pattr->attr_id == id) {
1408 			memcpy(attrib, pattr, sizeof(struct smart_attr));
1409 			break;
1410 		}
1411 
1412 	if (i == 29) {
1413 		dev_warn(&port->dd->pdev->dev,
1414 			"Query for invalid SMART attribute ID\n");
1415 		rv = -EINVAL;
1416 	}
1417 
1418 	return rv;
1419 }
1420 
1421 /*
1422  * Trim unused sectors
1423  *
1424  * @dd		pointer to driver_data structure
1425  * @lba		starting lba
1426  * @len		# of 512b sectors to trim
1427  *
1428  * return value
1429  *      -ENOMEM		Out of dma memory
1430  *      -EINVAL		Invalid parameters passed in, trim not supported
1431  *      -EIO		Error submitting trim request to hw
1432  */
1433 static int mtip_send_trim(struct driver_data *dd, unsigned int lba,
1434 				unsigned int len)
1435 {
1436 	int i, rv = 0;
1437 	u64 tlba, tlen, sect_left;
1438 	struct mtip_trim_entry *buf;
1439 	dma_addr_t dma_addr;
1440 	struct host_to_dev_fis fis;
1441 
1442 	if (!len || dd->trim_supp == false)
1443 		return -EINVAL;
1444 
1445 	/* Trim request too big */
1446 	WARN_ON(len > (MTIP_MAX_TRIM_ENTRY_LEN * MTIP_MAX_TRIM_ENTRIES));
1447 
1448 	/* Trim request not aligned on 4k boundary */
1449 	WARN_ON(len % 8 != 0);
1450 
1451 	/* Warn if vu_trim structure is too big */
1452 	WARN_ON(sizeof(struct mtip_trim) > ATA_SECT_SIZE);
1453 
1454 	/* Allocate a DMA buffer for the trim structure */
1455 	buf = dmam_alloc_coherent(&dd->pdev->dev, ATA_SECT_SIZE, &dma_addr,
1456 								GFP_KERNEL);
1457 	if (!buf)
1458 		return -ENOMEM;
1459 	memset(buf, 0, ATA_SECT_SIZE);
1460 
1461 	for (i = 0, sect_left = len, tlba = lba;
1462 			i < MTIP_MAX_TRIM_ENTRIES && sect_left;
1463 			i++) {
1464 		tlen = (sect_left >= MTIP_MAX_TRIM_ENTRY_LEN ?
1465 					MTIP_MAX_TRIM_ENTRY_LEN :
1466 					sect_left);
1467 		buf[i].lba = __force_bit2int cpu_to_le32(tlba);
1468 		buf[i].range = __force_bit2int cpu_to_le16(tlen);
1469 		tlba += tlen;
1470 		sect_left -= tlen;
1471 	}
1472 	WARN_ON(sect_left != 0);
1473 
1474 	/* Build the fis */
1475 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1476 	fis.type       = 0x27;
1477 	fis.opts       = 1 << 7;
1478 	fis.command    = 0xfb;
1479 	fis.features   = 0x60;
1480 	fis.sect_count = 1;
1481 	fis.device     = ATA_DEVICE_OBS;
1482 
1483 	if (mtip_exec_internal_command(dd->port,
1484 					&fis,
1485 					5,
1486 					dma_addr,
1487 					ATA_SECT_SIZE,
1488 					0,
1489 					MTIP_TRIM_TIMEOUT_MS) < 0)
1490 		rv = -EIO;
1491 
1492 	dmam_free_coherent(&dd->pdev->dev, ATA_SECT_SIZE, buf, dma_addr);
1493 	return rv;
1494 }
1495 
1496 /*
1497  * Get the drive capacity.
1498  *
1499  * @dd      Pointer to the device data structure.
1500  * @sectors Pointer to the variable that will receive the sector count.
1501  *
1502  * return value
1503  *	1 Capacity was returned successfully.
1504  *	0 The identify information is invalid.
1505  */
1506 static bool mtip_hw_get_capacity(struct driver_data *dd, sector_t *sectors)
1507 {
1508 	struct mtip_port *port = dd->port;
1509 	u64 total, raw0, raw1, raw2, raw3;
1510 	raw0 = port->identify[100];
1511 	raw1 = port->identify[101];
1512 	raw2 = port->identify[102];
1513 	raw3 = port->identify[103];
1514 	total = raw0 | raw1<<16 | raw2<<32 | raw3<<48;
1515 	*sectors = total;
1516 	return (bool) !!port->identify_valid;
1517 }
1518 
1519 /*
1520  * Display the identify command data.
1521  *
1522  * @port Pointer to the port data structure.
1523  *
1524  * return value
1525  *	None
1526  */
1527 static void mtip_dump_identify(struct mtip_port *port)
1528 {
1529 	sector_t sectors;
1530 	unsigned short revid;
1531 	char cbuf[42];
1532 
1533 	if (!port->identify_valid)
1534 		return;
1535 
1536 	strlcpy(cbuf, (char *)(port->identify+10), 21);
1537 	dev_info(&port->dd->pdev->dev,
1538 		"Serial No.: %s\n", cbuf);
1539 
1540 	strlcpy(cbuf, (char *)(port->identify+23), 9);
1541 	dev_info(&port->dd->pdev->dev,
1542 		"Firmware Ver.: %s\n", cbuf);
1543 
1544 	strlcpy(cbuf, (char *)(port->identify+27), 41);
1545 	dev_info(&port->dd->pdev->dev, "Model: %s\n", cbuf);
1546 
1547 	dev_info(&port->dd->pdev->dev, "Security: %04x %s\n",
1548 		port->identify[128],
1549 		port->identify[128] & 0x4 ? "(LOCKED)" : "");
1550 
1551 	if (mtip_hw_get_capacity(port->dd, &sectors))
1552 		dev_info(&port->dd->pdev->dev,
1553 			"Capacity: %llu sectors (%llu MB)\n",
1554 			 (u64)sectors,
1555 			 ((u64)sectors) * ATA_SECT_SIZE >> 20);
1556 
1557 	pci_read_config_word(port->dd->pdev, PCI_REVISION_ID, &revid);
1558 	switch (revid & 0xFF) {
1559 	case 0x1:
1560 		strlcpy(cbuf, "A0", 3);
1561 		break;
1562 	case 0x3:
1563 		strlcpy(cbuf, "A2", 3);
1564 		break;
1565 	default:
1566 		strlcpy(cbuf, "?", 2);
1567 		break;
1568 	}
1569 	dev_info(&port->dd->pdev->dev,
1570 		"Card Type: %s\n", cbuf);
1571 }
1572 
1573 /*
1574  * Map the commands scatter list into the command table.
1575  *
1576  * @command Pointer to the command.
1577  * @nents Number of scatter list entries.
1578  *
1579  * return value
1580  *	None
1581  */
1582 static inline void fill_command_sg(struct driver_data *dd,
1583 				struct mtip_cmd *command,
1584 				int nents)
1585 {
1586 	int n;
1587 	unsigned int dma_len;
1588 	struct mtip_cmd_sg *command_sg;
1589 	struct scatterlist *sg = command->sg;
1590 
1591 	command_sg = command->command + AHCI_CMD_TBL_HDR_SZ;
1592 
1593 	for (n = 0; n < nents; n++) {
1594 		dma_len = sg_dma_len(sg);
1595 		if (dma_len > 0x400000)
1596 			dev_err(&dd->pdev->dev,
1597 				"DMA segment length truncated\n");
1598 		command_sg->info = __force_bit2int
1599 			cpu_to_le32((dma_len-1) & 0x3FFFFF);
1600 		command_sg->dba	= __force_bit2int
1601 			cpu_to_le32(sg_dma_address(sg));
1602 		command_sg->dba_upper = __force_bit2int
1603 			cpu_to_le32((sg_dma_address(sg) >> 16) >> 16);
1604 		command_sg++;
1605 		sg++;
1606 	}
1607 }
1608 
1609 /*
1610  * @brief Execute a drive command.
1611  *
1612  * return value 0 The command completed successfully.
1613  * return value -1 An error occurred while executing the command.
1614  */
1615 static int exec_drive_task(struct mtip_port *port, u8 *command)
1616 {
1617 	struct host_to_dev_fis	fis;
1618 	struct host_to_dev_fis *reply = (port->rxfis + RX_FIS_D2H_REG);
1619 	unsigned int to;
1620 
1621 	/* Build the FIS. */
1622 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1623 	fis.type	= 0x27;
1624 	fis.opts	= 1 << 7;
1625 	fis.command	= command[0];
1626 	fis.features	= command[1];
1627 	fis.sect_count	= command[2];
1628 	fis.sector	= command[3];
1629 	fis.cyl_low	= command[4];
1630 	fis.cyl_hi	= command[5];
1631 	fis.device	= command[6] & ~0x10; /* Clear the dev bit*/
1632 
1633 	mtip_set_timeout(port->dd, &fis, &to, 0);
1634 
1635 	dbg_printk(MTIP_DRV_NAME " %s: User Command: cmd %x, feat %x, nsect %x, sect %x, lcyl %x, hcyl %x, sel %x\n",
1636 		__func__,
1637 		command[0],
1638 		command[1],
1639 		command[2],
1640 		command[3],
1641 		command[4],
1642 		command[5],
1643 		command[6]);
1644 
1645 	/* Execute the command. */
1646 	if (mtip_exec_internal_command(port,
1647 				 &fis,
1648 				 5,
1649 				 0,
1650 				 0,
1651 				 0,
1652 				 to) < 0) {
1653 		return -1;
1654 	}
1655 
1656 	command[0] = reply->command; /* Status*/
1657 	command[1] = reply->features; /* Error*/
1658 	command[4] = reply->cyl_low;
1659 	command[5] = reply->cyl_hi;
1660 
1661 	dbg_printk(MTIP_DRV_NAME " %s: Completion Status: stat %x, err %x , cyl_lo %x cyl_hi %x\n",
1662 		__func__,
1663 		command[0],
1664 		command[1],
1665 		command[4],
1666 		command[5]);
1667 
1668 	return 0;
1669 }
1670 
1671 /*
1672  * @brief Execute a drive command.
1673  *
1674  * @param port Pointer to the port data structure.
1675  * @param command Pointer to the user specified command parameters.
1676  * @param user_buffer Pointer to the user space buffer where read sector
1677  *                   data should be copied.
1678  *
1679  * return value 0 The command completed successfully.
1680  * return value -EFAULT An error occurred while copying the completion
1681  *                 data to the user space buffer.
1682  * return value -1 An error occurred while executing the command.
1683  */
1684 static int exec_drive_command(struct mtip_port *port, u8 *command,
1685 				void __user *user_buffer)
1686 {
1687 	struct host_to_dev_fis	fis;
1688 	struct host_to_dev_fis *reply;
1689 	u8 *buf = NULL;
1690 	dma_addr_t dma_addr = 0;
1691 	int rv = 0, xfer_sz = command[3];
1692 	unsigned int to;
1693 
1694 	if (xfer_sz) {
1695 		if (!user_buffer)
1696 			return -EFAULT;
1697 
1698 		buf = dmam_alloc_coherent(&port->dd->pdev->dev,
1699 				ATA_SECT_SIZE * xfer_sz,
1700 				&dma_addr,
1701 				GFP_KERNEL);
1702 		if (!buf) {
1703 			dev_err(&port->dd->pdev->dev,
1704 				"Memory allocation failed (%d bytes)\n",
1705 				ATA_SECT_SIZE * xfer_sz);
1706 			return -ENOMEM;
1707 		}
1708 		memset(buf, 0, ATA_SECT_SIZE * xfer_sz);
1709 	}
1710 
1711 	/* Build the FIS. */
1712 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1713 	fis.type	= 0x27;
1714 	fis.opts	= 1 << 7;
1715 	fis.command	= command[0];
1716 	fis.features	= command[2];
1717 	fis.sect_count	= command[3];
1718 	if (fis.command == ATA_CMD_SMART) {
1719 		fis.sector	= command[1];
1720 		fis.cyl_low	= 0x4F;
1721 		fis.cyl_hi	= 0xC2;
1722 	}
1723 
1724 	mtip_set_timeout(port->dd, &fis, &to, 0);
1725 
1726 	if (xfer_sz)
1727 		reply = (port->rxfis + RX_FIS_PIO_SETUP);
1728 	else
1729 		reply = (port->rxfis + RX_FIS_D2H_REG);
1730 
1731 	dbg_printk(MTIP_DRV_NAME
1732 		" %s: User Command: cmd %x, sect %x, "
1733 		"feat %x, sectcnt %x\n",
1734 		__func__,
1735 		command[0],
1736 		command[1],
1737 		command[2],
1738 		command[3]);
1739 
1740 	/* Execute the command. */
1741 	if (mtip_exec_internal_command(port,
1742 				&fis,
1743 				 5,
1744 				 (xfer_sz ? dma_addr : 0),
1745 				 (xfer_sz ? ATA_SECT_SIZE * xfer_sz : 0),
1746 				 0,
1747 				 to)
1748 				 < 0) {
1749 		rv = -EFAULT;
1750 		goto exit_drive_command;
1751 	}
1752 
1753 	/* Collect the completion status. */
1754 	command[0] = reply->command; /* Status*/
1755 	command[1] = reply->features; /* Error*/
1756 	command[2] = reply->sect_count;
1757 
1758 	dbg_printk(MTIP_DRV_NAME
1759 		" %s: Completion Status: stat %x, "
1760 		"err %x, nsect %x\n",
1761 		__func__,
1762 		command[0],
1763 		command[1],
1764 		command[2]);
1765 
1766 	if (xfer_sz) {
1767 		if (copy_to_user(user_buffer,
1768 				 buf,
1769 				 ATA_SECT_SIZE * command[3])) {
1770 			rv = -EFAULT;
1771 			goto exit_drive_command;
1772 		}
1773 	}
1774 exit_drive_command:
1775 	if (buf)
1776 		dmam_free_coherent(&port->dd->pdev->dev,
1777 				ATA_SECT_SIZE * xfer_sz, buf, dma_addr);
1778 	return rv;
1779 }
1780 
1781 /*
1782  *  Indicates whether a command has a single sector payload.
1783  *
1784  *  @command passed to the device to perform the certain event.
1785  *  @features passed to the device to perform the certain event.
1786  *
1787  *  return value
1788  *	1	command is one that always has a single sector payload,
1789  *		regardless of the value in the Sector Count field.
1790  *      0       otherwise
1791  *
1792  */
1793 static unsigned int implicit_sector(unsigned char command,
1794 				    unsigned char features)
1795 {
1796 	unsigned int rv = 0;
1797 
1798 	/* list of commands that have an implicit sector count of 1 */
1799 	switch (command) {
1800 	case ATA_CMD_SEC_SET_PASS:
1801 	case ATA_CMD_SEC_UNLOCK:
1802 	case ATA_CMD_SEC_ERASE_PREP:
1803 	case ATA_CMD_SEC_ERASE_UNIT:
1804 	case ATA_CMD_SEC_FREEZE_LOCK:
1805 	case ATA_CMD_SEC_DISABLE_PASS:
1806 	case ATA_CMD_PMP_READ:
1807 	case ATA_CMD_PMP_WRITE:
1808 		rv = 1;
1809 		break;
1810 	case ATA_CMD_SET_MAX:
1811 		if (features == ATA_SET_MAX_UNLOCK)
1812 			rv = 1;
1813 		break;
1814 	case ATA_CMD_SMART:
1815 		if ((features == ATA_SMART_READ_VALUES) ||
1816 				(features == ATA_SMART_READ_THRESHOLDS))
1817 			rv = 1;
1818 		break;
1819 	case ATA_CMD_CONF_OVERLAY:
1820 		if ((features == ATA_DCO_IDENTIFY) ||
1821 				(features == ATA_DCO_SET))
1822 			rv = 1;
1823 		break;
1824 	}
1825 	return rv;
1826 }
1827 
1828 /*
1829  * Executes a taskfile
1830  * See ide_taskfile_ioctl() for derivation
1831  */
1832 static int exec_drive_taskfile(struct driver_data *dd,
1833 			       void __user *buf,
1834 			       ide_task_request_t *req_task,
1835 			       int outtotal)
1836 {
1837 	struct host_to_dev_fis	fis;
1838 	struct host_to_dev_fis *reply;
1839 	u8 *outbuf = NULL;
1840 	u8 *inbuf = NULL;
1841 	dma_addr_t outbuf_dma = 0;
1842 	dma_addr_t inbuf_dma = 0;
1843 	dma_addr_t dma_buffer = 0;
1844 	int err = 0;
1845 	unsigned int taskin = 0;
1846 	unsigned int taskout = 0;
1847 	u8 nsect = 0;
1848 	unsigned int timeout;
1849 	unsigned int force_single_sector;
1850 	unsigned int transfer_size;
1851 	unsigned long task_file_data;
1852 	int intotal = outtotal + req_task->out_size;
1853 	int erasemode = 0;
1854 
1855 	taskout = req_task->out_size;
1856 	taskin = req_task->in_size;
1857 	/* 130560 = 512 * 0xFF*/
1858 	if (taskin > 130560 || taskout > 130560)
1859 		return -EINVAL;
1860 
1861 	if (taskout) {
1862 		outbuf = memdup_user(buf + outtotal, taskout);
1863 		if (IS_ERR(outbuf))
1864 			return PTR_ERR(outbuf);
1865 
1866 		outbuf_dma = pci_map_single(dd->pdev,
1867 					 outbuf,
1868 					 taskout,
1869 					 DMA_TO_DEVICE);
1870 		if (pci_dma_mapping_error(dd->pdev, outbuf_dma)) {
1871 			err = -ENOMEM;
1872 			goto abort;
1873 		}
1874 		dma_buffer = outbuf_dma;
1875 	}
1876 
1877 	if (taskin) {
1878 		inbuf = memdup_user(buf + intotal, taskin);
1879 		if (IS_ERR(inbuf)) {
1880 			err = PTR_ERR(inbuf);
1881 			inbuf = NULL;
1882 			goto abort;
1883 		}
1884 		inbuf_dma = pci_map_single(dd->pdev,
1885 					 inbuf,
1886 					 taskin, DMA_FROM_DEVICE);
1887 		if (pci_dma_mapping_error(dd->pdev, inbuf_dma)) {
1888 			err = -ENOMEM;
1889 			goto abort;
1890 		}
1891 		dma_buffer = inbuf_dma;
1892 	}
1893 
1894 	/* only supports PIO and non-data commands from this ioctl. */
1895 	switch (req_task->data_phase) {
1896 	case TASKFILE_OUT:
1897 		nsect = taskout / ATA_SECT_SIZE;
1898 		reply = (dd->port->rxfis + RX_FIS_PIO_SETUP);
1899 		break;
1900 	case TASKFILE_IN:
1901 		reply = (dd->port->rxfis + RX_FIS_PIO_SETUP);
1902 		break;
1903 	case TASKFILE_NO_DATA:
1904 		reply = (dd->port->rxfis + RX_FIS_D2H_REG);
1905 		break;
1906 	default:
1907 		err = -EINVAL;
1908 		goto abort;
1909 	}
1910 
1911 	/* Build the FIS. */
1912 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1913 
1914 	fis.type	= 0x27;
1915 	fis.opts	= 1 << 7;
1916 	fis.command	= req_task->io_ports[7];
1917 	fis.features	= req_task->io_ports[1];
1918 	fis.sect_count	= req_task->io_ports[2];
1919 	fis.lba_low	= req_task->io_ports[3];
1920 	fis.lba_mid	= req_task->io_ports[4];
1921 	fis.lba_hi	= req_task->io_ports[5];
1922 	 /* Clear the dev bit*/
1923 	fis.device	= req_task->io_ports[6] & ~0x10;
1924 
1925 	if ((req_task->in_flags.all == 0) && (req_task->out_flags.all & 1)) {
1926 		req_task->in_flags.all	=
1927 			IDE_TASKFILE_STD_IN_FLAGS |
1928 			(IDE_HOB_STD_IN_FLAGS << 8);
1929 		fis.lba_low_ex		= req_task->hob_ports[3];
1930 		fis.lba_mid_ex		= req_task->hob_ports[4];
1931 		fis.lba_hi_ex		= req_task->hob_ports[5];
1932 		fis.features_ex		= req_task->hob_ports[1];
1933 		fis.sect_cnt_ex		= req_task->hob_ports[2];
1934 
1935 	} else {
1936 		req_task->in_flags.all = IDE_TASKFILE_STD_IN_FLAGS;
1937 	}
1938 
1939 	force_single_sector = implicit_sector(fis.command, fis.features);
1940 
1941 	if ((taskin || taskout) && (!fis.sect_count)) {
1942 		if (nsect)
1943 			fis.sect_count = nsect;
1944 		else {
1945 			if (!force_single_sector) {
1946 				dev_warn(&dd->pdev->dev,
1947 					"data movement but "
1948 					"sect_count is 0\n");
1949 					err = -EINVAL;
1950 					goto abort;
1951 			}
1952 		}
1953 	}
1954 
1955 	dbg_printk(MTIP_DRV_NAME
1956 		" %s: cmd %x, feat %x, nsect %x,"
1957 		" sect/lbal %x, lcyl/lbam %x, hcyl/lbah %x,"
1958 		" head/dev %x\n",
1959 		__func__,
1960 		fis.command,
1961 		fis.features,
1962 		fis.sect_count,
1963 		fis.lba_low,
1964 		fis.lba_mid,
1965 		fis.lba_hi,
1966 		fis.device);
1967 
1968 	/* check for erase mode support during secure erase.*/
1969 	if ((fis.command == ATA_CMD_SEC_ERASE_UNIT) && outbuf &&
1970 					(outbuf[0] & MTIP_SEC_ERASE_MODE)) {
1971 		erasemode = 1;
1972 	}
1973 
1974 	mtip_set_timeout(dd, &fis, &timeout, erasemode);
1975 
1976 	/* Determine the correct transfer size.*/
1977 	if (force_single_sector)
1978 		transfer_size = ATA_SECT_SIZE;
1979 	else
1980 		transfer_size = ATA_SECT_SIZE * fis.sect_count;
1981 
1982 	/* Execute the command.*/
1983 	if (mtip_exec_internal_command(dd->port,
1984 				 &fis,
1985 				 5,
1986 				 dma_buffer,
1987 				 transfer_size,
1988 				 0,
1989 				 timeout) < 0) {
1990 		err = -EIO;
1991 		goto abort;
1992 	}
1993 
1994 	task_file_data = readl(dd->port->mmio+PORT_TFDATA);
1995 
1996 	if ((req_task->data_phase == TASKFILE_IN) && !(task_file_data & 1)) {
1997 		reply = dd->port->rxfis + RX_FIS_PIO_SETUP;
1998 		req_task->io_ports[7] = reply->control;
1999 	} else {
2000 		reply = dd->port->rxfis + RX_FIS_D2H_REG;
2001 		req_task->io_ports[7] = reply->command;
2002 	}
2003 
2004 	/* reclaim the DMA buffers.*/
2005 	if (inbuf_dma)
2006 		pci_unmap_single(dd->pdev, inbuf_dma,
2007 			taskin, DMA_FROM_DEVICE);
2008 	if (outbuf_dma)
2009 		pci_unmap_single(dd->pdev, outbuf_dma,
2010 			taskout, DMA_TO_DEVICE);
2011 	inbuf_dma  = 0;
2012 	outbuf_dma = 0;
2013 
2014 	/* return the ATA registers to the caller.*/
2015 	req_task->io_ports[1] = reply->features;
2016 	req_task->io_ports[2] = reply->sect_count;
2017 	req_task->io_ports[3] = reply->lba_low;
2018 	req_task->io_ports[4] = reply->lba_mid;
2019 	req_task->io_ports[5] = reply->lba_hi;
2020 	req_task->io_ports[6] = reply->device;
2021 
2022 	if (req_task->out_flags.all & 1)  {
2023 
2024 		req_task->hob_ports[3] = reply->lba_low_ex;
2025 		req_task->hob_ports[4] = reply->lba_mid_ex;
2026 		req_task->hob_ports[5] = reply->lba_hi_ex;
2027 		req_task->hob_ports[1] = reply->features_ex;
2028 		req_task->hob_ports[2] = reply->sect_cnt_ex;
2029 	}
2030 	dbg_printk(MTIP_DRV_NAME
2031 		" %s: Completion: stat %x,"
2032 		"err %x, sect_cnt %x, lbalo %x,"
2033 		"lbamid %x, lbahi %x, dev %x\n",
2034 		__func__,
2035 		req_task->io_ports[7],
2036 		req_task->io_ports[1],
2037 		req_task->io_ports[2],
2038 		req_task->io_ports[3],
2039 		req_task->io_ports[4],
2040 		req_task->io_ports[5],
2041 		req_task->io_ports[6]);
2042 
2043 	if (taskout) {
2044 		if (copy_to_user(buf + outtotal, outbuf, taskout)) {
2045 			err = -EFAULT;
2046 			goto abort;
2047 		}
2048 	}
2049 	if (taskin) {
2050 		if (copy_to_user(buf + intotal, inbuf, taskin)) {
2051 			err = -EFAULT;
2052 			goto abort;
2053 		}
2054 	}
2055 abort:
2056 	if (inbuf_dma)
2057 		pci_unmap_single(dd->pdev, inbuf_dma,
2058 					taskin, DMA_FROM_DEVICE);
2059 	if (outbuf_dma)
2060 		pci_unmap_single(dd->pdev, outbuf_dma,
2061 					taskout, DMA_TO_DEVICE);
2062 	kfree(outbuf);
2063 	kfree(inbuf);
2064 
2065 	return err;
2066 }
2067 
2068 /*
2069  * Handle IOCTL calls from the Block Layer.
2070  *
2071  * This function is called by the Block Layer when it receives an IOCTL
2072  * command that it does not understand. If the IOCTL command is not supported
2073  * this function returns -ENOTTY.
2074  *
2075  * @dd  Pointer to the driver data structure.
2076  * @cmd IOCTL command passed from the Block Layer.
2077  * @arg IOCTL argument passed from the Block Layer.
2078  *
2079  * return value
2080  *	0	The IOCTL completed successfully.
2081  *	-ENOTTY The specified command is not supported.
2082  *	-EFAULT An error occurred copying data to a user space buffer.
2083  *	-EIO	An error occurred while executing the command.
2084  */
2085 static int mtip_hw_ioctl(struct driver_data *dd, unsigned int cmd,
2086 			 unsigned long arg)
2087 {
2088 	switch (cmd) {
2089 	case HDIO_GET_IDENTITY:
2090 	{
2091 		if (copy_to_user((void __user *)arg, dd->port->identify,
2092 						sizeof(u16) * ATA_ID_WORDS))
2093 			return -EFAULT;
2094 		break;
2095 	}
2096 	case HDIO_DRIVE_CMD:
2097 	{
2098 		u8 drive_command[4];
2099 
2100 		/* Copy the user command info to our buffer. */
2101 		if (copy_from_user(drive_command,
2102 					 (void __user *) arg,
2103 					 sizeof(drive_command)))
2104 			return -EFAULT;
2105 
2106 		/* Execute the drive command. */
2107 		if (exec_drive_command(dd->port,
2108 					 drive_command,
2109 					 (void __user *) (arg+4)))
2110 			return -EIO;
2111 
2112 		/* Copy the status back to the users buffer. */
2113 		if (copy_to_user((void __user *) arg,
2114 					 drive_command,
2115 					 sizeof(drive_command)))
2116 			return -EFAULT;
2117 
2118 		break;
2119 	}
2120 	case HDIO_DRIVE_TASK:
2121 	{
2122 		u8 drive_command[7];
2123 
2124 		/* Copy the user command info to our buffer. */
2125 		if (copy_from_user(drive_command,
2126 					 (void __user *) arg,
2127 					 sizeof(drive_command)))
2128 			return -EFAULT;
2129 
2130 		/* Execute the drive command. */
2131 		if (exec_drive_task(dd->port, drive_command))
2132 			return -EIO;
2133 
2134 		/* Copy the status back to the users buffer. */
2135 		if (copy_to_user((void __user *) arg,
2136 					 drive_command,
2137 					 sizeof(drive_command)))
2138 			return -EFAULT;
2139 
2140 		break;
2141 	}
2142 	case HDIO_DRIVE_TASKFILE: {
2143 		ide_task_request_t req_task;
2144 		int ret, outtotal;
2145 
2146 		if (copy_from_user(&req_task, (void __user *) arg,
2147 					sizeof(req_task)))
2148 			return -EFAULT;
2149 
2150 		outtotal = sizeof(req_task);
2151 
2152 		ret = exec_drive_taskfile(dd, (void __user *) arg,
2153 						&req_task, outtotal);
2154 
2155 		if (copy_to_user((void __user *) arg, &req_task,
2156 							sizeof(req_task)))
2157 			return -EFAULT;
2158 
2159 		return ret;
2160 	}
2161 
2162 	default:
2163 		return -EINVAL;
2164 	}
2165 	return 0;
2166 }
2167 
2168 /*
2169  * Submit an IO to the hw
2170  *
2171  * This function is called by the block layer to issue an io
2172  * to the device. Upon completion, the callback function will
2173  * be called with the data parameter passed as the callback data.
2174  *
2175  * @dd       Pointer to the driver data structure.
2176  * @start    First sector to read.
2177  * @nsect    Number of sectors to read.
2178  * @nents    Number of entries in scatter list for the read command.
2179  * @tag      The tag of this read command.
2180  * @callback Pointer to the function that should be called
2181  *	     when the read completes.
2182  * @data     Callback data passed to the callback function
2183  *	     when the read completes.
2184  * @dir      Direction (read or write)
2185  *
2186  * return value
2187  *	None
2188  */
2189 static void mtip_hw_submit_io(struct driver_data *dd, struct request *rq,
2190 			      struct mtip_cmd *command, int nents,
2191 			      struct blk_mq_hw_ctx *hctx)
2192 {
2193 	struct host_to_dev_fis	*fis;
2194 	struct mtip_port *port = dd->port;
2195 	int dma_dir = rq_data_dir(rq) == READ ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2196 	u64 start = blk_rq_pos(rq);
2197 	unsigned int nsect = blk_rq_sectors(rq);
2198 
2199 	/* Map the scatter list for DMA access */
2200 	nents = dma_map_sg(&dd->pdev->dev, command->sg, nents, dma_dir);
2201 
2202 	prefetch(&port->flags);
2203 
2204 	command->scatter_ents = nents;
2205 
2206 	/*
2207 	 * The number of retries for this command before it is
2208 	 * reported as a failure to the upper layers.
2209 	 */
2210 	command->retries = MTIP_MAX_RETRIES;
2211 
2212 	/* Fill out fis */
2213 	fis = command->command;
2214 	fis->type        = 0x27;
2215 	fis->opts        = 1 << 7;
2216 	if (dma_dir == DMA_FROM_DEVICE)
2217 		fis->command = ATA_CMD_FPDMA_READ;
2218 	else
2219 		fis->command = ATA_CMD_FPDMA_WRITE;
2220 	fis->lba_low     = start & 0xFF;
2221 	fis->lba_mid     = (start >> 8) & 0xFF;
2222 	fis->lba_hi      = (start >> 16) & 0xFF;
2223 	fis->lba_low_ex  = (start >> 24) & 0xFF;
2224 	fis->lba_mid_ex  = (start >> 32) & 0xFF;
2225 	fis->lba_hi_ex   = (start >> 40) & 0xFF;
2226 	fis->device	 = 1 << 6;
2227 	fis->features    = nsect & 0xFF;
2228 	fis->features_ex = (nsect >> 8) & 0xFF;
2229 	fis->sect_count  = ((rq->tag << 3) | (rq->tag >> 5));
2230 	fis->sect_cnt_ex = 0;
2231 	fis->control     = 0;
2232 	fis->res2        = 0;
2233 	fis->res3        = 0;
2234 	fill_command_sg(dd, command, nents);
2235 
2236 	if (unlikely(command->unaligned))
2237 		fis->device |= 1 << 7;
2238 
2239 	/* Populate the command header */
2240 	command->command_header->opts =
2241 			__force_bit2int cpu_to_le32(
2242 				(nents << 16) | 5 | AHCI_CMD_PREFETCH);
2243 	command->command_header->byte_count = 0;
2244 
2245 	command->direction = dma_dir;
2246 
2247 	/*
2248 	 * To prevent this command from being issued
2249 	 * if an internal command is in progress or error handling is active.
2250 	 */
2251 	if (unlikely(port->flags & MTIP_PF_PAUSE_IO)) {
2252 		set_bit(rq->tag, port->cmds_to_issue);
2253 		set_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags);
2254 		return;
2255 	}
2256 
2257 	/* Issue the command to the hardware */
2258 	mtip_issue_ncq_command(port, rq->tag);
2259 }
2260 
2261 /*
2262  * Sysfs status dump.
2263  *
2264  * @dev  Pointer to the device structure, passed by the kernrel.
2265  * @attr Pointer to the device_attribute structure passed by the kernel.
2266  * @buf  Pointer to the char buffer that will receive the stats info.
2267  *
2268  * return value
2269  *	The size, in bytes, of the data copied into buf.
2270  */
2271 static ssize_t mtip_hw_show_status(struct device *dev,
2272 				struct device_attribute *attr,
2273 				char *buf)
2274 {
2275 	struct driver_data *dd = dev_to_disk(dev)->private_data;
2276 	int size = 0;
2277 
2278 	if (test_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag))
2279 		size += sprintf(buf, "%s", "thermal_shutdown\n");
2280 	else if (test_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag))
2281 		size += sprintf(buf, "%s", "write_protect\n");
2282 	else
2283 		size += sprintf(buf, "%s", "online\n");
2284 
2285 	return size;
2286 }
2287 
2288 static DEVICE_ATTR(status, 0444, mtip_hw_show_status, NULL);
2289 
2290 /* debugsfs entries */
2291 
2292 static ssize_t show_device_status(struct device_driver *drv, char *buf)
2293 {
2294 	int size = 0;
2295 	struct driver_data *dd, *tmp;
2296 	unsigned long flags;
2297 	char id_buf[42];
2298 	u16 status = 0;
2299 
2300 	spin_lock_irqsave(&dev_lock, flags);
2301 	size += sprintf(&buf[size], "Devices Present:\n");
2302 	list_for_each_entry_safe(dd, tmp, &online_list, online_list) {
2303 		if (dd->pdev) {
2304 			if (dd->port &&
2305 			    dd->port->identify &&
2306 			    dd->port->identify_valid) {
2307 				strlcpy(id_buf,
2308 					(char *) (dd->port->identify + 10), 21);
2309 				status = *(dd->port->identify + 141);
2310 			} else {
2311 				memset(id_buf, 0, 42);
2312 				status = 0;
2313 			}
2314 
2315 			if (dd->port &&
2316 			    test_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags)) {
2317 				size += sprintf(&buf[size],
2318 					" device %s %s (ftl rebuild %d %%)\n",
2319 					dev_name(&dd->pdev->dev),
2320 					id_buf,
2321 					status);
2322 			} else {
2323 				size += sprintf(&buf[size],
2324 					" device %s %s\n",
2325 					dev_name(&dd->pdev->dev),
2326 					id_buf);
2327 			}
2328 		}
2329 	}
2330 
2331 	size += sprintf(&buf[size], "Devices Being Removed:\n");
2332 	list_for_each_entry_safe(dd, tmp, &removing_list, remove_list) {
2333 		if (dd->pdev) {
2334 			if (dd->port &&
2335 			    dd->port->identify &&
2336 			    dd->port->identify_valid) {
2337 				strlcpy(id_buf,
2338 					(char *) (dd->port->identify+10), 21);
2339 				status = *(dd->port->identify + 141);
2340 			} else {
2341 				memset(id_buf, 0, 42);
2342 				status = 0;
2343 			}
2344 
2345 			if (dd->port &&
2346 			    test_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags)) {
2347 				size += sprintf(&buf[size],
2348 					" device %s %s (ftl rebuild %d %%)\n",
2349 					dev_name(&dd->pdev->dev),
2350 					id_buf,
2351 					status);
2352 			} else {
2353 				size += sprintf(&buf[size],
2354 					" device %s %s\n",
2355 					dev_name(&dd->pdev->dev),
2356 					id_buf);
2357 			}
2358 		}
2359 	}
2360 	spin_unlock_irqrestore(&dev_lock, flags);
2361 
2362 	return size;
2363 }
2364 
2365 static ssize_t mtip_hw_read_device_status(struct file *f, char __user *ubuf,
2366 						size_t len, loff_t *offset)
2367 {
2368 	struct driver_data *dd =  (struct driver_data *)f->private_data;
2369 	int size = *offset;
2370 	char *buf;
2371 	int rv = 0;
2372 
2373 	if (!len || *offset)
2374 		return 0;
2375 
2376 	buf = kzalloc(MTIP_DFS_MAX_BUF_SIZE, GFP_KERNEL);
2377 	if (!buf) {
2378 		dev_err(&dd->pdev->dev,
2379 			"Memory allocation: status buffer\n");
2380 		return -ENOMEM;
2381 	}
2382 
2383 	size += show_device_status(NULL, buf);
2384 
2385 	*offset = size <= len ? size : len;
2386 	size = copy_to_user(ubuf, buf, *offset);
2387 	if (size)
2388 		rv = -EFAULT;
2389 
2390 	kfree(buf);
2391 	return rv ? rv : *offset;
2392 }
2393 
2394 static ssize_t mtip_hw_read_registers(struct file *f, char __user *ubuf,
2395 				  size_t len, loff_t *offset)
2396 {
2397 	struct driver_data *dd =  (struct driver_data *)f->private_data;
2398 	char *buf;
2399 	u32 group_allocated;
2400 	int size = *offset;
2401 	int n, rv = 0;
2402 
2403 	if (!len || size)
2404 		return 0;
2405 
2406 	buf = kzalloc(MTIP_DFS_MAX_BUF_SIZE, GFP_KERNEL);
2407 	if (!buf) {
2408 		dev_err(&dd->pdev->dev,
2409 			"Memory allocation: register buffer\n");
2410 		return -ENOMEM;
2411 	}
2412 
2413 	size += sprintf(&buf[size], "H/ S ACTive      : [ 0x");
2414 
2415 	for (n = dd->slot_groups-1; n >= 0; n--)
2416 		size += sprintf(&buf[size], "%08X ",
2417 					 readl(dd->port->s_active[n]));
2418 
2419 	size += sprintf(&buf[size], "]\n");
2420 	size += sprintf(&buf[size], "H/ Command Issue : [ 0x");
2421 
2422 	for (n = dd->slot_groups-1; n >= 0; n--)
2423 		size += sprintf(&buf[size], "%08X ",
2424 					readl(dd->port->cmd_issue[n]));
2425 
2426 	size += sprintf(&buf[size], "]\n");
2427 	size += sprintf(&buf[size], "H/ Completed     : [ 0x");
2428 
2429 	for (n = dd->slot_groups-1; n >= 0; n--)
2430 		size += sprintf(&buf[size], "%08X ",
2431 				readl(dd->port->completed[n]));
2432 
2433 	size += sprintf(&buf[size], "]\n");
2434 	size += sprintf(&buf[size], "H/ PORT IRQ STAT : [ 0x%08X ]\n",
2435 				readl(dd->port->mmio + PORT_IRQ_STAT));
2436 	size += sprintf(&buf[size], "H/ HOST IRQ STAT : [ 0x%08X ]\n",
2437 				readl(dd->mmio + HOST_IRQ_STAT));
2438 	size += sprintf(&buf[size], "\n");
2439 
2440 	size += sprintf(&buf[size], "L/ Commands in Q : [ 0x");
2441 
2442 	for (n = dd->slot_groups-1; n >= 0; n--) {
2443 		if (sizeof(long) > sizeof(u32))
2444 			group_allocated =
2445 				dd->port->cmds_to_issue[n/2] >> (32*(n&1));
2446 		else
2447 			group_allocated = dd->port->cmds_to_issue[n];
2448 		size += sprintf(&buf[size], "%08X ", group_allocated);
2449 	}
2450 	size += sprintf(&buf[size], "]\n");
2451 
2452 	*offset = size <= len ? size : len;
2453 	size = copy_to_user(ubuf, buf, *offset);
2454 	if (size)
2455 		rv = -EFAULT;
2456 
2457 	kfree(buf);
2458 	return rv ? rv : *offset;
2459 }
2460 
2461 static ssize_t mtip_hw_read_flags(struct file *f, char __user *ubuf,
2462 				  size_t len, loff_t *offset)
2463 {
2464 	struct driver_data *dd =  (struct driver_data *)f->private_data;
2465 	char *buf;
2466 	int size = *offset;
2467 	int rv = 0;
2468 
2469 	if (!len || size)
2470 		return 0;
2471 
2472 	buf = kzalloc(MTIP_DFS_MAX_BUF_SIZE, GFP_KERNEL);
2473 	if (!buf) {
2474 		dev_err(&dd->pdev->dev,
2475 			"Memory allocation: flag buffer\n");
2476 		return -ENOMEM;
2477 	}
2478 
2479 	size += sprintf(&buf[size], "Flag-port : [ %08lX ]\n",
2480 							dd->port->flags);
2481 	size += sprintf(&buf[size], "Flag-dd   : [ %08lX ]\n",
2482 							dd->dd_flag);
2483 
2484 	*offset = size <= len ? size : len;
2485 	size = copy_to_user(ubuf, buf, *offset);
2486 	if (size)
2487 		rv = -EFAULT;
2488 
2489 	kfree(buf);
2490 	return rv ? rv : *offset;
2491 }
2492 
2493 static const struct file_operations mtip_device_status_fops = {
2494 	.owner  = THIS_MODULE,
2495 	.open   = simple_open,
2496 	.read   = mtip_hw_read_device_status,
2497 	.llseek = no_llseek,
2498 };
2499 
2500 static const struct file_operations mtip_regs_fops = {
2501 	.owner  = THIS_MODULE,
2502 	.open   = simple_open,
2503 	.read   = mtip_hw_read_registers,
2504 	.llseek = no_llseek,
2505 };
2506 
2507 static const struct file_operations mtip_flags_fops = {
2508 	.owner  = THIS_MODULE,
2509 	.open   = simple_open,
2510 	.read   = mtip_hw_read_flags,
2511 	.llseek = no_llseek,
2512 };
2513 
2514 /*
2515  * Create the sysfs related attributes.
2516  *
2517  * @dd   Pointer to the driver data structure.
2518  * @kobj Pointer to the kobj for the block device.
2519  *
2520  * return value
2521  *	0	Operation completed successfully.
2522  *	-EINVAL Invalid parameter.
2523  */
2524 static int mtip_hw_sysfs_init(struct driver_data *dd, struct kobject *kobj)
2525 {
2526 	if (!kobj || !dd)
2527 		return -EINVAL;
2528 
2529 	if (sysfs_create_file(kobj, &dev_attr_status.attr))
2530 		dev_warn(&dd->pdev->dev,
2531 			"Error creating 'status' sysfs entry\n");
2532 	return 0;
2533 }
2534 
2535 /*
2536  * Remove the sysfs related attributes.
2537  *
2538  * @dd   Pointer to the driver data structure.
2539  * @kobj Pointer to the kobj for the block device.
2540  *
2541  * return value
2542  *	0	Operation completed successfully.
2543  *	-EINVAL Invalid parameter.
2544  */
2545 static int mtip_hw_sysfs_exit(struct driver_data *dd, struct kobject *kobj)
2546 {
2547 	if (!kobj || !dd)
2548 		return -EINVAL;
2549 
2550 	sysfs_remove_file(kobj, &dev_attr_status.attr);
2551 
2552 	return 0;
2553 }
2554 
2555 static int mtip_hw_debugfs_init(struct driver_data *dd)
2556 {
2557 	if (!dfs_parent)
2558 		return -1;
2559 
2560 	dd->dfs_node = debugfs_create_dir(dd->disk->disk_name, dfs_parent);
2561 	if (IS_ERR_OR_NULL(dd->dfs_node)) {
2562 		dev_warn(&dd->pdev->dev,
2563 			"Error creating node %s under debugfs\n",
2564 						dd->disk->disk_name);
2565 		dd->dfs_node = NULL;
2566 		return -1;
2567 	}
2568 
2569 	debugfs_create_file("flags", 0444, dd->dfs_node, dd, &mtip_flags_fops);
2570 	debugfs_create_file("registers", 0444, dd->dfs_node, dd,
2571 			    &mtip_regs_fops);
2572 
2573 	return 0;
2574 }
2575 
2576 static void mtip_hw_debugfs_exit(struct driver_data *dd)
2577 {
2578 	if (dd->dfs_node)
2579 		debugfs_remove_recursive(dd->dfs_node);
2580 }
2581 
2582 /*
2583  * Perform any init/resume time hardware setup
2584  *
2585  * @dd Pointer to the driver data structure.
2586  *
2587  * return value
2588  *	None
2589  */
2590 static inline void hba_setup(struct driver_data *dd)
2591 {
2592 	u32 hwdata;
2593 	hwdata = readl(dd->mmio + HOST_HSORG);
2594 
2595 	/* interrupt bug workaround: use only 1 IS bit.*/
2596 	writel(hwdata |
2597 		HSORG_DISABLE_SLOTGRP_INTR |
2598 		HSORG_DISABLE_SLOTGRP_PXIS,
2599 		dd->mmio + HOST_HSORG);
2600 }
2601 
2602 static int mtip_device_unaligned_constrained(struct driver_data *dd)
2603 {
2604 	return (dd->pdev->device == P420M_DEVICE_ID ? 1 : 0);
2605 }
2606 
2607 /*
2608  * Detect the details of the product, and store anything needed
2609  * into the driver data structure.  This includes product type and
2610  * version and number of slot groups.
2611  *
2612  * @dd Pointer to the driver data structure.
2613  *
2614  * return value
2615  *	None
2616  */
2617 static void mtip_detect_product(struct driver_data *dd)
2618 {
2619 	u32 hwdata;
2620 	unsigned int rev, slotgroups;
2621 
2622 	/*
2623 	 * HBA base + 0xFC [15:0] - vendor-specific hardware interface
2624 	 * info register:
2625 	 * [15:8] hardware/software interface rev#
2626 	 * [   3] asic-style interface
2627 	 * [ 2:0] number of slot groups, minus 1 (only valid for asic-style).
2628 	 */
2629 	hwdata = readl(dd->mmio + HOST_HSORG);
2630 
2631 	dd->product_type = MTIP_PRODUCT_UNKNOWN;
2632 	dd->slot_groups = 1;
2633 
2634 	if (hwdata & 0x8) {
2635 		dd->product_type = MTIP_PRODUCT_ASICFPGA;
2636 		rev = (hwdata & HSORG_HWREV) >> 8;
2637 		slotgroups = (hwdata & HSORG_SLOTGROUPS) + 1;
2638 		dev_info(&dd->pdev->dev,
2639 			"ASIC-FPGA design, HS rev 0x%x, "
2640 			"%i slot groups [%i slots]\n",
2641 			 rev,
2642 			 slotgroups,
2643 			 slotgroups * 32);
2644 
2645 		if (slotgroups > MTIP_MAX_SLOT_GROUPS) {
2646 			dev_warn(&dd->pdev->dev,
2647 				"Warning: driver only supports "
2648 				"%i slot groups.\n", MTIP_MAX_SLOT_GROUPS);
2649 			slotgroups = MTIP_MAX_SLOT_GROUPS;
2650 		}
2651 		dd->slot_groups = slotgroups;
2652 		return;
2653 	}
2654 
2655 	dev_warn(&dd->pdev->dev, "Unrecognized product id\n");
2656 }
2657 
2658 /*
2659  * Blocking wait for FTL rebuild to complete
2660  *
2661  * @dd Pointer to the DRIVER_DATA structure.
2662  *
2663  * return value
2664  *	0	FTL rebuild completed successfully
2665  *	-EFAULT FTL rebuild error/timeout/interruption
2666  */
2667 static int mtip_ftl_rebuild_poll(struct driver_data *dd)
2668 {
2669 	unsigned long timeout, cnt = 0, start;
2670 
2671 	dev_warn(&dd->pdev->dev,
2672 		"FTL rebuild in progress. Polling for completion.\n");
2673 
2674 	start = jiffies;
2675 	timeout = jiffies + msecs_to_jiffies(MTIP_FTL_REBUILD_TIMEOUT_MS);
2676 
2677 	do {
2678 		if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
2679 				&dd->dd_flag)))
2680 			return -EFAULT;
2681 		if (mtip_check_surprise_removal(dd->pdev))
2682 			return -EFAULT;
2683 
2684 		if (mtip_get_identify(dd->port, NULL) < 0)
2685 			return -EFAULT;
2686 
2687 		if (*(dd->port->identify + MTIP_FTL_REBUILD_OFFSET) ==
2688 			MTIP_FTL_REBUILD_MAGIC) {
2689 			ssleep(1);
2690 			/* Print message every 3 minutes */
2691 			if (cnt++ >= 180) {
2692 				dev_warn(&dd->pdev->dev,
2693 				"FTL rebuild in progress (%d secs).\n",
2694 				jiffies_to_msecs(jiffies - start) / 1000);
2695 				cnt = 0;
2696 			}
2697 		} else {
2698 			dev_warn(&dd->pdev->dev,
2699 				"FTL rebuild complete (%d secs).\n",
2700 			jiffies_to_msecs(jiffies - start) / 1000);
2701 			mtip_block_initialize(dd);
2702 			return 0;
2703 		}
2704 	} while (time_before(jiffies, timeout));
2705 
2706 	/* Check for timeout */
2707 	dev_err(&dd->pdev->dev,
2708 		"Timed out waiting for FTL rebuild to complete (%d secs).\n",
2709 		jiffies_to_msecs(jiffies - start) / 1000);
2710 	return -EFAULT;
2711 }
2712 
2713 static void mtip_softirq_done_fn(struct request *rq)
2714 {
2715 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
2716 	struct driver_data *dd = rq->q->queuedata;
2717 
2718 	/* Unmap the DMA scatter list entries */
2719 	dma_unmap_sg(&dd->pdev->dev, cmd->sg, cmd->scatter_ents,
2720 							cmd->direction);
2721 
2722 	if (unlikely(cmd->unaligned))
2723 		up(&dd->port->cmd_slot_unal);
2724 
2725 	blk_mq_end_request(rq, cmd->status);
2726 }
2727 
2728 static void mtip_abort_cmd(struct request *req, void *data, bool reserved)
2729 {
2730 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(req);
2731 	struct driver_data *dd = data;
2732 
2733 	dbg_printk(MTIP_DRV_NAME " Aborting request, tag = %d\n", req->tag);
2734 
2735 	clear_bit(req->tag, dd->port->cmds_to_issue);
2736 	cmd->status = BLK_STS_IOERR;
2737 	mtip_softirq_done_fn(req);
2738 }
2739 
2740 static void mtip_queue_cmd(struct request *req, void *data, bool reserved)
2741 {
2742 	struct driver_data *dd = data;
2743 
2744 	set_bit(req->tag, dd->port->cmds_to_issue);
2745 	blk_abort_request(req);
2746 }
2747 
2748 /*
2749  * service thread to issue queued commands
2750  *
2751  * @data Pointer to the driver data structure.
2752  *
2753  * return value
2754  *	0
2755  */
2756 
2757 static int mtip_service_thread(void *data)
2758 {
2759 	struct driver_data *dd = (struct driver_data *)data;
2760 	unsigned long slot, slot_start, slot_wrap, to;
2761 	unsigned int num_cmd_slots = dd->slot_groups * 32;
2762 	struct mtip_port *port = dd->port;
2763 
2764 	while (1) {
2765 		if (kthread_should_stop() ||
2766 			test_bit(MTIP_PF_SVC_THD_STOP_BIT, &port->flags))
2767 			goto st_out;
2768 		clear_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags);
2769 
2770 		/*
2771 		 * the condition is to check neither an internal command is
2772 		 * is in progress nor error handling is active
2773 		 */
2774 		wait_event_interruptible(port->svc_wait, (port->flags) &&
2775 			(port->flags & MTIP_PF_SVC_THD_WORK));
2776 
2777 		if (kthread_should_stop() ||
2778 			test_bit(MTIP_PF_SVC_THD_STOP_BIT, &port->flags))
2779 			goto st_out;
2780 
2781 		if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
2782 				&dd->dd_flag)))
2783 			goto st_out;
2784 
2785 		set_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags);
2786 
2787 restart_eh:
2788 		/* Demux bits: start with error handling */
2789 		if (test_bit(MTIP_PF_EH_ACTIVE_BIT, &port->flags)) {
2790 			mtip_handle_tfe(dd);
2791 			clear_bit(MTIP_PF_EH_ACTIVE_BIT, &port->flags);
2792 		}
2793 
2794 		if (test_bit(MTIP_PF_EH_ACTIVE_BIT, &port->flags))
2795 			goto restart_eh;
2796 
2797 		if (test_bit(MTIP_PF_TO_ACTIVE_BIT, &port->flags)) {
2798 			to = jiffies + msecs_to_jiffies(5000);
2799 
2800 			do {
2801 				mdelay(100);
2802 			} while (atomic_read(&dd->irq_workers_active) != 0 &&
2803 				time_before(jiffies, to));
2804 
2805 			if (atomic_read(&dd->irq_workers_active) != 0)
2806 				dev_warn(&dd->pdev->dev,
2807 					"Completion workers still active!");
2808 
2809 			blk_mq_quiesce_queue(dd->queue);
2810 
2811 			spin_lock(dd->queue->queue_lock);
2812 			blk_mq_tagset_busy_iter(&dd->tags,
2813 							mtip_queue_cmd, dd);
2814 			spin_unlock(dd->queue->queue_lock);
2815 
2816 			set_bit(MTIP_PF_ISSUE_CMDS_BIT, &dd->port->flags);
2817 
2818 			if (mtip_device_reset(dd))
2819 				blk_mq_tagset_busy_iter(&dd->tags,
2820 							mtip_abort_cmd, dd);
2821 
2822 			clear_bit(MTIP_PF_TO_ACTIVE_BIT, &dd->port->flags);
2823 
2824 			blk_mq_unquiesce_queue(dd->queue);
2825 		}
2826 
2827 		if (test_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags)) {
2828 			slot = 1;
2829 			/* used to restrict the loop to one iteration */
2830 			slot_start = num_cmd_slots;
2831 			slot_wrap = 0;
2832 			while (1) {
2833 				slot = find_next_bit(port->cmds_to_issue,
2834 						num_cmd_slots, slot);
2835 				if (slot_wrap == 1) {
2836 					if ((slot_start >= slot) ||
2837 						(slot >= num_cmd_slots))
2838 						break;
2839 				}
2840 				if (unlikely(slot_start == num_cmd_slots))
2841 					slot_start = slot;
2842 
2843 				if (unlikely(slot == num_cmd_slots)) {
2844 					slot = 1;
2845 					slot_wrap = 1;
2846 					continue;
2847 				}
2848 
2849 				/* Issue the command to the hardware */
2850 				mtip_issue_ncq_command(port, slot);
2851 
2852 				clear_bit(slot, port->cmds_to_issue);
2853 			}
2854 
2855 			clear_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags);
2856 		}
2857 
2858 		if (test_bit(MTIP_PF_REBUILD_BIT, &port->flags)) {
2859 			if (mtip_ftl_rebuild_poll(dd) == 0)
2860 				clear_bit(MTIP_PF_REBUILD_BIT, &port->flags);
2861 		}
2862 	}
2863 
2864 st_out:
2865 	return 0;
2866 }
2867 
2868 /*
2869  * DMA region teardown
2870  *
2871  * @dd Pointer to driver_data structure
2872  *
2873  * return value
2874  *      None
2875  */
2876 static void mtip_dma_free(struct driver_data *dd)
2877 {
2878 	struct mtip_port *port = dd->port;
2879 
2880 	if (port->block1)
2881 		dmam_free_coherent(&dd->pdev->dev, BLOCK_DMA_ALLOC_SZ,
2882 					port->block1, port->block1_dma);
2883 
2884 	if (port->command_list) {
2885 		dmam_free_coherent(&dd->pdev->dev, AHCI_CMD_TBL_SZ,
2886 				port->command_list, port->command_list_dma);
2887 	}
2888 }
2889 
2890 /*
2891  * DMA region setup
2892  *
2893  * @dd Pointer to driver_data structure
2894  *
2895  * return value
2896  *      -ENOMEM Not enough free DMA region space to initialize driver
2897  */
2898 static int mtip_dma_alloc(struct driver_data *dd)
2899 {
2900 	struct mtip_port *port = dd->port;
2901 
2902 	/* Allocate dma memory for RX Fis, Identify, and Sector Bufffer */
2903 	port->block1 =
2904 		dmam_alloc_coherent(&dd->pdev->dev, BLOCK_DMA_ALLOC_SZ,
2905 					&port->block1_dma, GFP_KERNEL);
2906 	if (!port->block1)
2907 		return -ENOMEM;
2908 	memset(port->block1, 0, BLOCK_DMA_ALLOC_SZ);
2909 
2910 	/* Allocate dma memory for command list */
2911 	port->command_list =
2912 		dmam_alloc_coherent(&dd->pdev->dev, AHCI_CMD_TBL_SZ,
2913 					&port->command_list_dma, GFP_KERNEL);
2914 	if (!port->command_list) {
2915 		dmam_free_coherent(&dd->pdev->dev, BLOCK_DMA_ALLOC_SZ,
2916 					port->block1, port->block1_dma);
2917 		port->block1 = NULL;
2918 		port->block1_dma = 0;
2919 		return -ENOMEM;
2920 	}
2921 	memset(port->command_list, 0, AHCI_CMD_TBL_SZ);
2922 
2923 	/* Setup all pointers into first DMA region */
2924 	port->rxfis         = port->block1 + AHCI_RX_FIS_OFFSET;
2925 	port->rxfis_dma     = port->block1_dma + AHCI_RX_FIS_OFFSET;
2926 	port->identify      = port->block1 + AHCI_IDFY_OFFSET;
2927 	port->identify_dma  = port->block1_dma + AHCI_IDFY_OFFSET;
2928 	port->log_buf       = port->block1 + AHCI_SECTBUF_OFFSET;
2929 	port->log_buf_dma   = port->block1_dma + AHCI_SECTBUF_OFFSET;
2930 	port->smart_buf     = port->block1 + AHCI_SMARTBUF_OFFSET;
2931 	port->smart_buf_dma = port->block1_dma + AHCI_SMARTBUF_OFFSET;
2932 
2933 	return 0;
2934 }
2935 
2936 static int mtip_hw_get_identify(struct driver_data *dd)
2937 {
2938 	struct smart_attr attr242;
2939 	unsigned char *buf;
2940 	int rv;
2941 
2942 	if (mtip_get_identify(dd->port, NULL) < 0)
2943 		return -EFAULT;
2944 
2945 	if (*(dd->port->identify + MTIP_FTL_REBUILD_OFFSET) ==
2946 		MTIP_FTL_REBUILD_MAGIC) {
2947 		set_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags);
2948 		return MTIP_FTL_REBUILD_MAGIC;
2949 	}
2950 	mtip_dump_identify(dd->port);
2951 
2952 	/* check write protect, over temp and rebuild statuses */
2953 	rv = mtip_read_log_page(dd->port, ATA_LOG_SATA_NCQ,
2954 				dd->port->log_buf,
2955 				dd->port->log_buf_dma, 1);
2956 	if (rv) {
2957 		dev_warn(&dd->pdev->dev,
2958 			"Error in READ LOG EXT (10h) command\n");
2959 		/* non-critical error, don't fail the load */
2960 	} else {
2961 		buf = (unsigned char *)dd->port->log_buf;
2962 		if (buf[259] & 0x1) {
2963 			dev_info(&dd->pdev->dev,
2964 				"Write protect bit is set.\n");
2965 			set_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag);
2966 		}
2967 		if (buf[288] == 0xF7) {
2968 			dev_info(&dd->pdev->dev,
2969 				"Exceeded Tmax, drive in thermal shutdown.\n");
2970 			set_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag);
2971 		}
2972 		if (buf[288] == 0xBF) {
2973 			dev_info(&dd->pdev->dev,
2974 				"Drive indicates rebuild has failed.\n");
2975 			set_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag);
2976 		}
2977 	}
2978 
2979 	/* get write protect progess */
2980 	memset(&attr242, 0, sizeof(struct smart_attr));
2981 	if (mtip_get_smart_attr(dd->port, 242, &attr242))
2982 		dev_warn(&dd->pdev->dev,
2983 				"Unable to check write protect progress\n");
2984 	else
2985 		dev_info(&dd->pdev->dev,
2986 				"Write protect progress: %u%% (%u blocks)\n",
2987 				attr242.cur, le32_to_cpu(attr242.data));
2988 
2989 	return rv;
2990 }
2991 
2992 /*
2993  * Called once for each card.
2994  *
2995  * @dd Pointer to the driver data structure.
2996  *
2997  * return value
2998  *	0 on success, else an error code.
2999  */
3000 static int mtip_hw_init(struct driver_data *dd)
3001 {
3002 	int i;
3003 	int rv;
3004 	unsigned long timeout, timetaken;
3005 
3006 	dd->mmio = pcim_iomap_table(dd->pdev)[MTIP_ABAR];
3007 
3008 	mtip_detect_product(dd);
3009 	if (dd->product_type == MTIP_PRODUCT_UNKNOWN) {
3010 		rv = -EIO;
3011 		goto out1;
3012 	}
3013 
3014 	hba_setup(dd);
3015 
3016 	dd->port = kzalloc_node(sizeof(struct mtip_port), GFP_KERNEL,
3017 				dd->numa_node);
3018 	if (!dd->port) {
3019 		dev_err(&dd->pdev->dev,
3020 			"Memory allocation: port structure\n");
3021 		return -ENOMEM;
3022 	}
3023 
3024 	/* Continue workqueue setup */
3025 	for (i = 0; i < MTIP_MAX_SLOT_GROUPS; i++)
3026 		dd->work[i].port = dd->port;
3027 
3028 	/* Enable unaligned IO constraints for some devices */
3029 	if (mtip_device_unaligned_constrained(dd))
3030 		dd->unal_qdepth = MTIP_MAX_UNALIGNED_SLOTS;
3031 	else
3032 		dd->unal_qdepth = 0;
3033 
3034 	sema_init(&dd->port->cmd_slot_unal, dd->unal_qdepth);
3035 
3036 	/* Spinlock to prevent concurrent issue */
3037 	for (i = 0; i < MTIP_MAX_SLOT_GROUPS; i++)
3038 		spin_lock_init(&dd->port->cmd_issue_lock[i]);
3039 
3040 	/* Set the port mmio base address. */
3041 	dd->port->mmio	= dd->mmio + PORT_OFFSET;
3042 	dd->port->dd	= dd;
3043 
3044 	/* DMA allocations */
3045 	rv = mtip_dma_alloc(dd);
3046 	if (rv < 0)
3047 		goto out1;
3048 
3049 	/* Setup the pointers to the extended s_active and CI registers. */
3050 	for (i = 0; i < dd->slot_groups; i++) {
3051 		dd->port->s_active[i] =
3052 			dd->port->mmio + i*0x80 + PORT_SCR_ACT;
3053 		dd->port->cmd_issue[i] =
3054 			dd->port->mmio + i*0x80 + PORT_COMMAND_ISSUE;
3055 		dd->port->completed[i] =
3056 			dd->port->mmio + i*0x80 + PORT_SDBV;
3057 	}
3058 
3059 	timetaken = jiffies;
3060 	timeout = jiffies + msecs_to_jiffies(30000);
3061 	while (((readl(dd->port->mmio + PORT_SCR_STAT) & 0x0F) != 0x03) &&
3062 		 time_before(jiffies, timeout)) {
3063 		mdelay(100);
3064 	}
3065 	if (unlikely(mtip_check_surprise_removal(dd->pdev))) {
3066 		timetaken = jiffies - timetaken;
3067 		dev_warn(&dd->pdev->dev,
3068 			"Surprise removal detected at %u ms\n",
3069 			jiffies_to_msecs(timetaken));
3070 		rv = -ENODEV;
3071 		goto out2 ;
3072 	}
3073 	if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag))) {
3074 		timetaken = jiffies - timetaken;
3075 		dev_warn(&dd->pdev->dev,
3076 			"Removal detected at %u ms\n",
3077 			jiffies_to_msecs(timetaken));
3078 		rv = -EFAULT;
3079 		goto out2;
3080 	}
3081 
3082 	/* Conditionally reset the HBA. */
3083 	if (!(readl(dd->mmio + HOST_CAP) & HOST_CAP_NZDMA)) {
3084 		if (mtip_hba_reset(dd) < 0) {
3085 			dev_err(&dd->pdev->dev,
3086 				"Card did not reset within timeout\n");
3087 			rv = -EIO;
3088 			goto out2;
3089 		}
3090 	} else {
3091 		/* Clear any pending interrupts on the HBA */
3092 		writel(readl(dd->mmio + HOST_IRQ_STAT),
3093 			dd->mmio + HOST_IRQ_STAT);
3094 	}
3095 
3096 	mtip_init_port(dd->port);
3097 	mtip_start_port(dd->port);
3098 
3099 	/* Setup the ISR and enable interrupts. */
3100 	rv = devm_request_irq(&dd->pdev->dev,
3101 				dd->pdev->irq,
3102 				mtip_irq_handler,
3103 				IRQF_SHARED,
3104 				dev_driver_string(&dd->pdev->dev),
3105 				dd);
3106 
3107 	if (rv) {
3108 		dev_err(&dd->pdev->dev,
3109 			"Unable to allocate IRQ %d\n", dd->pdev->irq);
3110 		goto out2;
3111 	}
3112 	irq_set_affinity_hint(dd->pdev->irq, get_cpu_mask(dd->isr_binding));
3113 
3114 	/* Enable interrupts on the HBA. */
3115 	writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN,
3116 					dd->mmio + HOST_CTL);
3117 
3118 	init_waitqueue_head(&dd->port->svc_wait);
3119 
3120 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)) {
3121 		rv = -EFAULT;
3122 		goto out3;
3123 	}
3124 
3125 	return rv;
3126 
3127 out3:
3128 	/* Disable interrupts on the HBA. */
3129 	writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN,
3130 			dd->mmio + HOST_CTL);
3131 
3132 	/* Release the IRQ. */
3133 	irq_set_affinity_hint(dd->pdev->irq, NULL);
3134 	devm_free_irq(&dd->pdev->dev, dd->pdev->irq, dd);
3135 
3136 out2:
3137 	mtip_deinit_port(dd->port);
3138 	mtip_dma_free(dd);
3139 
3140 out1:
3141 	/* Free the memory allocated for the for structure. */
3142 	kfree(dd->port);
3143 
3144 	return rv;
3145 }
3146 
3147 static int mtip_standby_drive(struct driver_data *dd)
3148 {
3149 	int rv = 0;
3150 
3151 	if (dd->sr || !dd->port)
3152 		return -ENODEV;
3153 	/*
3154 	 * Send standby immediate (E0h) to the drive so that it
3155 	 * saves its state.
3156 	 */
3157 	if (!test_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags) &&
3158 	    !test_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag) &&
3159 	    !test_bit(MTIP_DDF_SEC_LOCK_BIT, &dd->dd_flag)) {
3160 		rv = mtip_standby_immediate(dd->port);
3161 		if (rv)
3162 			dev_warn(&dd->pdev->dev,
3163 				"STANDBY IMMEDIATE failed\n");
3164 	}
3165 	return rv;
3166 }
3167 
3168 /*
3169  * Called to deinitialize an interface.
3170  *
3171  * @dd Pointer to the driver data structure.
3172  *
3173  * return value
3174  *	0
3175  */
3176 static int mtip_hw_exit(struct driver_data *dd)
3177 {
3178 	if (!dd->sr) {
3179 		/* de-initialize the port. */
3180 		mtip_deinit_port(dd->port);
3181 
3182 		/* Disable interrupts on the HBA. */
3183 		writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN,
3184 				dd->mmio + HOST_CTL);
3185 	}
3186 
3187 	/* Release the IRQ. */
3188 	irq_set_affinity_hint(dd->pdev->irq, NULL);
3189 	devm_free_irq(&dd->pdev->dev, dd->pdev->irq, dd);
3190 	msleep(1000);
3191 
3192 	/* Free dma regions */
3193 	mtip_dma_free(dd);
3194 
3195 	/* Free the memory allocated for the for structure. */
3196 	kfree(dd->port);
3197 	dd->port = NULL;
3198 
3199 	return 0;
3200 }
3201 
3202 /*
3203  * Issue a Standby Immediate command to the device.
3204  *
3205  * This function is called by the Block Layer just before the
3206  * system powers off during a shutdown.
3207  *
3208  * @dd Pointer to the driver data structure.
3209  *
3210  * return value
3211  *	0
3212  */
3213 static int mtip_hw_shutdown(struct driver_data *dd)
3214 {
3215 	/*
3216 	 * Send standby immediate (E0h) to the drive so that it
3217 	 * saves its state.
3218 	 */
3219 	mtip_standby_drive(dd);
3220 
3221 	return 0;
3222 }
3223 
3224 /*
3225  * Suspend function
3226  *
3227  * This function is called by the Block Layer just before the
3228  * system hibernates.
3229  *
3230  * @dd Pointer to the driver data structure.
3231  *
3232  * return value
3233  *	0	Suspend was successful
3234  *	-EFAULT Suspend was not successful
3235  */
3236 static int mtip_hw_suspend(struct driver_data *dd)
3237 {
3238 	/*
3239 	 * Send standby immediate (E0h) to the drive
3240 	 * so that it saves its state.
3241 	 */
3242 	if (mtip_standby_drive(dd) != 0) {
3243 		dev_err(&dd->pdev->dev,
3244 			"Failed standby-immediate command\n");
3245 		return -EFAULT;
3246 	}
3247 
3248 	/* Disable interrupts on the HBA.*/
3249 	writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN,
3250 			dd->mmio + HOST_CTL);
3251 	mtip_deinit_port(dd->port);
3252 
3253 	return 0;
3254 }
3255 
3256 /*
3257  * Resume function
3258  *
3259  * This function is called by the Block Layer as the
3260  * system resumes.
3261  *
3262  * @dd Pointer to the driver data structure.
3263  *
3264  * return value
3265  *	0	Resume was successful
3266  *      -EFAULT Resume was not successful
3267  */
3268 static int mtip_hw_resume(struct driver_data *dd)
3269 {
3270 	/* Perform any needed hardware setup steps */
3271 	hba_setup(dd);
3272 
3273 	/* Reset the HBA */
3274 	if (mtip_hba_reset(dd) != 0) {
3275 		dev_err(&dd->pdev->dev,
3276 			"Unable to reset the HBA\n");
3277 		return -EFAULT;
3278 	}
3279 
3280 	/*
3281 	 * Enable the port, DMA engine, and FIS reception specific
3282 	 * h/w in controller.
3283 	 */
3284 	mtip_init_port(dd->port);
3285 	mtip_start_port(dd->port);
3286 
3287 	/* Enable interrupts on the HBA.*/
3288 	writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN,
3289 			dd->mmio + HOST_CTL);
3290 
3291 	return 0;
3292 }
3293 
3294 /*
3295  * Helper function for reusing disk name
3296  * upon hot insertion.
3297  */
3298 static int rssd_disk_name_format(char *prefix,
3299 				 int index,
3300 				 char *buf,
3301 				 int buflen)
3302 {
3303 	const int base = 'z' - 'a' + 1;
3304 	char *begin = buf + strlen(prefix);
3305 	char *end = buf + buflen;
3306 	char *p;
3307 	int unit;
3308 
3309 	p = end - 1;
3310 	*p = '\0';
3311 	unit = base;
3312 	do {
3313 		if (p == begin)
3314 			return -EINVAL;
3315 		*--p = 'a' + (index % unit);
3316 		index = (index / unit) - 1;
3317 	} while (index >= 0);
3318 
3319 	memmove(begin, p, end - p);
3320 	memcpy(buf, prefix, strlen(prefix));
3321 
3322 	return 0;
3323 }
3324 
3325 /*
3326  * Block layer IOCTL handler.
3327  *
3328  * @dev Pointer to the block_device structure.
3329  * @mode ignored
3330  * @cmd IOCTL command passed from the user application.
3331  * @arg Argument passed from the user application.
3332  *
3333  * return value
3334  *	0        IOCTL completed successfully.
3335  *	-ENOTTY  IOCTL not supported or invalid driver data
3336  *                 structure pointer.
3337  */
3338 static int mtip_block_ioctl(struct block_device *dev,
3339 			    fmode_t mode,
3340 			    unsigned cmd,
3341 			    unsigned long arg)
3342 {
3343 	struct driver_data *dd = dev->bd_disk->private_data;
3344 
3345 	if (!capable(CAP_SYS_ADMIN))
3346 		return -EACCES;
3347 
3348 	if (!dd)
3349 		return -ENOTTY;
3350 
3351 	if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)))
3352 		return -ENOTTY;
3353 
3354 	switch (cmd) {
3355 	case BLKFLSBUF:
3356 		return -ENOTTY;
3357 	default:
3358 		return mtip_hw_ioctl(dd, cmd, arg);
3359 	}
3360 }
3361 
3362 #ifdef CONFIG_COMPAT
3363 /*
3364  * Block layer compat IOCTL handler.
3365  *
3366  * @dev Pointer to the block_device structure.
3367  * @mode ignored
3368  * @cmd IOCTL command passed from the user application.
3369  * @arg Argument passed from the user application.
3370  *
3371  * return value
3372  *	0        IOCTL completed successfully.
3373  *	-ENOTTY  IOCTL not supported or invalid driver data
3374  *                 structure pointer.
3375  */
3376 static int mtip_block_compat_ioctl(struct block_device *dev,
3377 			    fmode_t mode,
3378 			    unsigned cmd,
3379 			    unsigned long arg)
3380 {
3381 	struct driver_data *dd = dev->bd_disk->private_data;
3382 
3383 	if (!capable(CAP_SYS_ADMIN))
3384 		return -EACCES;
3385 
3386 	if (!dd)
3387 		return -ENOTTY;
3388 
3389 	if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)))
3390 		return -ENOTTY;
3391 
3392 	switch (cmd) {
3393 	case BLKFLSBUF:
3394 		return -ENOTTY;
3395 	case HDIO_DRIVE_TASKFILE: {
3396 		struct mtip_compat_ide_task_request_s __user *compat_req_task;
3397 		ide_task_request_t req_task;
3398 		int compat_tasksize, outtotal, ret;
3399 
3400 		compat_tasksize =
3401 			sizeof(struct mtip_compat_ide_task_request_s);
3402 
3403 		compat_req_task =
3404 			(struct mtip_compat_ide_task_request_s __user *) arg;
3405 
3406 		if (copy_from_user(&req_task, (void __user *) arg,
3407 			compat_tasksize - (2 * sizeof(compat_long_t))))
3408 			return -EFAULT;
3409 
3410 		if (get_user(req_task.out_size, &compat_req_task->out_size))
3411 			return -EFAULT;
3412 
3413 		if (get_user(req_task.in_size, &compat_req_task->in_size))
3414 			return -EFAULT;
3415 
3416 		outtotal = sizeof(struct mtip_compat_ide_task_request_s);
3417 
3418 		ret = exec_drive_taskfile(dd, (void __user *) arg,
3419 						&req_task, outtotal);
3420 
3421 		if (copy_to_user((void __user *) arg, &req_task,
3422 				compat_tasksize -
3423 				(2 * sizeof(compat_long_t))))
3424 			return -EFAULT;
3425 
3426 		if (put_user(req_task.out_size, &compat_req_task->out_size))
3427 			return -EFAULT;
3428 
3429 		if (put_user(req_task.in_size, &compat_req_task->in_size))
3430 			return -EFAULT;
3431 
3432 		return ret;
3433 	}
3434 	default:
3435 		return mtip_hw_ioctl(dd, cmd, arg);
3436 	}
3437 }
3438 #endif
3439 
3440 /*
3441  * Obtain the geometry of the device.
3442  *
3443  * You may think that this function is obsolete, but some applications,
3444  * fdisk for example still used CHS values. This function describes the
3445  * device as having 224 heads and 56 sectors per cylinder. These values are
3446  * chosen so that each cylinder is aligned on a 4KB boundary. Since a
3447  * partition is described in terms of a start and end cylinder this means
3448  * that each partition is also 4KB aligned. Non-aligned partitions adversely
3449  * affects performance.
3450  *
3451  * @dev Pointer to the block_device strucutre.
3452  * @geo Pointer to a hd_geometry structure.
3453  *
3454  * return value
3455  *	0       Operation completed successfully.
3456  *	-ENOTTY An error occurred while reading the drive capacity.
3457  */
3458 static int mtip_block_getgeo(struct block_device *dev,
3459 				struct hd_geometry *geo)
3460 {
3461 	struct driver_data *dd = dev->bd_disk->private_data;
3462 	sector_t capacity;
3463 
3464 	if (!dd)
3465 		return -ENOTTY;
3466 
3467 	if (!(mtip_hw_get_capacity(dd, &capacity))) {
3468 		dev_warn(&dd->pdev->dev,
3469 			"Could not get drive capacity.\n");
3470 		return -ENOTTY;
3471 	}
3472 
3473 	geo->heads = 224;
3474 	geo->sectors = 56;
3475 	sector_div(capacity, (geo->heads * geo->sectors));
3476 	geo->cylinders = capacity;
3477 	return 0;
3478 }
3479 
3480 static int mtip_block_open(struct block_device *dev, fmode_t mode)
3481 {
3482 	struct driver_data *dd;
3483 
3484 	if (dev && dev->bd_disk) {
3485 		dd = (struct driver_data *) dev->bd_disk->private_data;
3486 
3487 		if (dd) {
3488 			if (test_bit(MTIP_DDF_REMOVAL_BIT,
3489 							&dd->dd_flag)) {
3490 				return -ENODEV;
3491 			}
3492 			return 0;
3493 		}
3494 	}
3495 	return -ENODEV;
3496 }
3497 
3498 static void mtip_block_release(struct gendisk *disk, fmode_t mode)
3499 {
3500 }
3501 
3502 /*
3503  * Block device operation function.
3504  *
3505  * This structure contains pointers to the functions required by the block
3506  * layer.
3507  */
3508 static const struct block_device_operations mtip_block_ops = {
3509 	.open		= mtip_block_open,
3510 	.release	= mtip_block_release,
3511 	.ioctl		= mtip_block_ioctl,
3512 #ifdef CONFIG_COMPAT
3513 	.compat_ioctl	= mtip_block_compat_ioctl,
3514 #endif
3515 	.getgeo		= mtip_block_getgeo,
3516 	.owner		= THIS_MODULE
3517 };
3518 
3519 static inline bool is_se_active(struct driver_data *dd)
3520 {
3521 	if (unlikely(test_bit(MTIP_PF_SE_ACTIVE_BIT, &dd->port->flags))) {
3522 		if (dd->port->ic_pause_timer) {
3523 			unsigned long to = dd->port->ic_pause_timer +
3524 							msecs_to_jiffies(1000);
3525 			if (time_after(jiffies, to)) {
3526 				clear_bit(MTIP_PF_SE_ACTIVE_BIT,
3527 							&dd->port->flags);
3528 				clear_bit(MTIP_DDF_SEC_LOCK_BIT, &dd->dd_flag);
3529 				dd->port->ic_pause_timer = 0;
3530 				wake_up_interruptible(&dd->port->svc_wait);
3531 				return false;
3532 			}
3533 		}
3534 		return true;
3535 	}
3536 	return false;
3537 }
3538 
3539 /*
3540  * Block layer make request function.
3541  *
3542  * This function is called by the kernel to process a BIO for
3543  * the P320 device.
3544  *
3545  * @queue Pointer to the request queue. Unused other than to obtain
3546  *              the driver data structure.
3547  * @rq    Pointer to the request.
3548  *
3549  */
3550 static int mtip_submit_request(struct blk_mq_hw_ctx *hctx, struct request *rq)
3551 {
3552 	struct driver_data *dd = hctx->queue->queuedata;
3553 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3554 	unsigned int nents;
3555 
3556 	if (is_se_active(dd))
3557 		return -ENODATA;
3558 
3559 	if (unlikely(dd->dd_flag & MTIP_DDF_STOP_IO)) {
3560 		if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
3561 							&dd->dd_flag))) {
3562 			return -ENXIO;
3563 		}
3564 		if (unlikely(test_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag))) {
3565 			return -ENODATA;
3566 		}
3567 		if (unlikely(test_bit(MTIP_DDF_WRITE_PROTECT_BIT,
3568 							&dd->dd_flag) &&
3569 				rq_data_dir(rq))) {
3570 			return -ENODATA;
3571 		}
3572 		if (unlikely(test_bit(MTIP_DDF_SEC_LOCK_BIT, &dd->dd_flag) ||
3573 			test_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag)))
3574 			return -ENODATA;
3575 	}
3576 
3577 	if (req_op(rq) == REQ_OP_DISCARD) {
3578 		int err;
3579 
3580 		err = mtip_send_trim(dd, blk_rq_pos(rq), blk_rq_sectors(rq));
3581 		blk_mq_end_request(rq, err ? BLK_STS_IOERR : BLK_STS_OK);
3582 		return 0;
3583 	}
3584 
3585 	/* Create the scatter list for this request. */
3586 	nents = blk_rq_map_sg(hctx->queue, rq, cmd->sg);
3587 
3588 	/* Issue the read/write. */
3589 	mtip_hw_submit_io(dd, rq, cmd, nents, hctx);
3590 	return 0;
3591 }
3592 
3593 static bool mtip_check_unal_depth(struct blk_mq_hw_ctx *hctx,
3594 				  struct request *rq)
3595 {
3596 	struct driver_data *dd = hctx->queue->queuedata;
3597 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3598 
3599 	if (rq_data_dir(rq) == READ || !dd->unal_qdepth)
3600 		return false;
3601 
3602 	/*
3603 	 * If unaligned depth must be limited on this controller, mark it
3604 	 * as unaligned if the IO isn't on a 4k boundary (start of length).
3605 	 */
3606 	if (blk_rq_sectors(rq) <= 64) {
3607 		if ((blk_rq_pos(rq) & 7) || (blk_rq_sectors(rq) & 7))
3608 			cmd->unaligned = 1;
3609 	}
3610 
3611 	if (cmd->unaligned && down_trylock(&dd->port->cmd_slot_unal))
3612 		return true;
3613 
3614 	return false;
3615 }
3616 
3617 static blk_status_t mtip_issue_reserved_cmd(struct blk_mq_hw_ctx *hctx,
3618 		struct request *rq)
3619 {
3620 	struct driver_data *dd = hctx->queue->queuedata;
3621 	struct mtip_int_cmd *icmd = rq->special;
3622 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3623 	struct mtip_cmd_sg *command_sg;
3624 
3625 	if (mtip_commands_active(dd->port))
3626 		return BLK_STS_RESOURCE;
3627 
3628 	/* Populate the SG list */
3629 	cmd->command_header->opts =
3630 		 __force_bit2int cpu_to_le32(icmd->opts | icmd->fis_len);
3631 	if (icmd->buf_len) {
3632 		command_sg = cmd->command + AHCI_CMD_TBL_HDR_SZ;
3633 
3634 		command_sg->info =
3635 			__force_bit2int cpu_to_le32((icmd->buf_len-1) & 0x3FFFFF);
3636 		command_sg->dba	=
3637 			__force_bit2int cpu_to_le32(icmd->buffer & 0xFFFFFFFF);
3638 		command_sg->dba_upper =
3639 			__force_bit2int cpu_to_le32((icmd->buffer >> 16) >> 16);
3640 
3641 		cmd->command_header->opts |=
3642 			__force_bit2int cpu_to_le32((1 << 16));
3643 	}
3644 
3645 	/* Populate the command header */
3646 	cmd->command_header->byte_count = 0;
3647 
3648 	blk_mq_start_request(rq);
3649 	mtip_issue_non_ncq_command(dd->port, rq->tag);
3650 	return 0;
3651 }
3652 
3653 static blk_status_t mtip_queue_rq(struct blk_mq_hw_ctx *hctx,
3654 			 const struct blk_mq_queue_data *bd)
3655 {
3656 	struct request *rq = bd->rq;
3657 	int ret;
3658 
3659 	mtip_init_cmd_header(rq);
3660 
3661 	if (blk_rq_is_passthrough(rq))
3662 		return mtip_issue_reserved_cmd(hctx, rq);
3663 
3664 	if (unlikely(mtip_check_unal_depth(hctx, rq)))
3665 		return BLK_STS_RESOURCE;
3666 
3667 	blk_mq_start_request(rq);
3668 
3669 	ret = mtip_submit_request(hctx, rq);
3670 	if (likely(!ret))
3671 		return BLK_STS_OK;
3672 	return BLK_STS_IOERR;
3673 }
3674 
3675 static void mtip_free_cmd(struct blk_mq_tag_set *set, struct request *rq,
3676 			  unsigned int hctx_idx)
3677 {
3678 	struct driver_data *dd = set->driver_data;
3679 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3680 
3681 	if (!cmd->command)
3682 		return;
3683 
3684 	dmam_free_coherent(&dd->pdev->dev, CMD_DMA_ALLOC_SZ,
3685 				cmd->command, cmd->command_dma);
3686 }
3687 
3688 static int mtip_init_cmd(struct blk_mq_tag_set *set, struct request *rq,
3689 			 unsigned int hctx_idx, unsigned int numa_node)
3690 {
3691 	struct driver_data *dd = set->driver_data;
3692 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3693 
3694 	cmd->command = dmam_alloc_coherent(&dd->pdev->dev, CMD_DMA_ALLOC_SZ,
3695 			&cmd->command_dma, GFP_KERNEL);
3696 	if (!cmd->command)
3697 		return -ENOMEM;
3698 
3699 	memset(cmd->command, 0, CMD_DMA_ALLOC_SZ);
3700 
3701 	sg_init_table(cmd->sg, MTIP_MAX_SG);
3702 	return 0;
3703 }
3704 
3705 static enum blk_eh_timer_return mtip_cmd_timeout(struct request *req,
3706 								bool reserved)
3707 {
3708 	struct driver_data *dd = req->q->queuedata;
3709 
3710 	if (reserved) {
3711 		struct mtip_cmd *cmd = blk_mq_rq_to_pdu(req);
3712 
3713 		cmd->status = BLK_STS_TIMEOUT;
3714 		blk_mq_complete_request(req);
3715 		return BLK_EH_DONE;
3716 	}
3717 
3718 	if (test_bit(req->tag, dd->port->cmds_to_issue))
3719 		goto exit_handler;
3720 
3721 	if (test_and_set_bit(MTIP_PF_TO_ACTIVE_BIT, &dd->port->flags))
3722 		goto exit_handler;
3723 
3724 	wake_up_interruptible(&dd->port->svc_wait);
3725 exit_handler:
3726 	return BLK_EH_RESET_TIMER;
3727 }
3728 
3729 static const struct blk_mq_ops mtip_mq_ops = {
3730 	.queue_rq	= mtip_queue_rq,
3731 	.init_request	= mtip_init_cmd,
3732 	.exit_request	= mtip_free_cmd,
3733 	.complete	= mtip_softirq_done_fn,
3734 	.timeout        = mtip_cmd_timeout,
3735 };
3736 
3737 /*
3738  * Block layer initialization function.
3739  *
3740  * This function is called once by the PCI layer for each P320
3741  * device that is connected to the system.
3742  *
3743  * @dd Pointer to the driver data structure.
3744  *
3745  * return value
3746  *	0 on success else an error code.
3747  */
3748 static int mtip_block_initialize(struct driver_data *dd)
3749 {
3750 	int rv = 0, wait_for_rebuild = 0;
3751 	sector_t capacity;
3752 	unsigned int index = 0;
3753 	struct kobject *kobj;
3754 
3755 	if (dd->disk)
3756 		goto skip_create_disk; /* hw init done, before rebuild */
3757 
3758 	if (mtip_hw_init(dd)) {
3759 		rv = -EINVAL;
3760 		goto protocol_init_error;
3761 	}
3762 
3763 	dd->disk = alloc_disk_node(MTIP_MAX_MINORS, dd->numa_node);
3764 	if (dd->disk  == NULL) {
3765 		dev_err(&dd->pdev->dev,
3766 			"Unable to allocate gendisk structure\n");
3767 		rv = -EINVAL;
3768 		goto alloc_disk_error;
3769 	}
3770 
3771 	/* Generate the disk name, implemented same as in sd.c */
3772 	do {
3773 		if (!ida_pre_get(&rssd_index_ida, GFP_KERNEL)) {
3774 			rv = -ENOMEM;
3775 			goto ida_get_error;
3776 		}
3777 
3778 		spin_lock(&rssd_index_lock);
3779 		rv = ida_get_new(&rssd_index_ida, &index);
3780 		spin_unlock(&rssd_index_lock);
3781 	} while (rv == -EAGAIN);
3782 
3783 	if (rv)
3784 		goto ida_get_error;
3785 
3786 	rv = rssd_disk_name_format("rssd",
3787 				index,
3788 				dd->disk->disk_name,
3789 				DISK_NAME_LEN);
3790 	if (rv)
3791 		goto disk_index_error;
3792 
3793 	dd->disk->major		= dd->major;
3794 	dd->disk->first_minor	= index * MTIP_MAX_MINORS;
3795 	dd->disk->minors 	= MTIP_MAX_MINORS;
3796 	dd->disk->fops		= &mtip_block_ops;
3797 	dd->disk->private_data	= dd;
3798 	dd->index		= index;
3799 
3800 	mtip_hw_debugfs_init(dd);
3801 
3802 	memset(&dd->tags, 0, sizeof(dd->tags));
3803 	dd->tags.ops = &mtip_mq_ops;
3804 	dd->tags.nr_hw_queues = 1;
3805 	dd->tags.queue_depth = MTIP_MAX_COMMAND_SLOTS;
3806 	dd->tags.reserved_tags = 1;
3807 	dd->tags.cmd_size = sizeof(struct mtip_cmd);
3808 	dd->tags.numa_node = dd->numa_node;
3809 	dd->tags.flags = BLK_MQ_F_SHOULD_MERGE;
3810 	dd->tags.driver_data = dd;
3811 	dd->tags.timeout = MTIP_NCQ_CMD_TIMEOUT_MS;
3812 
3813 	rv = blk_mq_alloc_tag_set(&dd->tags);
3814 	if (rv) {
3815 		dev_err(&dd->pdev->dev,
3816 			"Unable to allocate request queue\n");
3817 		goto block_queue_alloc_tag_error;
3818 	}
3819 
3820 	/* Allocate the request queue. */
3821 	dd->queue = blk_mq_init_queue(&dd->tags);
3822 	if (IS_ERR(dd->queue)) {
3823 		dev_err(&dd->pdev->dev,
3824 			"Unable to allocate request queue\n");
3825 		rv = -ENOMEM;
3826 		goto block_queue_alloc_init_error;
3827 	}
3828 
3829 	dd->disk->queue		= dd->queue;
3830 	dd->queue->queuedata	= dd;
3831 
3832 skip_create_disk:
3833 	/* Initialize the protocol layer. */
3834 	wait_for_rebuild = mtip_hw_get_identify(dd);
3835 	if (wait_for_rebuild < 0) {
3836 		dev_err(&dd->pdev->dev,
3837 			"Protocol layer initialization failed\n");
3838 		rv = -EINVAL;
3839 		goto init_hw_cmds_error;
3840 	}
3841 
3842 	/*
3843 	 * if rebuild pending, start the service thread, and delay the block
3844 	 * queue creation and device_add_disk()
3845 	 */
3846 	if (wait_for_rebuild == MTIP_FTL_REBUILD_MAGIC)
3847 		goto start_service_thread;
3848 
3849 	/* Set device limits. */
3850 	blk_queue_flag_set(QUEUE_FLAG_NONROT, dd->queue);
3851 	blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, dd->queue);
3852 	blk_queue_max_segments(dd->queue, MTIP_MAX_SG);
3853 	blk_queue_physical_block_size(dd->queue, 4096);
3854 	blk_queue_max_hw_sectors(dd->queue, 0xffff);
3855 	blk_queue_max_segment_size(dd->queue, 0x400000);
3856 	blk_queue_io_min(dd->queue, 4096);
3857 
3858 	/* Signal trim support */
3859 	if (dd->trim_supp == true) {
3860 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, dd->queue);
3861 		dd->queue->limits.discard_granularity = 4096;
3862 		blk_queue_max_discard_sectors(dd->queue,
3863 			MTIP_MAX_TRIM_ENTRY_LEN * MTIP_MAX_TRIM_ENTRIES);
3864 	}
3865 
3866 	/* Set the capacity of the device in 512 byte sectors. */
3867 	if (!(mtip_hw_get_capacity(dd, &capacity))) {
3868 		dev_warn(&dd->pdev->dev,
3869 			"Could not read drive capacity\n");
3870 		rv = -EIO;
3871 		goto read_capacity_error;
3872 	}
3873 	set_capacity(dd->disk, capacity);
3874 
3875 	/* Enable the block device and add it to /dev */
3876 	device_add_disk(&dd->pdev->dev, dd->disk);
3877 
3878 	dd->bdev = bdget_disk(dd->disk, 0);
3879 	/*
3880 	 * Now that the disk is active, initialize any sysfs attributes
3881 	 * managed by the protocol layer.
3882 	 */
3883 	kobj = kobject_get(&disk_to_dev(dd->disk)->kobj);
3884 	if (kobj) {
3885 		mtip_hw_sysfs_init(dd, kobj);
3886 		kobject_put(kobj);
3887 	}
3888 
3889 	if (dd->mtip_svc_handler) {
3890 		set_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag);
3891 		return rv; /* service thread created for handling rebuild */
3892 	}
3893 
3894 start_service_thread:
3895 	dd->mtip_svc_handler = kthread_create_on_node(mtip_service_thread,
3896 						dd, dd->numa_node,
3897 						"mtip_svc_thd_%02d", index);
3898 
3899 	if (IS_ERR(dd->mtip_svc_handler)) {
3900 		dev_err(&dd->pdev->dev, "service thread failed to start\n");
3901 		dd->mtip_svc_handler = NULL;
3902 		rv = -EFAULT;
3903 		goto kthread_run_error;
3904 	}
3905 	wake_up_process(dd->mtip_svc_handler);
3906 	if (wait_for_rebuild == MTIP_FTL_REBUILD_MAGIC)
3907 		rv = wait_for_rebuild;
3908 
3909 	return rv;
3910 
3911 kthread_run_error:
3912 	bdput(dd->bdev);
3913 	dd->bdev = NULL;
3914 
3915 	/* Delete our gendisk. This also removes the device from /dev */
3916 	del_gendisk(dd->disk);
3917 
3918 read_capacity_error:
3919 init_hw_cmds_error:
3920 	blk_cleanup_queue(dd->queue);
3921 block_queue_alloc_init_error:
3922 	blk_mq_free_tag_set(&dd->tags);
3923 block_queue_alloc_tag_error:
3924 	mtip_hw_debugfs_exit(dd);
3925 disk_index_error:
3926 	spin_lock(&rssd_index_lock);
3927 	ida_remove(&rssd_index_ida, index);
3928 	spin_unlock(&rssd_index_lock);
3929 
3930 ida_get_error:
3931 	put_disk(dd->disk);
3932 
3933 alloc_disk_error:
3934 	mtip_hw_exit(dd); /* De-initialize the protocol layer. */
3935 
3936 protocol_init_error:
3937 	return rv;
3938 }
3939 
3940 static void mtip_no_dev_cleanup(struct request *rq, void *data, bool reserv)
3941 {
3942 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3943 
3944 	cmd->status = BLK_STS_IOERR;
3945 	blk_mq_complete_request(rq);
3946 }
3947 
3948 /*
3949  * Block layer deinitialization function.
3950  *
3951  * Called by the PCI layer as each P320 device is removed.
3952  *
3953  * @dd Pointer to the driver data structure.
3954  *
3955  * return value
3956  *	0
3957  */
3958 static int mtip_block_remove(struct driver_data *dd)
3959 {
3960 	struct kobject *kobj;
3961 
3962 	mtip_hw_debugfs_exit(dd);
3963 
3964 	if (dd->mtip_svc_handler) {
3965 		set_bit(MTIP_PF_SVC_THD_STOP_BIT, &dd->port->flags);
3966 		wake_up_interruptible(&dd->port->svc_wait);
3967 		kthread_stop(dd->mtip_svc_handler);
3968 	}
3969 
3970 	/* Clean up the sysfs attributes, if created */
3971 	if (test_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag)) {
3972 		kobj = kobject_get(&disk_to_dev(dd->disk)->kobj);
3973 		if (kobj) {
3974 			mtip_hw_sysfs_exit(dd, kobj);
3975 			kobject_put(kobj);
3976 		}
3977 	}
3978 
3979 	if (!dd->sr) {
3980 		/*
3981 		 * Explicitly wait here for IOs to quiesce,
3982 		 * as mtip_standby_drive usually won't wait for IOs.
3983 		 */
3984 		if (!mtip_quiesce_io(dd->port, MTIP_QUIESCE_IO_TIMEOUT_MS))
3985 			mtip_standby_drive(dd);
3986 	}
3987 	else
3988 		dev_info(&dd->pdev->dev, "device %s surprise removal\n",
3989 						dd->disk->disk_name);
3990 
3991 	blk_freeze_queue_start(dd->queue);
3992 	blk_mq_quiesce_queue(dd->queue);
3993 	blk_mq_tagset_busy_iter(&dd->tags, mtip_no_dev_cleanup, dd);
3994 	blk_mq_unquiesce_queue(dd->queue);
3995 
3996 	/*
3997 	 * Delete our gendisk structure. This also removes the device
3998 	 * from /dev
3999 	 */
4000 	if (dd->bdev) {
4001 		bdput(dd->bdev);
4002 		dd->bdev = NULL;
4003 	}
4004 	if (dd->disk) {
4005 		if (test_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag))
4006 			del_gendisk(dd->disk);
4007 		if (dd->disk->queue) {
4008 			blk_cleanup_queue(dd->queue);
4009 			blk_mq_free_tag_set(&dd->tags);
4010 			dd->queue = NULL;
4011 		}
4012 		put_disk(dd->disk);
4013 	}
4014 	dd->disk  = NULL;
4015 
4016 	spin_lock(&rssd_index_lock);
4017 	ida_remove(&rssd_index_ida, dd->index);
4018 	spin_unlock(&rssd_index_lock);
4019 
4020 	/* De-initialize the protocol layer. */
4021 	mtip_hw_exit(dd);
4022 
4023 	return 0;
4024 }
4025 
4026 /*
4027  * Function called by the PCI layer when just before the
4028  * machine shuts down.
4029  *
4030  * If a protocol layer shutdown function is present it will be called
4031  * by this function.
4032  *
4033  * @dd Pointer to the driver data structure.
4034  *
4035  * return value
4036  *	0
4037  */
4038 static int mtip_block_shutdown(struct driver_data *dd)
4039 {
4040 	mtip_hw_shutdown(dd);
4041 
4042 	/* Delete our gendisk structure, and cleanup the blk queue. */
4043 	if (dd->disk) {
4044 		dev_info(&dd->pdev->dev,
4045 			"Shutting down %s ...\n", dd->disk->disk_name);
4046 
4047 		if (test_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag))
4048 			del_gendisk(dd->disk);
4049 		if (dd->disk->queue) {
4050 			blk_cleanup_queue(dd->queue);
4051 			blk_mq_free_tag_set(&dd->tags);
4052 		}
4053 		put_disk(dd->disk);
4054 		dd->disk  = NULL;
4055 		dd->queue = NULL;
4056 	}
4057 
4058 	spin_lock(&rssd_index_lock);
4059 	ida_remove(&rssd_index_ida, dd->index);
4060 	spin_unlock(&rssd_index_lock);
4061 	return 0;
4062 }
4063 
4064 static int mtip_block_suspend(struct driver_data *dd)
4065 {
4066 	dev_info(&dd->pdev->dev,
4067 		"Suspending %s ...\n", dd->disk->disk_name);
4068 	mtip_hw_suspend(dd);
4069 	return 0;
4070 }
4071 
4072 static int mtip_block_resume(struct driver_data *dd)
4073 {
4074 	dev_info(&dd->pdev->dev, "Resuming %s ...\n",
4075 		dd->disk->disk_name);
4076 	mtip_hw_resume(dd);
4077 	return 0;
4078 }
4079 
4080 static void drop_cpu(int cpu)
4081 {
4082 	cpu_use[cpu]--;
4083 }
4084 
4085 static int get_least_used_cpu_on_node(int node)
4086 {
4087 	int cpu, least_used_cpu, least_cnt;
4088 	const struct cpumask *node_mask;
4089 
4090 	node_mask = cpumask_of_node(node);
4091 	least_used_cpu = cpumask_first(node_mask);
4092 	least_cnt = cpu_use[least_used_cpu];
4093 	cpu = least_used_cpu;
4094 
4095 	for_each_cpu(cpu, node_mask) {
4096 		if (cpu_use[cpu] < least_cnt) {
4097 			least_used_cpu = cpu;
4098 			least_cnt = cpu_use[cpu];
4099 		}
4100 	}
4101 	cpu_use[least_used_cpu]++;
4102 	return least_used_cpu;
4103 }
4104 
4105 /* Helper for selecting a node in round robin mode */
4106 static inline int mtip_get_next_rr_node(void)
4107 {
4108 	static int next_node = -1;
4109 
4110 	if (next_node == -1) {
4111 		next_node = first_online_node;
4112 		return next_node;
4113 	}
4114 
4115 	next_node = next_online_node(next_node);
4116 	if (next_node == MAX_NUMNODES)
4117 		next_node = first_online_node;
4118 	return next_node;
4119 }
4120 
4121 static DEFINE_HANDLER(0);
4122 static DEFINE_HANDLER(1);
4123 static DEFINE_HANDLER(2);
4124 static DEFINE_HANDLER(3);
4125 static DEFINE_HANDLER(4);
4126 static DEFINE_HANDLER(5);
4127 static DEFINE_HANDLER(6);
4128 static DEFINE_HANDLER(7);
4129 
4130 static void mtip_disable_link_opts(struct driver_data *dd, struct pci_dev *pdev)
4131 {
4132 	int pos;
4133 	unsigned short pcie_dev_ctrl;
4134 
4135 	pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4136 	if (pos) {
4137 		pci_read_config_word(pdev,
4138 			pos + PCI_EXP_DEVCTL,
4139 			&pcie_dev_ctrl);
4140 		if (pcie_dev_ctrl & (1 << 11) ||
4141 		    pcie_dev_ctrl & (1 << 4)) {
4142 			dev_info(&dd->pdev->dev,
4143 				"Disabling ERO/No-Snoop on bridge device %04x:%04x\n",
4144 					pdev->vendor, pdev->device);
4145 			pcie_dev_ctrl &= ~(PCI_EXP_DEVCTL_NOSNOOP_EN |
4146 						PCI_EXP_DEVCTL_RELAX_EN);
4147 			pci_write_config_word(pdev,
4148 				pos + PCI_EXP_DEVCTL,
4149 				pcie_dev_ctrl);
4150 		}
4151 	}
4152 }
4153 
4154 static void mtip_fix_ero_nosnoop(struct driver_data *dd, struct pci_dev *pdev)
4155 {
4156 	/*
4157 	 * This workaround is specific to AMD/ATI chipset with a PCI upstream
4158 	 * device with device id 0x5aXX
4159 	 */
4160 	if (pdev->bus && pdev->bus->self) {
4161 		if (pdev->bus->self->vendor == PCI_VENDOR_ID_ATI &&
4162 		    ((pdev->bus->self->device & 0xff00) == 0x5a00)) {
4163 			mtip_disable_link_opts(dd, pdev->bus->self);
4164 		} else {
4165 			/* Check further up the topology */
4166 			struct pci_dev *parent_dev = pdev->bus->self;
4167 			if (parent_dev->bus &&
4168 				parent_dev->bus->parent &&
4169 				parent_dev->bus->parent->self &&
4170 				parent_dev->bus->parent->self->vendor ==
4171 					 PCI_VENDOR_ID_ATI &&
4172 				(parent_dev->bus->parent->self->device &
4173 					0xff00) == 0x5a00) {
4174 				mtip_disable_link_opts(dd,
4175 					parent_dev->bus->parent->self);
4176 			}
4177 		}
4178 	}
4179 }
4180 
4181 /*
4182  * Called for each supported PCI device detected.
4183  *
4184  * This function allocates the private data structure, enables the
4185  * PCI device and then calls the block layer initialization function.
4186  *
4187  * return value
4188  *	0 on success else an error code.
4189  */
4190 static int mtip_pci_probe(struct pci_dev *pdev,
4191 			const struct pci_device_id *ent)
4192 {
4193 	int rv = 0;
4194 	struct driver_data *dd = NULL;
4195 	char cpu_list[256];
4196 	const struct cpumask *node_mask;
4197 	int cpu, i = 0, j = 0;
4198 	int my_node = NUMA_NO_NODE;
4199 	unsigned long flags;
4200 
4201 	/* Allocate memory for this devices private data. */
4202 	my_node = pcibus_to_node(pdev->bus);
4203 	if (my_node != NUMA_NO_NODE) {
4204 		if (!node_online(my_node))
4205 			my_node = mtip_get_next_rr_node();
4206 	} else {
4207 		dev_info(&pdev->dev, "Kernel not reporting proximity, choosing a node\n");
4208 		my_node = mtip_get_next_rr_node();
4209 	}
4210 	dev_info(&pdev->dev, "NUMA node %d (closest: %d,%d, probe on %d:%d)\n",
4211 		my_node, pcibus_to_node(pdev->bus), dev_to_node(&pdev->dev),
4212 		cpu_to_node(raw_smp_processor_id()), raw_smp_processor_id());
4213 
4214 	dd = kzalloc_node(sizeof(struct driver_data), GFP_KERNEL, my_node);
4215 	if (dd == NULL) {
4216 		dev_err(&pdev->dev,
4217 			"Unable to allocate memory for driver data\n");
4218 		return -ENOMEM;
4219 	}
4220 
4221 	/* Attach the private data to this PCI device.  */
4222 	pci_set_drvdata(pdev, dd);
4223 
4224 	rv = pcim_enable_device(pdev);
4225 	if (rv < 0) {
4226 		dev_err(&pdev->dev, "Unable to enable device\n");
4227 		goto iomap_err;
4228 	}
4229 
4230 	/* Map BAR5 to memory. */
4231 	rv = pcim_iomap_regions(pdev, 1 << MTIP_ABAR, MTIP_DRV_NAME);
4232 	if (rv < 0) {
4233 		dev_err(&pdev->dev, "Unable to map regions\n");
4234 		goto iomap_err;
4235 	}
4236 
4237 	if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
4238 		rv = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4239 
4240 		if (rv) {
4241 			rv = pci_set_consistent_dma_mask(pdev,
4242 						DMA_BIT_MASK(32));
4243 			if (rv) {
4244 				dev_warn(&pdev->dev,
4245 					"64-bit DMA enable failed\n");
4246 				goto setmask_err;
4247 			}
4248 		}
4249 	}
4250 
4251 	/* Copy the info we may need later into the private data structure. */
4252 	dd->major	= mtip_major;
4253 	dd->instance	= instance;
4254 	dd->pdev	= pdev;
4255 	dd->numa_node	= my_node;
4256 
4257 	INIT_LIST_HEAD(&dd->online_list);
4258 	INIT_LIST_HEAD(&dd->remove_list);
4259 
4260 	memset(dd->workq_name, 0, 32);
4261 	snprintf(dd->workq_name, 31, "mtipq%d", dd->instance);
4262 
4263 	dd->isr_workq = create_workqueue(dd->workq_name);
4264 	if (!dd->isr_workq) {
4265 		dev_warn(&pdev->dev, "Can't create wq %d\n", dd->instance);
4266 		rv = -ENOMEM;
4267 		goto setmask_err;
4268 	}
4269 
4270 	memset(cpu_list, 0, sizeof(cpu_list));
4271 
4272 	node_mask = cpumask_of_node(dd->numa_node);
4273 	if (!cpumask_empty(node_mask)) {
4274 		for_each_cpu(cpu, node_mask)
4275 		{
4276 			snprintf(&cpu_list[j], 256 - j, "%d ", cpu);
4277 			j = strlen(cpu_list);
4278 		}
4279 
4280 		dev_info(&pdev->dev, "Node %d on package %d has %d cpu(s): %s\n",
4281 			dd->numa_node,
4282 			topology_physical_package_id(cpumask_first(node_mask)),
4283 			nr_cpus_node(dd->numa_node),
4284 			cpu_list);
4285 	} else
4286 		dev_dbg(&pdev->dev, "mtip32xx: node_mask empty\n");
4287 
4288 	dd->isr_binding = get_least_used_cpu_on_node(dd->numa_node);
4289 	dev_info(&pdev->dev, "Initial IRQ binding node:cpu %d:%d\n",
4290 		cpu_to_node(dd->isr_binding), dd->isr_binding);
4291 
4292 	/* first worker context always runs in ISR */
4293 	dd->work[0].cpu_binding = dd->isr_binding;
4294 	dd->work[1].cpu_binding = get_least_used_cpu_on_node(dd->numa_node);
4295 	dd->work[2].cpu_binding = get_least_used_cpu_on_node(dd->numa_node);
4296 	dd->work[3].cpu_binding = dd->work[0].cpu_binding;
4297 	dd->work[4].cpu_binding = dd->work[1].cpu_binding;
4298 	dd->work[5].cpu_binding = dd->work[2].cpu_binding;
4299 	dd->work[6].cpu_binding = dd->work[2].cpu_binding;
4300 	dd->work[7].cpu_binding = dd->work[1].cpu_binding;
4301 
4302 	/* Log the bindings */
4303 	for_each_present_cpu(cpu) {
4304 		memset(cpu_list, 0, sizeof(cpu_list));
4305 		for (i = 0, j = 0; i < MTIP_MAX_SLOT_GROUPS; i++) {
4306 			if (dd->work[i].cpu_binding == cpu) {
4307 				snprintf(&cpu_list[j], 256 - j, "%d ", i);
4308 				j = strlen(cpu_list);
4309 			}
4310 		}
4311 		if (j)
4312 			dev_info(&pdev->dev, "CPU %d: WQs %s\n", cpu, cpu_list);
4313 	}
4314 
4315 	INIT_WORK(&dd->work[0].work, mtip_workq_sdbf0);
4316 	INIT_WORK(&dd->work[1].work, mtip_workq_sdbf1);
4317 	INIT_WORK(&dd->work[2].work, mtip_workq_sdbf2);
4318 	INIT_WORK(&dd->work[3].work, mtip_workq_sdbf3);
4319 	INIT_WORK(&dd->work[4].work, mtip_workq_sdbf4);
4320 	INIT_WORK(&dd->work[5].work, mtip_workq_sdbf5);
4321 	INIT_WORK(&dd->work[6].work, mtip_workq_sdbf6);
4322 	INIT_WORK(&dd->work[7].work, mtip_workq_sdbf7);
4323 
4324 	pci_set_master(pdev);
4325 	rv = pci_enable_msi(pdev);
4326 	if (rv) {
4327 		dev_warn(&pdev->dev,
4328 			"Unable to enable MSI interrupt.\n");
4329 		goto msi_initialize_err;
4330 	}
4331 
4332 	mtip_fix_ero_nosnoop(dd, pdev);
4333 
4334 	/* Initialize the block layer. */
4335 	rv = mtip_block_initialize(dd);
4336 	if (rv < 0) {
4337 		dev_err(&pdev->dev,
4338 			"Unable to initialize block layer\n");
4339 		goto block_initialize_err;
4340 	}
4341 
4342 	/*
4343 	 * Increment the instance count so that each device has a unique
4344 	 * instance number.
4345 	 */
4346 	instance++;
4347 	if (rv != MTIP_FTL_REBUILD_MAGIC)
4348 		set_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag);
4349 	else
4350 		rv = 0; /* device in rebuild state, return 0 from probe */
4351 
4352 	/* Add to online list even if in ftl rebuild */
4353 	spin_lock_irqsave(&dev_lock, flags);
4354 	list_add(&dd->online_list, &online_list);
4355 	spin_unlock_irqrestore(&dev_lock, flags);
4356 
4357 	goto done;
4358 
4359 block_initialize_err:
4360 	pci_disable_msi(pdev);
4361 
4362 msi_initialize_err:
4363 	if (dd->isr_workq) {
4364 		flush_workqueue(dd->isr_workq);
4365 		destroy_workqueue(dd->isr_workq);
4366 		drop_cpu(dd->work[0].cpu_binding);
4367 		drop_cpu(dd->work[1].cpu_binding);
4368 		drop_cpu(dd->work[2].cpu_binding);
4369 	}
4370 setmask_err:
4371 	pcim_iounmap_regions(pdev, 1 << MTIP_ABAR);
4372 
4373 iomap_err:
4374 	kfree(dd);
4375 	pci_set_drvdata(pdev, NULL);
4376 	return rv;
4377 done:
4378 	return rv;
4379 }
4380 
4381 /*
4382  * Called for each probed device when the device is removed or the
4383  * driver is unloaded.
4384  *
4385  * return value
4386  *	None
4387  */
4388 static void mtip_pci_remove(struct pci_dev *pdev)
4389 {
4390 	struct driver_data *dd = pci_get_drvdata(pdev);
4391 	unsigned long flags, to;
4392 
4393 	set_bit(MTIP_DDF_REMOVAL_BIT, &dd->dd_flag);
4394 
4395 	spin_lock_irqsave(&dev_lock, flags);
4396 	list_del_init(&dd->online_list);
4397 	list_add(&dd->remove_list, &removing_list);
4398 	spin_unlock_irqrestore(&dev_lock, flags);
4399 
4400 	mtip_check_surprise_removal(pdev);
4401 	synchronize_irq(dd->pdev->irq);
4402 
4403 	/* Spin until workers are done */
4404 	to = jiffies + msecs_to_jiffies(4000);
4405 	do {
4406 		msleep(20);
4407 	} while (atomic_read(&dd->irq_workers_active) != 0 &&
4408 		time_before(jiffies, to));
4409 
4410 	if (!dd->sr)
4411 		fsync_bdev(dd->bdev);
4412 
4413 	if (atomic_read(&dd->irq_workers_active) != 0) {
4414 		dev_warn(&dd->pdev->dev,
4415 			"Completion workers still active!\n");
4416 	}
4417 
4418 	blk_set_queue_dying(dd->queue);
4419 	set_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag);
4420 
4421 	/* Clean up the block layer. */
4422 	mtip_block_remove(dd);
4423 
4424 	if (dd->isr_workq) {
4425 		flush_workqueue(dd->isr_workq);
4426 		destroy_workqueue(dd->isr_workq);
4427 		drop_cpu(dd->work[0].cpu_binding);
4428 		drop_cpu(dd->work[1].cpu_binding);
4429 		drop_cpu(dd->work[2].cpu_binding);
4430 	}
4431 
4432 	pci_disable_msi(pdev);
4433 
4434 	spin_lock_irqsave(&dev_lock, flags);
4435 	list_del_init(&dd->remove_list);
4436 	spin_unlock_irqrestore(&dev_lock, flags);
4437 
4438 	kfree(dd);
4439 
4440 	pcim_iounmap_regions(pdev, 1 << MTIP_ABAR);
4441 	pci_set_drvdata(pdev, NULL);
4442 }
4443 
4444 /*
4445  * Called for each probed device when the device is suspended.
4446  *
4447  * return value
4448  *	0  Success
4449  *	<0 Error
4450  */
4451 static int mtip_pci_suspend(struct pci_dev *pdev, pm_message_t mesg)
4452 {
4453 	int rv = 0;
4454 	struct driver_data *dd = pci_get_drvdata(pdev);
4455 
4456 	if (!dd) {
4457 		dev_err(&pdev->dev,
4458 			"Driver private datastructure is NULL\n");
4459 		return -EFAULT;
4460 	}
4461 
4462 	set_bit(MTIP_DDF_RESUME_BIT, &dd->dd_flag);
4463 
4464 	/* Disable ports & interrupts then send standby immediate */
4465 	rv = mtip_block_suspend(dd);
4466 	if (rv < 0) {
4467 		dev_err(&pdev->dev,
4468 			"Failed to suspend controller\n");
4469 		return rv;
4470 	}
4471 
4472 	/*
4473 	 * Save the pci config space to pdev structure &
4474 	 * disable the device
4475 	 */
4476 	pci_save_state(pdev);
4477 	pci_disable_device(pdev);
4478 
4479 	/* Move to Low power state*/
4480 	pci_set_power_state(pdev, PCI_D3hot);
4481 
4482 	return rv;
4483 }
4484 
4485 /*
4486  * Called for each probed device when the device is resumed.
4487  *
4488  * return value
4489  *      0  Success
4490  *      <0 Error
4491  */
4492 static int mtip_pci_resume(struct pci_dev *pdev)
4493 {
4494 	int rv = 0;
4495 	struct driver_data *dd;
4496 
4497 	dd = pci_get_drvdata(pdev);
4498 	if (!dd) {
4499 		dev_err(&pdev->dev,
4500 			"Driver private datastructure is NULL\n");
4501 		return -EFAULT;
4502 	}
4503 
4504 	/* Move the device to active State */
4505 	pci_set_power_state(pdev, PCI_D0);
4506 
4507 	/* Restore PCI configuration space */
4508 	pci_restore_state(pdev);
4509 
4510 	/* Enable the PCI device*/
4511 	rv = pcim_enable_device(pdev);
4512 	if (rv < 0) {
4513 		dev_err(&pdev->dev,
4514 			"Failed to enable card during resume\n");
4515 		goto err;
4516 	}
4517 	pci_set_master(pdev);
4518 
4519 	/*
4520 	 * Calls hbaReset, initPort, & startPort function
4521 	 * then enables interrupts
4522 	 */
4523 	rv = mtip_block_resume(dd);
4524 	if (rv < 0)
4525 		dev_err(&pdev->dev, "Unable to resume\n");
4526 
4527 err:
4528 	clear_bit(MTIP_DDF_RESUME_BIT, &dd->dd_flag);
4529 
4530 	return rv;
4531 }
4532 
4533 /*
4534  * Shutdown routine
4535  *
4536  * return value
4537  *      None
4538  */
4539 static void mtip_pci_shutdown(struct pci_dev *pdev)
4540 {
4541 	struct driver_data *dd = pci_get_drvdata(pdev);
4542 	if (dd)
4543 		mtip_block_shutdown(dd);
4544 }
4545 
4546 /* Table of device ids supported by this driver. */
4547 static const struct pci_device_id mtip_pci_tbl[] = {
4548 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P320H_DEVICE_ID) },
4549 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P320M_DEVICE_ID) },
4550 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P320S_DEVICE_ID) },
4551 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P325M_DEVICE_ID) },
4552 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P420H_DEVICE_ID) },
4553 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P420M_DEVICE_ID) },
4554 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P425M_DEVICE_ID) },
4555 	{ 0 }
4556 };
4557 
4558 /* Structure that describes the PCI driver functions. */
4559 static struct pci_driver mtip_pci_driver = {
4560 	.name			= MTIP_DRV_NAME,
4561 	.id_table		= mtip_pci_tbl,
4562 	.probe			= mtip_pci_probe,
4563 	.remove			= mtip_pci_remove,
4564 	.suspend		= mtip_pci_suspend,
4565 	.resume			= mtip_pci_resume,
4566 	.shutdown		= mtip_pci_shutdown,
4567 };
4568 
4569 MODULE_DEVICE_TABLE(pci, mtip_pci_tbl);
4570 
4571 /*
4572  * Module initialization function.
4573  *
4574  * Called once when the module is loaded. This function allocates a major
4575  * block device number to the Cyclone devices and registers the PCI layer
4576  * of the driver.
4577  *
4578  * Return value
4579  *      0 on success else error code.
4580  */
4581 static int __init mtip_init(void)
4582 {
4583 	int error;
4584 
4585 	pr_info(MTIP_DRV_NAME " Version " MTIP_DRV_VERSION "\n");
4586 
4587 	spin_lock_init(&dev_lock);
4588 
4589 	INIT_LIST_HEAD(&online_list);
4590 	INIT_LIST_HEAD(&removing_list);
4591 
4592 	/* Allocate a major block device number to use with this driver. */
4593 	error = register_blkdev(0, MTIP_DRV_NAME);
4594 	if (error <= 0) {
4595 		pr_err("Unable to register block device (%d)\n",
4596 		error);
4597 		return -EBUSY;
4598 	}
4599 	mtip_major = error;
4600 
4601 	dfs_parent = debugfs_create_dir("rssd", NULL);
4602 	if (IS_ERR_OR_NULL(dfs_parent)) {
4603 		pr_warn("Error creating debugfs parent\n");
4604 		dfs_parent = NULL;
4605 	}
4606 	if (dfs_parent) {
4607 		dfs_device_status = debugfs_create_file("device_status",
4608 					0444, dfs_parent, NULL,
4609 					&mtip_device_status_fops);
4610 		if (IS_ERR_OR_NULL(dfs_device_status)) {
4611 			pr_err("Error creating device_status node\n");
4612 			dfs_device_status = NULL;
4613 		}
4614 	}
4615 
4616 	/* Register our PCI operations. */
4617 	error = pci_register_driver(&mtip_pci_driver);
4618 	if (error) {
4619 		debugfs_remove(dfs_parent);
4620 		unregister_blkdev(mtip_major, MTIP_DRV_NAME);
4621 	}
4622 
4623 	return error;
4624 }
4625 
4626 /*
4627  * Module de-initialization function.
4628  *
4629  * Called once when the module is unloaded. This function deallocates
4630  * the major block device number allocated by mtip_init() and
4631  * unregisters the PCI layer of the driver.
4632  *
4633  * Return value
4634  *      none
4635  */
4636 static void __exit mtip_exit(void)
4637 {
4638 	/* Release the allocated major block device number. */
4639 	unregister_blkdev(mtip_major, MTIP_DRV_NAME);
4640 
4641 	/* Unregister the PCI driver. */
4642 	pci_unregister_driver(&mtip_pci_driver);
4643 
4644 	debugfs_remove_recursive(dfs_parent);
4645 }
4646 
4647 MODULE_AUTHOR("Micron Technology, Inc");
4648 MODULE_DESCRIPTION("Micron RealSSD PCIe Block Driver");
4649 MODULE_LICENSE("GPL");
4650 MODULE_VERSION(MTIP_DRV_VERSION);
4651 
4652 module_init(mtip_init);
4653 module_exit(mtip_exit);
4654