1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Driver for the Micron P320 SSD
4  *   Copyright (C) 2011 Micron Technology, Inc.
5  *
6  * Portions of this code were derived from works subjected to the
7  * following copyright:
8  *    Copyright (C) 2009 Integrated Device Technology, Inc.
9  */
10 
11 #include <linux/pci.h>
12 #include <linux/interrupt.h>
13 #include <linux/ata.h>
14 #include <linux/delay.h>
15 #include <linux/hdreg.h>
16 #include <linux/uaccess.h>
17 #include <linux/random.h>
18 #include <linux/smp.h>
19 #include <linux/compat.h>
20 #include <linux/fs.h>
21 #include <linux/module.h>
22 #include <linux/genhd.h>
23 #include <linux/blkdev.h>
24 #include <linux/blk-mq.h>
25 #include <linux/bio.h>
26 #include <linux/dma-mapping.h>
27 #include <linux/idr.h>
28 #include <linux/kthread.h>
29 #include <../drivers/ata/ahci.h>
30 #include <linux/export.h>
31 #include <linux/debugfs.h>
32 #include <linux/prefetch.h>
33 #include <linux/numa.h>
34 #include "mtip32xx.h"
35 
36 #define HW_CMD_SLOT_SZ		(MTIP_MAX_COMMAND_SLOTS * 32)
37 
38 /* DMA region containing RX Fis, Identify, RLE10, and SMART buffers */
39 #define AHCI_RX_FIS_SZ          0x100
40 #define AHCI_RX_FIS_OFFSET      0x0
41 #define AHCI_IDFY_SZ            ATA_SECT_SIZE
42 #define AHCI_IDFY_OFFSET        0x400
43 #define AHCI_SECTBUF_SZ         ATA_SECT_SIZE
44 #define AHCI_SECTBUF_OFFSET     0x800
45 #define AHCI_SMARTBUF_SZ        ATA_SECT_SIZE
46 #define AHCI_SMARTBUF_OFFSET    0xC00
47 /* 0x100 + 0x200 + 0x200 + 0x200 is smaller than 4k but we pad it out */
48 #define BLOCK_DMA_ALLOC_SZ      4096
49 
50 /* DMA region containing command table (should be 8192 bytes) */
51 #define AHCI_CMD_SLOT_SZ        sizeof(struct mtip_cmd_hdr)
52 #define AHCI_CMD_TBL_SZ         (MTIP_MAX_COMMAND_SLOTS * AHCI_CMD_SLOT_SZ)
53 #define AHCI_CMD_TBL_OFFSET     0x0
54 
55 /* DMA region per command (contains header and SGL) */
56 #define AHCI_CMD_TBL_HDR_SZ     0x80
57 #define AHCI_CMD_TBL_HDR_OFFSET 0x0
58 #define AHCI_CMD_TBL_SGL_SZ     (MTIP_MAX_SG * sizeof(struct mtip_cmd_sg))
59 #define AHCI_CMD_TBL_SGL_OFFSET AHCI_CMD_TBL_HDR_SZ
60 #define CMD_DMA_ALLOC_SZ        (AHCI_CMD_TBL_SGL_SZ + AHCI_CMD_TBL_HDR_SZ)
61 
62 
63 #define HOST_CAP_NZDMA		(1 << 19)
64 #define HOST_HSORG		0xFC
65 #define HSORG_DISABLE_SLOTGRP_INTR (1<<24)
66 #define HSORG_DISABLE_SLOTGRP_PXIS (1<<16)
67 #define HSORG_HWREV		0xFF00
68 #define HSORG_STYLE		0x8
69 #define HSORG_SLOTGROUPS	0x7
70 
71 #define PORT_COMMAND_ISSUE	0x38
72 #define PORT_SDBV		0x7C
73 
74 #define PORT_OFFSET		0x100
75 #define PORT_MEM_SIZE		0x80
76 
77 #define PORT_IRQ_ERR \
78 	(PORT_IRQ_HBUS_ERR | PORT_IRQ_IF_ERR | PORT_IRQ_CONNECT | \
79 	 PORT_IRQ_PHYRDY | PORT_IRQ_UNK_FIS | PORT_IRQ_BAD_PMP | \
80 	 PORT_IRQ_TF_ERR | PORT_IRQ_HBUS_DATA_ERR | PORT_IRQ_IF_NONFATAL | \
81 	 PORT_IRQ_OVERFLOW)
82 #define PORT_IRQ_LEGACY \
83 	(PORT_IRQ_PIOS_FIS | PORT_IRQ_D2H_REG_FIS)
84 #define PORT_IRQ_HANDLED \
85 	(PORT_IRQ_SDB_FIS | PORT_IRQ_LEGACY | \
86 	 PORT_IRQ_TF_ERR | PORT_IRQ_IF_ERR | \
87 	 PORT_IRQ_CONNECT | PORT_IRQ_PHYRDY)
88 #define DEF_PORT_IRQ \
89 	(PORT_IRQ_ERR | PORT_IRQ_LEGACY | PORT_IRQ_SDB_FIS)
90 
91 /* product numbers */
92 #define MTIP_PRODUCT_UNKNOWN	0x00
93 #define MTIP_PRODUCT_ASICFPGA	0x11
94 
95 /* Device instance number, incremented each time a device is probed. */
96 static int instance;
97 
98 static LIST_HEAD(online_list);
99 static LIST_HEAD(removing_list);
100 static DEFINE_SPINLOCK(dev_lock);
101 
102 /*
103  * Global variable used to hold the major block device number
104  * allocated in mtip_init().
105  */
106 static int mtip_major;
107 static struct dentry *dfs_parent;
108 static struct dentry *dfs_device_status;
109 
110 static u32 cpu_use[NR_CPUS];
111 
112 static DEFINE_IDA(rssd_index_ida);
113 
114 static int mtip_block_initialize(struct driver_data *dd);
115 
116 #ifdef CONFIG_COMPAT
117 struct mtip_compat_ide_task_request_s {
118 	__u8		io_ports[8];
119 	__u8		hob_ports[8];
120 	ide_reg_valid_t	out_flags;
121 	ide_reg_valid_t	in_flags;
122 	int		data_phase;
123 	int		req_cmd;
124 	compat_ulong_t	out_size;
125 	compat_ulong_t	in_size;
126 };
127 #endif
128 
129 /*
130  * This function check_for_surprise_removal is called
131  * while card is removed from the system and it will
132  * read the vendor id from the configuration space
133  *
134  * @pdev Pointer to the pci_dev structure.
135  *
136  * return value
137  *	 true if device removed, else false
138  */
139 static bool mtip_check_surprise_removal(struct pci_dev *pdev)
140 {
141 	u16 vendor_id = 0;
142 	struct driver_data *dd = pci_get_drvdata(pdev);
143 
144 	if (dd->sr)
145 		return true;
146 
147        /* Read the vendorID from the configuration space */
148 	pci_read_config_word(pdev, 0x00, &vendor_id);
149 	if (vendor_id == 0xFFFF) {
150 		dd->sr = true;
151 		if (dd->queue)
152 			blk_queue_flag_set(QUEUE_FLAG_DEAD, dd->queue);
153 		else
154 			dev_warn(&dd->pdev->dev,
155 				"%s: dd->queue is NULL\n", __func__);
156 		return true; /* device removed */
157 	}
158 
159 	return false; /* device present */
160 }
161 
162 static struct mtip_cmd *mtip_cmd_from_tag(struct driver_data *dd,
163 					  unsigned int tag)
164 {
165 	struct blk_mq_hw_ctx *hctx = dd->queue->queue_hw_ctx[0];
166 
167 	return blk_mq_rq_to_pdu(blk_mq_tag_to_rq(hctx->tags, tag));
168 }
169 
170 /*
171  * Reset the HBA (without sleeping)
172  *
173  * @dd Pointer to the driver data structure.
174  *
175  * return value
176  *	0	The reset was successful.
177  *	-1	The HBA Reset bit did not clear.
178  */
179 static int mtip_hba_reset(struct driver_data *dd)
180 {
181 	unsigned long timeout;
182 
183 	/* Set the reset bit */
184 	writel(HOST_RESET, dd->mmio + HOST_CTL);
185 
186 	/* Flush */
187 	readl(dd->mmio + HOST_CTL);
188 
189 	/*
190 	 * Spin for up to 10 seconds waiting for reset acknowledgement. Spec
191 	 * is 1 sec but in LUN failure conditions, up to 10 secs are required
192 	 */
193 	timeout = jiffies + msecs_to_jiffies(10000);
194 	do {
195 		mdelay(10);
196 		if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag))
197 			return -1;
198 
199 	} while ((readl(dd->mmio + HOST_CTL) & HOST_RESET)
200 		 && time_before(jiffies, timeout));
201 
202 	if (readl(dd->mmio + HOST_CTL) & HOST_RESET)
203 		return -1;
204 
205 	return 0;
206 }
207 
208 /*
209  * Issue a command to the hardware.
210  *
211  * Set the appropriate bit in the s_active and Command Issue hardware
212  * registers, causing hardware command processing to begin.
213  *
214  * @port Pointer to the port structure.
215  * @tag  The tag of the command to be issued.
216  *
217  * return value
218  *      None
219  */
220 static inline void mtip_issue_ncq_command(struct mtip_port *port, int tag)
221 {
222 	int group = tag >> 5;
223 
224 	/* guard SACT and CI registers */
225 	spin_lock(&port->cmd_issue_lock[group]);
226 	writel((1 << MTIP_TAG_BIT(tag)),
227 			port->s_active[MTIP_TAG_INDEX(tag)]);
228 	writel((1 << MTIP_TAG_BIT(tag)),
229 			port->cmd_issue[MTIP_TAG_INDEX(tag)]);
230 	spin_unlock(&port->cmd_issue_lock[group]);
231 }
232 
233 /*
234  * Enable/disable the reception of FIS
235  *
236  * @port   Pointer to the port data structure
237  * @enable 1 to enable, 0 to disable
238  *
239  * return value
240  *	Previous state: 1 enabled, 0 disabled
241  */
242 static int mtip_enable_fis(struct mtip_port *port, int enable)
243 {
244 	u32 tmp;
245 
246 	/* enable FIS reception */
247 	tmp = readl(port->mmio + PORT_CMD);
248 	if (enable)
249 		writel(tmp | PORT_CMD_FIS_RX, port->mmio + PORT_CMD);
250 	else
251 		writel(tmp & ~PORT_CMD_FIS_RX, port->mmio + PORT_CMD);
252 
253 	/* Flush */
254 	readl(port->mmio + PORT_CMD);
255 
256 	return (((tmp & PORT_CMD_FIS_RX) == PORT_CMD_FIS_RX));
257 }
258 
259 /*
260  * Enable/disable the DMA engine
261  *
262  * @port   Pointer to the port data structure
263  * @enable 1 to enable, 0 to disable
264  *
265  * return value
266  *	Previous state: 1 enabled, 0 disabled.
267  */
268 static int mtip_enable_engine(struct mtip_port *port, int enable)
269 {
270 	u32 tmp;
271 
272 	/* enable FIS reception */
273 	tmp = readl(port->mmio + PORT_CMD);
274 	if (enable)
275 		writel(tmp | PORT_CMD_START, port->mmio + PORT_CMD);
276 	else
277 		writel(tmp & ~PORT_CMD_START, port->mmio + PORT_CMD);
278 
279 	readl(port->mmio + PORT_CMD);
280 	return (((tmp & PORT_CMD_START) == PORT_CMD_START));
281 }
282 
283 /*
284  * Enables the port DMA engine and FIS reception.
285  *
286  * return value
287  *	None
288  */
289 static inline void mtip_start_port(struct mtip_port *port)
290 {
291 	/* Enable FIS reception */
292 	mtip_enable_fis(port, 1);
293 
294 	/* Enable the DMA engine */
295 	mtip_enable_engine(port, 1);
296 }
297 
298 /*
299  * Deinitialize a port by disabling port interrupts, the DMA engine,
300  * and FIS reception.
301  *
302  * @port Pointer to the port structure
303  *
304  * return value
305  *	None
306  */
307 static inline void mtip_deinit_port(struct mtip_port *port)
308 {
309 	/* Disable interrupts on this port */
310 	writel(0, port->mmio + PORT_IRQ_MASK);
311 
312 	/* Disable the DMA engine */
313 	mtip_enable_engine(port, 0);
314 
315 	/* Disable FIS reception */
316 	mtip_enable_fis(port, 0);
317 }
318 
319 /*
320  * Initialize a port.
321  *
322  * This function deinitializes the port by calling mtip_deinit_port() and
323  * then initializes it by setting the command header and RX FIS addresses,
324  * clearing the SError register and any pending port interrupts before
325  * re-enabling the default set of port interrupts.
326  *
327  * @port Pointer to the port structure.
328  *
329  * return value
330  *	None
331  */
332 static void mtip_init_port(struct mtip_port *port)
333 {
334 	int i;
335 	mtip_deinit_port(port);
336 
337 	/* Program the command list base and FIS base addresses */
338 	if (readl(port->dd->mmio + HOST_CAP) & HOST_CAP_64) {
339 		writel((port->command_list_dma >> 16) >> 16,
340 			 port->mmio + PORT_LST_ADDR_HI);
341 		writel((port->rxfis_dma >> 16) >> 16,
342 			 port->mmio + PORT_FIS_ADDR_HI);
343 		set_bit(MTIP_PF_HOST_CAP_64, &port->flags);
344 	}
345 
346 	writel(port->command_list_dma & 0xFFFFFFFF,
347 			port->mmio + PORT_LST_ADDR);
348 	writel(port->rxfis_dma & 0xFFFFFFFF, port->mmio + PORT_FIS_ADDR);
349 
350 	/* Clear SError */
351 	writel(readl(port->mmio + PORT_SCR_ERR), port->mmio + PORT_SCR_ERR);
352 
353 	/* reset the completed registers.*/
354 	for (i = 0; i < port->dd->slot_groups; i++)
355 		writel(0xFFFFFFFF, port->completed[i]);
356 
357 	/* Clear any pending interrupts for this port */
358 	writel(readl(port->mmio + PORT_IRQ_STAT), port->mmio + PORT_IRQ_STAT);
359 
360 	/* Clear any pending interrupts on the HBA. */
361 	writel(readl(port->dd->mmio + HOST_IRQ_STAT),
362 					port->dd->mmio + HOST_IRQ_STAT);
363 
364 	/* Enable port interrupts */
365 	writel(DEF_PORT_IRQ, port->mmio + PORT_IRQ_MASK);
366 }
367 
368 /*
369  * Restart a port
370  *
371  * @port Pointer to the port data structure.
372  *
373  * return value
374  *	None
375  */
376 static void mtip_restart_port(struct mtip_port *port)
377 {
378 	unsigned long timeout;
379 
380 	/* Disable the DMA engine */
381 	mtip_enable_engine(port, 0);
382 
383 	/* Chip quirk: wait up to 500ms for PxCMD.CR == 0 */
384 	timeout = jiffies + msecs_to_jiffies(500);
385 	while ((readl(port->mmio + PORT_CMD) & PORT_CMD_LIST_ON)
386 		 && time_before(jiffies, timeout))
387 		;
388 
389 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag))
390 		return;
391 
392 	/*
393 	 * Chip quirk: escalate to hba reset if
394 	 * PxCMD.CR not clear after 500 ms
395 	 */
396 	if (readl(port->mmio + PORT_CMD) & PORT_CMD_LIST_ON) {
397 		dev_warn(&port->dd->pdev->dev,
398 			"PxCMD.CR not clear, escalating reset\n");
399 
400 		if (mtip_hba_reset(port->dd))
401 			dev_err(&port->dd->pdev->dev,
402 				"HBA reset escalation failed.\n");
403 
404 		/* 30 ms delay before com reset to quiesce chip */
405 		mdelay(30);
406 	}
407 
408 	dev_warn(&port->dd->pdev->dev, "Issuing COM reset\n");
409 
410 	/* Set PxSCTL.DET */
411 	writel(readl(port->mmio + PORT_SCR_CTL) |
412 			 1, port->mmio + PORT_SCR_CTL);
413 	readl(port->mmio + PORT_SCR_CTL);
414 
415 	/* Wait 1 ms to quiesce chip function */
416 	timeout = jiffies + msecs_to_jiffies(1);
417 	while (time_before(jiffies, timeout))
418 		;
419 
420 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag))
421 		return;
422 
423 	/* Clear PxSCTL.DET */
424 	writel(readl(port->mmio + PORT_SCR_CTL) & ~1,
425 			 port->mmio + PORT_SCR_CTL);
426 	readl(port->mmio + PORT_SCR_CTL);
427 
428 	/* Wait 500 ms for bit 0 of PORT_SCR_STS to be set */
429 	timeout = jiffies + msecs_to_jiffies(500);
430 	while (((readl(port->mmio + PORT_SCR_STAT) & 0x01) == 0)
431 			 && time_before(jiffies, timeout))
432 		;
433 
434 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag))
435 		return;
436 
437 	if ((readl(port->mmio + PORT_SCR_STAT) & 0x01) == 0)
438 		dev_warn(&port->dd->pdev->dev,
439 			"COM reset failed\n");
440 
441 	mtip_init_port(port);
442 	mtip_start_port(port);
443 
444 }
445 
446 static int mtip_device_reset(struct driver_data *dd)
447 {
448 	int rv = 0;
449 
450 	if (mtip_check_surprise_removal(dd->pdev))
451 		return 0;
452 
453 	if (mtip_hba_reset(dd) < 0)
454 		rv = -EFAULT;
455 
456 	mdelay(1);
457 	mtip_init_port(dd->port);
458 	mtip_start_port(dd->port);
459 
460 	/* Enable interrupts on the HBA. */
461 	writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN,
462 					dd->mmio + HOST_CTL);
463 	return rv;
464 }
465 
466 /*
467  * Helper function for tag logging
468  */
469 static void print_tags(struct driver_data *dd,
470 			char *msg,
471 			unsigned long *tagbits,
472 			int cnt)
473 {
474 	unsigned char tagmap[128];
475 	int group, tagmap_len = 0;
476 
477 	memset(tagmap, 0, sizeof(tagmap));
478 	for (group = SLOTBITS_IN_LONGS; group > 0; group--)
479 		tagmap_len += sprintf(tagmap + tagmap_len, "%016lX ",
480 						tagbits[group-1]);
481 	dev_warn(&dd->pdev->dev,
482 			"%d command(s) %s: tagmap [%s]", cnt, msg, tagmap);
483 }
484 
485 static int mtip_read_log_page(struct mtip_port *port, u8 page, u16 *buffer,
486 				dma_addr_t buffer_dma, unsigned int sectors);
487 static int mtip_get_smart_attr(struct mtip_port *port, unsigned int id,
488 						struct smart_attr *attrib);
489 
490 static void mtip_complete_command(struct mtip_cmd *cmd, blk_status_t status)
491 {
492 	struct request *req = blk_mq_rq_from_pdu(cmd);
493 
494 	cmd->status = status;
495 	if (likely(!blk_should_fake_timeout(req->q)))
496 		blk_mq_complete_request(req);
497 }
498 
499 /*
500  * Handle an error.
501  *
502  * @dd Pointer to the DRIVER_DATA structure.
503  *
504  * return value
505  *	None
506  */
507 static void mtip_handle_tfe(struct driver_data *dd)
508 {
509 	int group, tag, bit, reissue, rv;
510 	struct mtip_port *port;
511 	struct mtip_cmd  *cmd;
512 	u32 completed;
513 	struct host_to_dev_fis *fis;
514 	unsigned long tagaccum[SLOTBITS_IN_LONGS];
515 	unsigned int cmd_cnt = 0;
516 	unsigned char *buf;
517 	char *fail_reason = NULL;
518 	int fail_all_ncq_write = 0, fail_all_ncq_cmds = 0;
519 
520 	dev_warn(&dd->pdev->dev, "Taskfile error\n");
521 
522 	port = dd->port;
523 
524 	if (test_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags)) {
525 		cmd = mtip_cmd_from_tag(dd, MTIP_TAG_INTERNAL);
526 		dbg_printk(MTIP_DRV_NAME " TFE for the internal command\n");
527 		mtip_complete_command(cmd, BLK_STS_IOERR);
528 		return;
529 	}
530 
531 	/* clear the tag accumulator */
532 	memset(tagaccum, 0, SLOTBITS_IN_LONGS * sizeof(long));
533 
534 	/* Loop through all the groups */
535 	for (group = 0; group < dd->slot_groups; group++) {
536 		completed = readl(port->completed[group]);
537 
538 		dev_warn(&dd->pdev->dev, "g=%u, comp=%x\n", group, completed);
539 
540 		/* clear completed status register in the hardware.*/
541 		writel(completed, port->completed[group]);
542 
543 		/* Process successfully completed commands */
544 		for (bit = 0; bit < 32 && completed; bit++) {
545 			if (!(completed & (1<<bit)))
546 				continue;
547 			tag = (group << 5) + bit;
548 
549 			/* Skip the internal command slot */
550 			if (tag == MTIP_TAG_INTERNAL)
551 				continue;
552 
553 			cmd = mtip_cmd_from_tag(dd, tag);
554 			mtip_complete_command(cmd, 0);
555 			set_bit(tag, tagaccum);
556 			cmd_cnt++;
557 		}
558 	}
559 
560 	print_tags(dd, "completed (TFE)", tagaccum, cmd_cnt);
561 
562 	/* Restart the port */
563 	mdelay(20);
564 	mtip_restart_port(port);
565 
566 	/* Trying to determine the cause of the error */
567 	rv = mtip_read_log_page(dd->port, ATA_LOG_SATA_NCQ,
568 				dd->port->log_buf,
569 				dd->port->log_buf_dma, 1);
570 	if (rv) {
571 		dev_warn(&dd->pdev->dev,
572 			"Error in READ LOG EXT (10h) command\n");
573 		/* non-critical error, don't fail the load */
574 	} else {
575 		buf = (unsigned char *)dd->port->log_buf;
576 		if (buf[259] & 0x1) {
577 			dev_info(&dd->pdev->dev,
578 				"Write protect bit is set.\n");
579 			set_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag);
580 			fail_all_ncq_write = 1;
581 			fail_reason = "write protect";
582 		}
583 		if (buf[288] == 0xF7) {
584 			dev_info(&dd->pdev->dev,
585 				"Exceeded Tmax, drive in thermal shutdown.\n");
586 			set_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag);
587 			fail_all_ncq_cmds = 1;
588 			fail_reason = "thermal shutdown";
589 		}
590 		if (buf[288] == 0xBF) {
591 			set_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag);
592 			dev_info(&dd->pdev->dev,
593 				"Drive indicates rebuild has failed. Secure erase required.\n");
594 			fail_all_ncq_cmds = 1;
595 			fail_reason = "rebuild failed";
596 		}
597 	}
598 
599 	/* clear the tag accumulator */
600 	memset(tagaccum, 0, SLOTBITS_IN_LONGS * sizeof(long));
601 
602 	/* Loop through all the groups */
603 	for (group = 0; group < dd->slot_groups; group++) {
604 		for (bit = 0; bit < 32; bit++) {
605 			reissue = 1;
606 			tag = (group << 5) + bit;
607 			cmd = mtip_cmd_from_tag(dd, tag);
608 
609 			fis = (struct host_to_dev_fis *)cmd->command;
610 
611 			/* Should re-issue? */
612 			if (tag == MTIP_TAG_INTERNAL ||
613 			    fis->command == ATA_CMD_SET_FEATURES)
614 				reissue = 0;
615 			else {
616 				if (fail_all_ncq_cmds ||
617 					(fail_all_ncq_write &&
618 					fis->command == ATA_CMD_FPDMA_WRITE)) {
619 					dev_warn(&dd->pdev->dev,
620 					"  Fail: %s w/tag %d [%s].\n",
621 					fis->command == ATA_CMD_FPDMA_WRITE ?
622 						"write" : "read",
623 					tag,
624 					fail_reason != NULL ?
625 						fail_reason : "unknown");
626 					mtip_complete_command(cmd, BLK_STS_MEDIUM);
627 					continue;
628 				}
629 			}
630 
631 			/*
632 			 * First check if this command has
633 			 *  exceeded its retries.
634 			 */
635 			if (reissue && (cmd->retries-- > 0)) {
636 
637 				set_bit(tag, tagaccum);
638 
639 				/* Re-issue the command. */
640 				mtip_issue_ncq_command(port, tag);
641 
642 				continue;
643 			}
644 
645 			/* Retire a command that will not be reissued */
646 			dev_warn(&port->dd->pdev->dev,
647 				"retiring tag %d\n", tag);
648 
649 			mtip_complete_command(cmd, BLK_STS_IOERR);
650 		}
651 	}
652 	print_tags(dd, "reissued (TFE)", tagaccum, cmd_cnt);
653 }
654 
655 /*
656  * Handle a set device bits interrupt
657  */
658 static inline void mtip_workq_sdbfx(struct mtip_port *port, int group,
659 							u32 completed)
660 {
661 	struct driver_data *dd = port->dd;
662 	int tag, bit;
663 	struct mtip_cmd *command;
664 
665 	if (!completed) {
666 		WARN_ON_ONCE(!completed);
667 		return;
668 	}
669 	/* clear completed status register in the hardware.*/
670 	writel(completed, port->completed[group]);
671 
672 	/* Process completed commands. */
673 	for (bit = 0; (bit < 32) && completed; bit++) {
674 		if (completed & 0x01) {
675 			tag = (group << 5) | bit;
676 
677 			/* skip internal command slot. */
678 			if (unlikely(tag == MTIP_TAG_INTERNAL))
679 				continue;
680 
681 			command = mtip_cmd_from_tag(dd, tag);
682 			mtip_complete_command(command, 0);
683 		}
684 		completed >>= 1;
685 	}
686 
687 	/* If last, re-enable interrupts */
688 	if (atomic_dec_return(&dd->irq_workers_active) == 0)
689 		writel(0xffffffff, dd->mmio + HOST_IRQ_STAT);
690 }
691 
692 /*
693  * Process legacy pio and d2h interrupts
694  */
695 static inline void mtip_process_legacy(struct driver_data *dd, u32 port_stat)
696 {
697 	struct mtip_port *port = dd->port;
698 	struct mtip_cmd *cmd = mtip_cmd_from_tag(dd, MTIP_TAG_INTERNAL);
699 
700 	if (test_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags) && cmd) {
701 		int group = MTIP_TAG_INDEX(MTIP_TAG_INTERNAL);
702 		int status = readl(port->cmd_issue[group]);
703 
704 		if (!(status & (1 << MTIP_TAG_BIT(MTIP_TAG_INTERNAL))))
705 			mtip_complete_command(cmd, 0);
706 	}
707 }
708 
709 /*
710  * Demux and handle errors
711  */
712 static inline void mtip_process_errors(struct driver_data *dd, u32 port_stat)
713 {
714 	if (unlikely(port_stat & PORT_IRQ_CONNECT)) {
715 		dev_warn(&dd->pdev->dev,
716 			"Clearing PxSERR.DIAG.x\n");
717 		writel((1 << 26), dd->port->mmio + PORT_SCR_ERR);
718 	}
719 
720 	if (unlikely(port_stat & PORT_IRQ_PHYRDY)) {
721 		dev_warn(&dd->pdev->dev,
722 			"Clearing PxSERR.DIAG.n\n");
723 		writel((1 << 16), dd->port->mmio + PORT_SCR_ERR);
724 	}
725 
726 	if (unlikely(port_stat & ~PORT_IRQ_HANDLED)) {
727 		dev_warn(&dd->pdev->dev,
728 			"Port stat errors %x unhandled\n",
729 			(port_stat & ~PORT_IRQ_HANDLED));
730 		if (mtip_check_surprise_removal(dd->pdev))
731 			return;
732 	}
733 	if (likely(port_stat & (PORT_IRQ_TF_ERR | PORT_IRQ_IF_ERR))) {
734 		set_bit(MTIP_PF_EH_ACTIVE_BIT, &dd->port->flags);
735 		wake_up_interruptible(&dd->port->svc_wait);
736 	}
737 }
738 
739 static inline irqreturn_t mtip_handle_irq(struct driver_data *data)
740 {
741 	struct driver_data *dd = (struct driver_data *) data;
742 	struct mtip_port *port = dd->port;
743 	u32 hba_stat, port_stat;
744 	int rv = IRQ_NONE;
745 	int do_irq_enable = 1, i, workers;
746 	struct mtip_work *twork;
747 
748 	hba_stat = readl(dd->mmio + HOST_IRQ_STAT);
749 	if (hba_stat) {
750 		rv = IRQ_HANDLED;
751 
752 		/* Acknowledge the interrupt status on the port.*/
753 		port_stat = readl(port->mmio + PORT_IRQ_STAT);
754 		if (unlikely(port_stat == 0xFFFFFFFF)) {
755 			mtip_check_surprise_removal(dd->pdev);
756 			return IRQ_HANDLED;
757 		}
758 		writel(port_stat, port->mmio + PORT_IRQ_STAT);
759 
760 		/* Demux port status */
761 		if (likely(port_stat & PORT_IRQ_SDB_FIS)) {
762 			do_irq_enable = 0;
763 			WARN_ON_ONCE(atomic_read(&dd->irq_workers_active) != 0);
764 
765 			/* Start at 1: group zero is always local? */
766 			for (i = 0, workers = 0; i < MTIP_MAX_SLOT_GROUPS;
767 									i++) {
768 				twork = &dd->work[i];
769 				twork->completed = readl(port->completed[i]);
770 				if (twork->completed)
771 					workers++;
772 			}
773 
774 			atomic_set(&dd->irq_workers_active, workers);
775 			if (workers) {
776 				for (i = 1; i < MTIP_MAX_SLOT_GROUPS; i++) {
777 					twork = &dd->work[i];
778 					if (twork->completed)
779 						queue_work_on(
780 							twork->cpu_binding,
781 							dd->isr_workq,
782 							&twork->work);
783 				}
784 
785 				if (likely(dd->work[0].completed))
786 					mtip_workq_sdbfx(port, 0,
787 							dd->work[0].completed);
788 
789 			} else {
790 				/*
791 				 * Chip quirk: SDB interrupt but nothing
792 				 * to complete
793 				 */
794 				do_irq_enable = 1;
795 			}
796 		}
797 
798 		if (unlikely(port_stat & PORT_IRQ_ERR)) {
799 			if (unlikely(mtip_check_surprise_removal(dd->pdev))) {
800 				/* don't proceed further */
801 				return IRQ_HANDLED;
802 			}
803 			if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
804 							&dd->dd_flag))
805 				return rv;
806 
807 			mtip_process_errors(dd, port_stat & PORT_IRQ_ERR);
808 		}
809 
810 		if (unlikely(port_stat & PORT_IRQ_LEGACY))
811 			mtip_process_legacy(dd, port_stat & PORT_IRQ_LEGACY);
812 	}
813 
814 	/* acknowledge interrupt */
815 	if (unlikely(do_irq_enable))
816 		writel(hba_stat, dd->mmio + HOST_IRQ_STAT);
817 
818 	return rv;
819 }
820 
821 /*
822  * HBA interrupt subroutine.
823  *
824  * @irq		IRQ number.
825  * @instance	Pointer to the driver data structure.
826  *
827  * return value
828  *	IRQ_HANDLED	A HBA interrupt was pending and handled.
829  *	IRQ_NONE	This interrupt was not for the HBA.
830  */
831 static irqreturn_t mtip_irq_handler(int irq, void *instance)
832 {
833 	struct driver_data *dd = instance;
834 
835 	return mtip_handle_irq(dd);
836 }
837 
838 static void mtip_issue_non_ncq_command(struct mtip_port *port, int tag)
839 {
840 	writel(1 << MTIP_TAG_BIT(tag), port->cmd_issue[MTIP_TAG_INDEX(tag)]);
841 }
842 
843 static bool mtip_pause_ncq(struct mtip_port *port,
844 				struct host_to_dev_fis *fis)
845 {
846 	unsigned long task_file_data;
847 
848 	task_file_data = readl(port->mmio+PORT_TFDATA);
849 	if ((task_file_data & 1))
850 		return false;
851 
852 	if (fis->command == ATA_CMD_SEC_ERASE_PREP) {
853 		port->ic_pause_timer = jiffies;
854 		return true;
855 	} else if ((fis->command == ATA_CMD_DOWNLOAD_MICRO) &&
856 					(fis->features == 0x03)) {
857 		set_bit(MTIP_PF_DM_ACTIVE_BIT, &port->flags);
858 		port->ic_pause_timer = jiffies;
859 		return true;
860 	} else if ((fis->command == ATA_CMD_SEC_ERASE_UNIT) ||
861 		((fis->command == 0xFC) &&
862 			(fis->features == 0x27 || fis->features == 0x72 ||
863 			 fis->features == 0x62 || fis->features == 0x26))) {
864 		clear_bit(MTIP_DDF_SEC_LOCK_BIT, &port->dd->dd_flag);
865 		clear_bit(MTIP_DDF_REBUILD_FAILED_BIT, &port->dd->dd_flag);
866 		/* Com reset after secure erase or lowlevel format */
867 		mtip_restart_port(port);
868 		clear_bit(MTIP_PF_SE_ACTIVE_BIT, &port->flags);
869 		return false;
870 	}
871 
872 	return false;
873 }
874 
875 static bool mtip_commands_active(struct mtip_port *port)
876 {
877 	unsigned int active;
878 	unsigned int n;
879 
880 	/*
881 	 * Ignore s_active bit 0 of array element 0.
882 	 * This bit will always be set
883 	 */
884 	active = readl(port->s_active[0]) & 0xFFFFFFFE;
885 	for (n = 1; n < port->dd->slot_groups; n++)
886 		active |= readl(port->s_active[n]);
887 
888 	return active != 0;
889 }
890 
891 /*
892  * Wait for port to quiesce
893  *
894  * @port    Pointer to port data structure
895  * @timeout Max duration to wait (ms)
896  *
897  * return value
898  *	0	Success
899  *	-EBUSY  Commands still active
900  */
901 static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout)
902 {
903 	unsigned long to;
904 	bool active = true;
905 
906 	blk_mq_quiesce_queue(port->dd->queue);
907 
908 	to = jiffies + msecs_to_jiffies(timeout);
909 	do {
910 		if (test_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags) &&
911 			test_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags)) {
912 			msleep(20);
913 			continue; /* svc thd is actively issuing commands */
914 		}
915 
916 		msleep(100);
917 
918 		if (mtip_check_surprise_removal(port->dd->pdev))
919 			goto err_fault;
920 
921 		active = mtip_commands_active(port);
922 		if (!active)
923 			break;
924 	} while (time_before(jiffies, to));
925 
926 	blk_mq_unquiesce_queue(port->dd->queue);
927 	return active ? -EBUSY : 0;
928 err_fault:
929 	blk_mq_unquiesce_queue(port->dd->queue);
930 	return -EFAULT;
931 }
932 
933 struct mtip_int_cmd {
934 	int fis_len;
935 	dma_addr_t buffer;
936 	int buf_len;
937 	u32 opts;
938 };
939 
940 /*
941  * Execute an internal command and wait for the completion.
942  *
943  * @port    Pointer to the port data structure.
944  * @fis     Pointer to the FIS that describes the command.
945  * @fis_len  Length in WORDS of the FIS.
946  * @buffer  DMA accessible for command data.
947  * @buf_len  Length, in bytes, of the data buffer.
948  * @opts    Command header options, excluding the FIS length
949  *             and the number of PRD entries.
950  * @timeout Time in ms to wait for the command to complete.
951  *
952  * return value
953  *	0	 Command completed successfully.
954  *	-EFAULT  The buffer address is not correctly aligned.
955  *	-EBUSY   Internal command or other IO in progress.
956  *	-EAGAIN  Time out waiting for command to complete.
957  */
958 static int mtip_exec_internal_command(struct mtip_port *port,
959 					struct host_to_dev_fis *fis,
960 					int fis_len,
961 					dma_addr_t buffer,
962 					int buf_len,
963 					u32 opts,
964 					unsigned long timeout)
965 {
966 	struct mtip_cmd *int_cmd;
967 	struct driver_data *dd = port->dd;
968 	struct request *rq;
969 	struct mtip_int_cmd icmd = {
970 		.fis_len = fis_len,
971 		.buffer = buffer,
972 		.buf_len = buf_len,
973 		.opts = opts
974 	};
975 	int rv = 0;
976 
977 	/* Make sure the buffer is 8 byte aligned. This is asic specific. */
978 	if (buffer & 0x00000007) {
979 		dev_err(&dd->pdev->dev, "SG buffer is not 8 byte aligned\n");
980 		return -EFAULT;
981 	}
982 
983 	if (mtip_check_surprise_removal(dd->pdev))
984 		return -EFAULT;
985 
986 	rq = blk_mq_alloc_request(dd->queue, REQ_OP_DRV_IN, BLK_MQ_REQ_RESERVED);
987 	if (IS_ERR(rq)) {
988 		dbg_printk(MTIP_DRV_NAME "Unable to allocate tag for PIO cmd\n");
989 		return -EFAULT;
990 	}
991 
992 	set_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags);
993 
994 	if (fis->command == ATA_CMD_SEC_ERASE_PREP)
995 		set_bit(MTIP_PF_SE_ACTIVE_BIT, &port->flags);
996 
997 	clear_bit(MTIP_PF_DM_ACTIVE_BIT, &port->flags);
998 
999 	if (fis->command != ATA_CMD_STANDBYNOW1) {
1000 		/* wait for io to complete if non atomic */
1001 		if (mtip_quiesce_io(port, MTIP_QUIESCE_IO_TIMEOUT_MS) < 0) {
1002 			dev_warn(&dd->pdev->dev, "Failed to quiesce IO\n");
1003 			blk_mq_free_request(rq);
1004 			clear_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags);
1005 			wake_up_interruptible(&port->svc_wait);
1006 			return -EBUSY;
1007 		}
1008 	}
1009 
1010 	/* Copy the command to the command table */
1011 	int_cmd = blk_mq_rq_to_pdu(rq);
1012 	int_cmd->icmd = &icmd;
1013 	memcpy(int_cmd->command, fis, fis_len*4);
1014 
1015 	rq->timeout = timeout;
1016 
1017 	/* insert request and run queue */
1018 	blk_execute_rq(NULL, rq, true);
1019 
1020 	if (int_cmd->status) {
1021 		dev_err(&dd->pdev->dev, "Internal command [%02X] failed %d\n",
1022 				fis->command, int_cmd->status);
1023 		rv = -EIO;
1024 
1025 		if (mtip_check_surprise_removal(dd->pdev) ||
1026 			test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
1027 					&dd->dd_flag)) {
1028 			dev_err(&dd->pdev->dev,
1029 				"Internal command [%02X] wait returned due to SR\n",
1030 				fis->command);
1031 			rv = -ENXIO;
1032 			goto exec_ic_exit;
1033 		}
1034 		mtip_device_reset(dd); /* recover from timeout issue */
1035 		rv = -EAGAIN;
1036 		goto exec_ic_exit;
1037 	}
1038 
1039 	if (readl(port->cmd_issue[MTIP_TAG_INDEX(MTIP_TAG_INTERNAL)])
1040 			& (1 << MTIP_TAG_BIT(MTIP_TAG_INTERNAL))) {
1041 		rv = -ENXIO;
1042 		if (!test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)) {
1043 			mtip_device_reset(dd);
1044 			rv = -EAGAIN;
1045 		}
1046 	}
1047 exec_ic_exit:
1048 	/* Clear the allocated and active bits for the internal command. */
1049 	blk_mq_free_request(rq);
1050 	clear_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags);
1051 	if (rv >= 0 && mtip_pause_ncq(port, fis)) {
1052 		/* NCQ paused */
1053 		return rv;
1054 	}
1055 	wake_up_interruptible(&port->svc_wait);
1056 
1057 	return rv;
1058 }
1059 
1060 /*
1061  * Byte-swap ATA ID strings.
1062  *
1063  * ATA identify data contains strings in byte-swapped 16-bit words.
1064  * They must be swapped (on all architectures) to be usable as C strings.
1065  * This function swaps bytes in-place.
1066  *
1067  * @buf The buffer location of the string
1068  * @len The number of bytes to swap
1069  *
1070  * return value
1071  *	None
1072  */
1073 static inline void ata_swap_string(u16 *buf, unsigned int len)
1074 {
1075 	int i;
1076 	for (i = 0; i < (len/2); i++)
1077 		be16_to_cpus(&buf[i]);
1078 }
1079 
1080 static void mtip_set_timeout(struct driver_data *dd,
1081 					struct host_to_dev_fis *fis,
1082 					unsigned int *timeout, u8 erasemode)
1083 {
1084 	switch (fis->command) {
1085 	case ATA_CMD_DOWNLOAD_MICRO:
1086 		*timeout = 120000; /* 2 minutes */
1087 		break;
1088 	case ATA_CMD_SEC_ERASE_UNIT:
1089 	case 0xFC:
1090 		if (erasemode)
1091 			*timeout = ((*(dd->port->identify + 90) * 2) * 60000);
1092 		else
1093 			*timeout = ((*(dd->port->identify + 89) * 2) * 60000);
1094 		break;
1095 	case ATA_CMD_STANDBYNOW1:
1096 		*timeout = 120000;  /* 2 minutes */
1097 		break;
1098 	case 0xF7:
1099 	case 0xFA:
1100 		*timeout = 60000;  /* 60 seconds */
1101 		break;
1102 	case ATA_CMD_SMART:
1103 		*timeout = 15000;  /* 15 seconds */
1104 		break;
1105 	default:
1106 		*timeout = MTIP_IOCTL_CMD_TIMEOUT_MS;
1107 		break;
1108 	}
1109 }
1110 
1111 /*
1112  * Request the device identity information.
1113  *
1114  * If a user space buffer is not specified, i.e. is NULL, the
1115  * identify information is still read from the drive and placed
1116  * into the identify data buffer (@e port->identify) in the
1117  * port data structure.
1118  * When the identify buffer contains valid identify information @e
1119  * port->identify_valid is non-zero.
1120  *
1121  * @port	 Pointer to the port structure.
1122  * @user_buffer  A user space buffer where the identify data should be
1123  *                    copied.
1124  *
1125  * return value
1126  *	0	Command completed successfully.
1127  *	-EFAULT An error occurred while coping data to the user buffer.
1128  *	-1	Command failed.
1129  */
1130 static int mtip_get_identify(struct mtip_port *port, void __user *user_buffer)
1131 {
1132 	int rv = 0;
1133 	struct host_to_dev_fis fis;
1134 
1135 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag))
1136 		return -EFAULT;
1137 
1138 	/* Build the FIS. */
1139 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1140 	fis.type	= 0x27;
1141 	fis.opts	= 1 << 7;
1142 	fis.command	= ATA_CMD_ID_ATA;
1143 
1144 	/* Set the identify information as invalid. */
1145 	port->identify_valid = 0;
1146 
1147 	/* Clear the identify information. */
1148 	memset(port->identify, 0, sizeof(u16) * ATA_ID_WORDS);
1149 
1150 	/* Execute the command. */
1151 	if (mtip_exec_internal_command(port,
1152 				&fis,
1153 				5,
1154 				port->identify_dma,
1155 				sizeof(u16) * ATA_ID_WORDS,
1156 				0,
1157 				MTIP_INT_CMD_TIMEOUT_MS)
1158 				< 0) {
1159 		rv = -1;
1160 		goto out;
1161 	}
1162 
1163 	/*
1164 	 * Perform any necessary byte-swapping.  Yes, the kernel does in fact
1165 	 * perform field-sensitive swapping on the string fields.
1166 	 * See the kernel use of ata_id_string() for proof of this.
1167 	 */
1168 #ifdef __LITTLE_ENDIAN
1169 	ata_swap_string(port->identify + 27, 40);  /* model string*/
1170 	ata_swap_string(port->identify + 23, 8);   /* firmware string*/
1171 	ata_swap_string(port->identify + 10, 20);  /* serial# string*/
1172 #else
1173 	{
1174 		int i;
1175 		for (i = 0; i < ATA_ID_WORDS; i++)
1176 			port->identify[i] = le16_to_cpu(port->identify[i]);
1177 	}
1178 #endif
1179 
1180 	/* Check security locked state */
1181 	if (port->identify[128] & 0x4)
1182 		set_bit(MTIP_DDF_SEC_LOCK_BIT, &port->dd->dd_flag);
1183 	else
1184 		clear_bit(MTIP_DDF_SEC_LOCK_BIT, &port->dd->dd_flag);
1185 
1186 	/* Set the identify buffer as valid. */
1187 	port->identify_valid = 1;
1188 
1189 	if (user_buffer) {
1190 		if (copy_to_user(
1191 			user_buffer,
1192 			port->identify,
1193 			ATA_ID_WORDS * sizeof(u16))) {
1194 			rv = -EFAULT;
1195 			goto out;
1196 		}
1197 	}
1198 
1199 out:
1200 	return rv;
1201 }
1202 
1203 /*
1204  * Issue a standby immediate command to the device.
1205  *
1206  * @port Pointer to the port structure.
1207  *
1208  * return value
1209  *	0	Command was executed successfully.
1210  *	-1	An error occurred while executing the command.
1211  */
1212 static int mtip_standby_immediate(struct mtip_port *port)
1213 {
1214 	int rv;
1215 	struct host_to_dev_fis	fis;
1216 	unsigned long __maybe_unused start;
1217 	unsigned int timeout;
1218 
1219 	/* Build the FIS. */
1220 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1221 	fis.type	= 0x27;
1222 	fis.opts	= 1 << 7;
1223 	fis.command	= ATA_CMD_STANDBYNOW1;
1224 
1225 	mtip_set_timeout(port->dd, &fis, &timeout, 0);
1226 
1227 	start = jiffies;
1228 	rv = mtip_exec_internal_command(port,
1229 					&fis,
1230 					5,
1231 					0,
1232 					0,
1233 					0,
1234 					timeout);
1235 	dbg_printk(MTIP_DRV_NAME "Time taken to complete standby cmd: %d ms\n",
1236 			jiffies_to_msecs(jiffies - start));
1237 	if (rv)
1238 		dev_warn(&port->dd->pdev->dev,
1239 			"STANDBY IMMEDIATE command failed.\n");
1240 
1241 	return rv;
1242 }
1243 
1244 /*
1245  * Issue a READ LOG EXT command to the device.
1246  *
1247  * @port	pointer to the port structure.
1248  * @page	page number to fetch
1249  * @buffer	pointer to buffer
1250  * @buffer_dma	dma address corresponding to @buffer
1251  * @sectors	page length to fetch, in sectors
1252  *
1253  * return value
1254  *	@rv	return value from mtip_exec_internal_command()
1255  */
1256 static int mtip_read_log_page(struct mtip_port *port, u8 page, u16 *buffer,
1257 				dma_addr_t buffer_dma, unsigned int sectors)
1258 {
1259 	struct host_to_dev_fis fis;
1260 
1261 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1262 	fis.type	= 0x27;
1263 	fis.opts	= 1 << 7;
1264 	fis.command	= ATA_CMD_READ_LOG_EXT;
1265 	fis.sect_count	= sectors & 0xFF;
1266 	fis.sect_cnt_ex	= (sectors >> 8) & 0xFF;
1267 	fis.lba_low	= page;
1268 	fis.lba_mid	= 0;
1269 	fis.device	= ATA_DEVICE_OBS;
1270 
1271 	memset(buffer, 0, sectors * ATA_SECT_SIZE);
1272 
1273 	return mtip_exec_internal_command(port,
1274 					&fis,
1275 					5,
1276 					buffer_dma,
1277 					sectors * ATA_SECT_SIZE,
1278 					0,
1279 					MTIP_INT_CMD_TIMEOUT_MS);
1280 }
1281 
1282 /*
1283  * Issue a SMART READ DATA command to the device.
1284  *
1285  * @port	pointer to the port structure.
1286  * @buffer	pointer to buffer
1287  * @buffer_dma	dma address corresponding to @buffer
1288  *
1289  * return value
1290  *	@rv	return value from mtip_exec_internal_command()
1291  */
1292 static int mtip_get_smart_data(struct mtip_port *port, u8 *buffer,
1293 					dma_addr_t buffer_dma)
1294 {
1295 	struct host_to_dev_fis fis;
1296 
1297 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1298 	fis.type	= 0x27;
1299 	fis.opts	= 1 << 7;
1300 	fis.command	= ATA_CMD_SMART;
1301 	fis.features	= 0xD0;
1302 	fis.sect_count	= 1;
1303 	fis.lba_mid	= 0x4F;
1304 	fis.lba_hi	= 0xC2;
1305 	fis.device	= ATA_DEVICE_OBS;
1306 
1307 	return mtip_exec_internal_command(port,
1308 					&fis,
1309 					5,
1310 					buffer_dma,
1311 					ATA_SECT_SIZE,
1312 					0,
1313 					15000);
1314 }
1315 
1316 /*
1317  * Get the value of a smart attribute
1318  *
1319  * @port	pointer to the port structure
1320  * @id		attribute number
1321  * @attrib	pointer to return attrib information corresponding to @id
1322  *
1323  * return value
1324  *	-EINVAL	NULL buffer passed or unsupported attribute @id.
1325  *	-EPERM	Identify data not valid, SMART not supported or not enabled
1326  */
1327 static int mtip_get_smart_attr(struct mtip_port *port, unsigned int id,
1328 						struct smart_attr *attrib)
1329 {
1330 	int rv, i;
1331 	struct smart_attr *pattr;
1332 
1333 	if (!attrib)
1334 		return -EINVAL;
1335 
1336 	if (!port->identify_valid) {
1337 		dev_warn(&port->dd->pdev->dev, "IDENTIFY DATA not valid\n");
1338 		return -EPERM;
1339 	}
1340 	if (!(port->identify[82] & 0x1)) {
1341 		dev_warn(&port->dd->pdev->dev, "SMART not supported\n");
1342 		return -EPERM;
1343 	}
1344 	if (!(port->identify[85] & 0x1)) {
1345 		dev_warn(&port->dd->pdev->dev, "SMART not enabled\n");
1346 		return -EPERM;
1347 	}
1348 
1349 	memset(port->smart_buf, 0, ATA_SECT_SIZE);
1350 	rv = mtip_get_smart_data(port, port->smart_buf, port->smart_buf_dma);
1351 	if (rv) {
1352 		dev_warn(&port->dd->pdev->dev, "Failed to ge SMART data\n");
1353 		return rv;
1354 	}
1355 
1356 	pattr = (struct smart_attr *)(port->smart_buf + 2);
1357 	for (i = 0; i < 29; i++, pattr++)
1358 		if (pattr->attr_id == id) {
1359 			memcpy(attrib, pattr, sizeof(struct smart_attr));
1360 			break;
1361 		}
1362 
1363 	if (i == 29) {
1364 		dev_warn(&port->dd->pdev->dev,
1365 			"Query for invalid SMART attribute ID\n");
1366 		rv = -EINVAL;
1367 	}
1368 
1369 	return rv;
1370 }
1371 
1372 /*
1373  * Get the drive capacity.
1374  *
1375  * @dd      Pointer to the device data structure.
1376  * @sectors Pointer to the variable that will receive the sector count.
1377  *
1378  * return value
1379  *	1 Capacity was returned successfully.
1380  *	0 The identify information is invalid.
1381  */
1382 static bool mtip_hw_get_capacity(struct driver_data *dd, sector_t *sectors)
1383 {
1384 	struct mtip_port *port = dd->port;
1385 	u64 total, raw0, raw1, raw2, raw3;
1386 	raw0 = port->identify[100];
1387 	raw1 = port->identify[101];
1388 	raw2 = port->identify[102];
1389 	raw3 = port->identify[103];
1390 	total = raw0 | raw1<<16 | raw2<<32 | raw3<<48;
1391 	*sectors = total;
1392 	return (bool) !!port->identify_valid;
1393 }
1394 
1395 /*
1396  * Display the identify command data.
1397  *
1398  * @port Pointer to the port data structure.
1399  *
1400  * return value
1401  *	None
1402  */
1403 static void mtip_dump_identify(struct mtip_port *port)
1404 {
1405 	sector_t sectors;
1406 	unsigned short revid;
1407 	char cbuf[42];
1408 
1409 	if (!port->identify_valid)
1410 		return;
1411 
1412 	strlcpy(cbuf, (char *)(port->identify+10), 21);
1413 	dev_info(&port->dd->pdev->dev,
1414 		"Serial No.: %s\n", cbuf);
1415 
1416 	strlcpy(cbuf, (char *)(port->identify+23), 9);
1417 	dev_info(&port->dd->pdev->dev,
1418 		"Firmware Ver.: %s\n", cbuf);
1419 
1420 	strlcpy(cbuf, (char *)(port->identify+27), 41);
1421 	dev_info(&port->dd->pdev->dev, "Model: %s\n", cbuf);
1422 
1423 	dev_info(&port->dd->pdev->dev, "Security: %04x %s\n",
1424 		port->identify[128],
1425 		port->identify[128] & 0x4 ? "(LOCKED)" : "");
1426 
1427 	if (mtip_hw_get_capacity(port->dd, &sectors))
1428 		dev_info(&port->dd->pdev->dev,
1429 			"Capacity: %llu sectors (%llu MB)\n",
1430 			 (u64)sectors,
1431 			 ((u64)sectors) * ATA_SECT_SIZE >> 20);
1432 
1433 	pci_read_config_word(port->dd->pdev, PCI_REVISION_ID, &revid);
1434 	switch (revid & 0xFF) {
1435 	case 0x1:
1436 		strlcpy(cbuf, "A0", 3);
1437 		break;
1438 	case 0x3:
1439 		strlcpy(cbuf, "A2", 3);
1440 		break;
1441 	default:
1442 		strlcpy(cbuf, "?", 2);
1443 		break;
1444 	}
1445 	dev_info(&port->dd->pdev->dev,
1446 		"Card Type: %s\n", cbuf);
1447 }
1448 
1449 /*
1450  * Map the commands scatter list into the command table.
1451  *
1452  * @command Pointer to the command.
1453  * @nents Number of scatter list entries.
1454  *
1455  * return value
1456  *	None
1457  */
1458 static inline void fill_command_sg(struct driver_data *dd,
1459 				struct mtip_cmd *command,
1460 				int nents)
1461 {
1462 	int n;
1463 	unsigned int dma_len;
1464 	struct mtip_cmd_sg *command_sg;
1465 	struct scatterlist *sg;
1466 
1467 	command_sg = command->command + AHCI_CMD_TBL_HDR_SZ;
1468 
1469 	for_each_sg(command->sg, sg, nents, n) {
1470 		dma_len = sg_dma_len(sg);
1471 		if (dma_len > 0x400000)
1472 			dev_err(&dd->pdev->dev,
1473 				"DMA segment length truncated\n");
1474 		command_sg->info = cpu_to_le32((dma_len-1) & 0x3FFFFF);
1475 		command_sg->dba	=  cpu_to_le32(sg_dma_address(sg));
1476 		command_sg->dba_upper =
1477 			cpu_to_le32((sg_dma_address(sg) >> 16) >> 16);
1478 		command_sg++;
1479 	}
1480 }
1481 
1482 /*
1483  * @brief Execute a drive command.
1484  *
1485  * return value 0 The command completed successfully.
1486  * return value -1 An error occurred while executing the command.
1487  */
1488 static int exec_drive_task(struct mtip_port *port, u8 *command)
1489 {
1490 	struct host_to_dev_fis	fis;
1491 	struct host_to_dev_fis *reply = (port->rxfis + RX_FIS_D2H_REG);
1492 	unsigned int to;
1493 
1494 	/* Build the FIS. */
1495 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1496 	fis.type	= 0x27;
1497 	fis.opts	= 1 << 7;
1498 	fis.command	= command[0];
1499 	fis.features	= command[1];
1500 	fis.sect_count	= command[2];
1501 	fis.sector	= command[3];
1502 	fis.cyl_low	= command[4];
1503 	fis.cyl_hi	= command[5];
1504 	fis.device	= command[6] & ~0x10; /* Clear the dev bit*/
1505 
1506 	mtip_set_timeout(port->dd, &fis, &to, 0);
1507 
1508 	dbg_printk(MTIP_DRV_NAME " %s: User Command: cmd %x, feat %x, nsect %x, sect %x, lcyl %x, hcyl %x, sel %x\n",
1509 		__func__,
1510 		command[0],
1511 		command[1],
1512 		command[2],
1513 		command[3],
1514 		command[4],
1515 		command[5],
1516 		command[6]);
1517 
1518 	/* Execute the command. */
1519 	if (mtip_exec_internal_command(port,
1520 				 &fis,
1521 				 5,
1522 				 0,
1523 				 0,
1524 				 0,
1525 				 to) < 0) {
1526 		return -1;
1527 	}
1528 
1529 	command[0] = reply->command; /* Status*/
1530 	command[1] = reply->features; /* Error*/
1531 	command[4] = reply->cyl_low;
1532 	command[5] = reply->cyl_hi;
1533 
1534 	dbg_printk(MTIP_DRV_NAME " %s: Completion Status: stat %x, err %x , cyl_lo %x cyl_hi %x\n",
1535 		__func__,
1536 		command[0],
1537 		command[1],
1538 		command[4],
1539 		command[5]);
1540 
1541 	return 0;
1542 }
1543 
1544 /*
1545  * @brief Execute a drive command.
1546  *
1547  * @param port Pointer to the port data structure.
1548  * @param command Pointer to the user specified command parameters.
1549  * @param user_buffer Pointer to the user space buffer where read sector
1550  *                   data should be copied.
1551  *
1552  * return value 0 The command completed successfully.
1553  * return value -EFAULT An error occurred while copying the completion
1554  *                 data to the user space buffer.
1555  * return value -1 An error occurred while executing the command.
1556  */
1557 static int exec_drive_command(struct mtip_port *port, u8 *command,
1558 				void __user *user_buffer)
1559 {
1560 	struct host_to_dev_fis	fis;
1561 	struct host_to_dev_fis *reply;
1562 	u8 *buf = NULL;
1563 	dma_addr_t dma_addr = 0;
1564 	int rv = 0, xfer_sz = command[3];
1565 	unsigned int to;
1566 
1567 	if (xfer_sz) {
1568 		if (!user_buffer)
1569 			return -EFAULT;
1570 
1571 		buf = dma_alloc_coherent(&port->dd->pdev->dev,
1572 				ATA_SECT_SIZE * xfer_sz,
1573 				&dma_addr,
1574 				GFP_KERNEL);
1575 		if (!buf) {
1576 			dev_err(&port->dd->pdev->dev,
1577 				"Memory allocation failed (%d bytes)\n",
1578 				ATA_SECT_SIZE * xfer_sz);
1579 			return -ENOMEM;
1580 		}
1581 	}
1582 
1583 	/* Build the FIS. */
1584 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1585 	fis.type	= 0x27;
1586 	fis.opts	= 1 << 7;
1587 	fis.command	= command[0];
1588 	fis.features	= command[2];
1589 	fis.sect_count	= command[3];
1590 	if (fis.command == ATA_CMD_SMART) {
1591 		fis.sector	= command[1];
1592 		fis.cyl_low	= 0x4F;
1593 		fis.cyl_hi	= 0xC2;
1594 	}
1595 
1596 	mtip_set_timeout(port->dd, &fis, &to, 0);
1597 
1598 	if (xfer_sz)
1599 		reply = (port->rxfis + RX_FIS_PIO_SETUP);
1600 	else
1601 		reply = (port->rxfis + RX_FIS_D2H_REG);
1602 
1603 	dbg_printk(MTIP_DRV_NAME
1604 		" %s: User Command: cmd %x, sect %x, "
1605 		"feat %x, sectcnt %x\n",
1606 		__func__,
1607 		command[0],
1608 		command[1],
1609 		command[2],
1610 		command[3]);
1611 
1612 	/* Execute the command. */
1613 	if (mtip_exec_internal_command(port,
1614 				&fis,
1615 				 5,
1616 				 (xfer_sz ? dma_addr : 0),
1617 				 (xfer_sz ? ATA_SECT_SIZE * xfer_sz : 0),
1618 				 0,
1619 				 to)
1620 				 < 0) {
1621 		rv = -EFAULT;
1622 		goto exit_drive_command;
1623 	}
1624 
1625 	/* Collect the completion status. */
1626 	command[0] = reply->command; /* Status*/
1627 	command[1] = reply->features; /* Error*/
1628 	command[2] = reply->sect_count;
1629 
1630 	dbg_printk(MTIP_DRV_NAME
1631 		" %s: Completion Status: stat %x, "
1632 		"err %x, nsect %x\n",
1633 		__func__,
1634 		command[0],
1635 		command[1],
1636 		command[2]);
1637 
1638 	if (xfer_sz) {
1639 		if (copy_to_user(user_buffer,
1640 				 buf,
1641 				 ATA_SECT_SIZE * command[3])) {
1642 			rv = -EFAULT;
1643 			goto exit_drive_command;
1644 		}
1645 	}
1646 exit_drive_command:
1647 	if (buf)
1648 		dma_free_coherent(&port->dd->pdev->dev,
1649 				ATA_SECT_SIZE * xfer_sz, buf, dma_addr);
1650 	return rv;
1651 }
1652 
1653 /*
1654  *  Indicates whether a command has a single sector payload.
1655  *
1656  *  @command passed to the device to perform the certain event.
1657  *  @features passed to the device to perform the certain event.
1658  *
1659  *  return value
1660  *	1	command is one that always has a single sector payload,
1661  *		regardless of the value in the Sector Count field.
1662  *      0       otherwise
1663  *
1664  */
1665 static unsigned int implicit_sector(unsigned char command,
1666 				    unsigned char features)
1667 {
1668 	unsigned int rv = 0;
1669 
1670 	/* list of commands that have an implicit sector count of 1 */
1671 	switch (command) {
1672 	case ATA_CMD_SEC_SET_PASS:
1673 	case ATA_CMD_SEC_UNLOCK:
1674 	case ATA_CMD_SEC_ERASE_PREP:
1675 	case ATA_CMD_SEC_ERASE_UNIT:
1676 	case ATA_CMD_SEC_FREEZE_LOCK:
1677 	case ATA_CMD_SEC_DISABLE_PASS:
1678 	case ATA_CMD_PMP_READ:
1679 	case ATA_CMD_PMP_WRITE:
1680 		rv = 1;
1681 		break;
1682 	case ATA_CMD_SET_MAX:
1683 		if (features == ATA_SET_MAX_UNLOCK)
1684 			rv = 1;
1685 		break;
1686 	case ATA_CMD_SMART:
1687 		if ((features == ATA_SMART_READ_VALUES) ||
1688 				(features == ATA_SMART_READ_THRESHOLDS))
1689 			rv = 1;
1690 		break;
1691 	case ATA_CMD_CONF_OVERLAY:
1692 		if ((features == ATA_DCO_IDENTIFY) ||
1693 				(features == ATA_DCO_SET))
1694 			rv = 1;
1695 		break;
1696 	}
1697 	return rv;
1698 }
1699 
1700 /*
1701  * Executes a taskfile
1702  * See ide_taskfile_ioctl() for derivation
1703  */
1704 static int exec_drive_taskfile(struct driver_data *dd,
1705 			       void __user *buf,
1706 			       ide_task_request_t *req_task,
1707 			       int outtotal)
1708 {
1709 	struct host_to_dev_fis	fis;
1710 	struct host_to_dev_fis *reply;
1711 	u8 *outbuf = NULL;
1712 	u8 *inbuf = NULL;
1713 	dma_addr_t outbuf_dma = 0;
1714 	dma_addr_t inbuf_dma = 0;
1715 	dma_addr_t dma_buffer = 0;
1716 	int err = 0;
1717 	unsigned int taskin = 0;
1718 	unsigned int taskout = 0;
1719 	u8 nsect = 0;
1720 	unsigned int timeout;
1721 	unsigned int force_single_sector;
1722 	unsigned int transfer_size;
1723 	unsigned long task_file_data;
1724 	int intotal = outtotal + req_task->out_size;
1725 	int erasemode = 0;
1726 
1727 	taskout = req_task->out_size;
1728 	taskin = req_task->in_size;
1729 	/* 130560 = 512 * 0xFF*/
1730 	if (taskin > 130560 || taskout > 130560)
1731 		return -EINVAL;
1732 
1733 	if (taskout) {
1734 		outbuf = memdup_user(buf + outtotal, taskout);
1735 		if (IS_ERR(outbuf))
1736 			return PTR_ERR(outbuf);
1737 
1738 		outbuf_dma = dma_map_single(&dd->pdev->dev, outbuf,
1739 					    taskout, DMA_TO_DEVICE);
1740 		if (dma_mapping_error(&dd->pdev->dev, outbuf_dma)) {
1741 			err = -ENOMEM;
1742 			goto abort;
1743 		}
1744 		dma_buffer = outbuf_dma;
1745 	}
1746 
1747 	if (taskin) {
1748 		inbuf = memdup_user(buf + intotal, taskin);
1749 		if (IS_ERR(inbuf)) {
1750 			err = PTR_ERR(inbuf);
1751 			inbuf = NULL;
1752 			goto abort;
1753 		}
1754 		inbuf_dma = dma_map_single(&dd->pdev->dev, inbuf,
1755 					   taskin, DMA_FROM_DEVICE);
1756 		if (dma_mapping_error(&dd->pdev->dev, inbuf_dma)) {
1757 			err = -ENOMEM;
1758 			goto abort;
1759 		}
1760 		dma_buffer = inbuf_dma;
1761 	}
1762 
1763 	/* only supports PIO and non-data commands from this ioctl. */
1764 	switch (req_task->data_phase) {
1765 	case TASKFILE_OUT:
1766 		nsect = taskout / ATA_SECT_SIZE;
1767 		reply = (dd->port->rxfis + RX_FIS_PIO_SETUP);
1768 		break;
1769 	case TASKFILE_IN:
1770 		reply = (dd->port->rxfis + RX_FIS_PIO_SETUP);
1771 		break;
1772 	case TASKFILE_NO_DATA:
1773 		reply = (dd->port->rxfis + RX_FIS_D2H_REG);
1774 		break;
1775 	default:
1776 		err = -EINVAL;
1777 		goto abort;
1778 	}
1779 
1780 	/* Build the FIS. */
1781 	memset(&fis, 0, sizeof(struct host_to_dev_fis));
1782 
1783 	fis.type	= 0x27;
1784 	fis.opts	= 1 << 7;
1785 	fis.command	= req_task->io_ports[7];
1786 	fis.features	= req_task->io_ports[1];
1787 	fis.sect_count	= req_task->io_ports[2];
1788 	fis.lba_low	= req_task->io_ports[3];
1789 	fis.lba_mid	= req_task->io_ports[4];
1790 	fis.lba_hi	= req_task->io_ports[5];
1791 	 /* Clear the dev bit*/
1792 	fis.device	= req_task->io_ports[6] & ~0x10;
1793 
1794 	if ((req_task->in_flags.all == 0) && (req_task->out_flags.all & 1)) {
1795 		req_task->in_flags.all	=
1796 			IDE_TASKFILE_STD_IN_FLAGS |
1797 			(IDE_HOB_STD_IN_FLAGS << 8);
1798 		fis.lba_low_ex		= req_task->hob_ports[3];
1799 		fis.lba_mid_ex		= req_task->hob_ports[4];
1800 		fis.lba_hi_ex		= req_task->hob_ports[5];
1801 		fis.features_ex		= req_task->hob_ports[1];
1802 		fis.sect_cnt_ex		= req_task->hob_ports[2];
1803 
1804 	} else {
1805 		req_task->in_flags.all = IDE_TASKFILE_STD_IN_FLAGS;
1806 	}
1807 
1808 	force_single_sector = implicit_sector(fis.command, fis.features);
1809 
1810 	if ((taskin || taskout) && (!fis.sect_count)) {
1811 		if (nsect)
1812 			fis.sect_count = nsect;
1813 		else {
1814 			if (!force_single_sector) {
1815 				dev_warn(&dd->pdev->dev,
1816 					"data movement but "
1817 					"sect_count is 0\n");
1818 				err = -EINVAL;
1819 				goto abort;
1820 			}
1821 		}
1822 	}
1823 
1824 	dbg_printk(MTIP_DRV_NAME
1825 		" %s: cmd %x, feat %x, nsect %x,"
1826 		" sect/lbal %x, lcyl/lbam %x, hcyl/lbah %x,"
1827 		" head/dev %x\n",
1828 		__func__,
1829 		fis.command,
1830 		fis.features,
1831 		fis.sect_count,
1832 		fis.lba_low,
1833 		fis.lba_mid,
1834 		fis.lba_hi,
1835 		fis.device);
1836 
1837 	/* check for erase mode support during secure erase.*/
1838 	if ((fis.command == ATA_CMD_SEC_ERASE_UNIT) && outbuf &&
1839 					(outbuf[0] & MTIP_SEC_ERASE_MODE)) {
1840 		erasemode = 1;
1841 	}
1842 
1843 	mtip_set_timeout(dd, &fis, &timeout, erasemode);
1844 
1845 	/* Determine the correct transfer size.*/
1846 	if (force_single_sector)
1847 		transfer_size = ATA_SECT_SIZE;
1848 	else
1849 		transfer_size = ATA_SECT_SIZE * fis.sect_count;
1850 
1851 	/* Execute the command.*/
1852 	if (mtip_exec_internal_command(dd->port,
1853 				 &fis,
1854 				 5,
1855 				 dma_buffer,
1856 				 transfer_size,
1857 				 0,
1858 				 timeout) < 0) {
1859 		err = -EIO;
1860 		goto abort;
1861 	}
1862 
1863 	task_file_data = readl(dd->port->mmio+PORT_TFDATA);
1864 
1865 	if ((req_task->data_phase == TASKFILE_IN) && !(task_file_data & 1)) {
1866 		reply = dd->port->rxfis + RX_FIS_PIO_SETUP;
1867 		req_task->io_ports[7] = reply->control;
1868 	} else {
1869 		reply = dd->port->rxfis + RX_FIS_D2H_REG;
1870 		req_task->io_ports[7] = reply->command;
1871 	}
1872 
1873 	/* reclaim the DMA buffers.*/
1874 	if (inbuf_dma)
1875 		dma_unmap_single(&dd->pdev->dev, inbuf_dma, taskin,
1876 				 DMA_FROM_DEVICE);
1877 	if (outbuf_dma)
1878 		dma_unmap_single(&dd->pdev->dev, outbuf_dma, taskout,
1879 				 DMA_TO_DEVICE);
1880 	inbuf_dma  = 0;
1881 	outbuf_dma = 0;
1882 
1883 	/* return the ATA registers to the caller.*/
1884 	req_task->io_ports[1] = reply->features;
1885 	req_task->io_ports[2] = reply->sect_count;
1886 	req_task->io_ports[3] = reply->lba_low;
1887 	req_task->io_ports[4] = reply->lba_mid;
1888 	req_task->io_ports[5] = reply->lba_hi;
1889 	req_task->io_ports[6] = reply->device;
1890 
1891 	if (req_task->out_flags.all & 1)  {
1892 
1893 		req_task->hob_ports[3] = reply->lba_low_ex;
1894 		req_task->hob_ports[4] = reply->lba_mid_ex;
1895 		req_task->hob_ports[5] = reply->lba_hi_ex;
1896 		req_task->hob_ports[1] = reply->features_ex;
1897 		req_task->hob_ports[2] = reply->sect_cnt_ex;
1898 	}
1899 	dbg_printk(MTIP_DRV_NAME
1900 		" %s: Completion: stat %x,"
1901 		"err %x, sect_cnt %x, lbalo %x,"
1902 		"lbamid %x, lbahi %x, dev %x\n",
1903 		__func__,
1904 		req_task->io_ports[7],
1905 		req_task->io_ports[1],
1906 		req_task->io_ports[2],
1907 		req_task->io_ports[3],
1908 		req_task->io_ports[4],
1909 		req_task->io_ports[5],
1910 		req_task->io_ports[6]);
1911 
1912 	if (taskout) {
1913 		if (copy_to_user(buf + outtotal, outbuf, taskout)) {
1914 			err = -EFAULT;
1915 			goto abort;
1916 		}
1917 	}
1918 	if (taskin) {
1919 		if (copy_to_user(buf + intotal, inbuf, taskin)) {
1920 			err = -EFAULT;
1921 			goto abort;
1922 		}
1923 	}
1924 abort:
1925 	if (inbuf_dma)
1926 		dma_unmap_single(&dd->pdev->dev, inbuf_dma, taskin,
1927 				 DMA_FROM_DEVICE);
1928 	if (outbuf_dma)
1929 		dma_unmap_single(&dd->pdev->dev, outbuf_dma, taskout,
1930 				 DMA_TO_DEVICE);
1931 	kfree(outbuf);
1932 	kfree(inbuf);
1933 
1934 	return err;
1935 }
1936 
1937 /*
1938  * Handle IOCTL calls from the Block Layer.
1939  *
1940  * This function is called by the Block Layer when it receives an IOCTL
1941  * command that it does not understand. If the IOCTL command is not supported
1942  * this function returns -ENOTTY.
1943  *
1944  * @dd  Pointer to the driver data structure.
1945  * @cmd IOCTL command passed from the Block Layer.
1946  * @arg IOCTL argument passed from the Block Layer.
1947  *
1948  * return value
1949  *	0	The IOCTL completed successfully.
1950  *	-ENOTTY The specified command is not supported.
1951  *	-EFAULT An error occurred copying data to a user space buffer.
1952  *	-EIO	An error occurred while executing the command.
1953  */
1954 static int mtip_hw_ioctl(struct driver_data *dd, unsigned int cmd,
1955 			 unsigned long arg)
1956 {
1957 	switch (cmd) {
1958 	case HDIO_GET_IDENTITY:
1959 	{
1960 		if (copy_to_user((void __user *)arg, dd->port->identify,
1961 						sizeof(u16) * ATA_ID_WORDS))
1962 			return -EFAULT;
1963 		break;
1964 	}
1965 	case HDIO_DRIVE_CMD:
1966 	{
1967 		u8 drive_command[4];
1968 
1969 		/* Copy the user command info to our buffer. */
1970 		if (copy_from_user(drive_command,
1971 					 (void __user *) arg,
1972 					 sizeof(drive_command)))
1973 			return -EFAULT;
1974 
1975 		/* Execute the drive command. */
1976 		if (exec_drive_command(dd->port,
1977 					 drive_command,
1978 					 (void __user *) (arg+4)))
1979 			return -EIO;
1980 
1981 		/* Copy the status back to the users buffer. */
1982 		if (copy_to_user((void __user *) arg,
1983 					 drive_command,
1984 					 sizeof(drive_command)))
1985 			return -EFAULT;
1986 
1987 		break;
1988 	}
1989 	case HDIO_DRIVE_TASK:
1990 	{
1991 		u8 drive_command[7];
1992 
1993 		/* Copy the user command info to our buffer. */
1994 		if (copy_from_user(drive_command,
1995 					 (void __user *) arg,
1996 					 sizeof(drive_command)))
1997 			return -EFAULT;
1998 
1999 		/* Execute the drive command. */
2000 		if (exec_drive_task(dd->port, drive_command))
2001 			return -EIO;
2002 
2003 		/* Copy the status back to the users buffer. */
2004 		if (copy_to_user((void __user *) arg,
2005 					 drive_command,
2006 					 sizeof(drive_command)))
2007 			return -EFAULT;
2008 
2009 		break;
2010 	}
2011 	case HDIO_DRIVE_TASKFILE: {
2012 		ide_task_request_t req_task;
2013 		int ret, outtotal;
2014 
2015 		if (copy_from_user(&req_task, (void __user *) arg,
2016 					sizeof(req_task)))
2017 			return -EFAULT;
2018 
2019 		outtotal = sizeof(req_task);
2020 
2021 		ret = exec_drive_taskfile(dd, (void __user *) arg,
2022 						&req_task, outtotal);
2023 
2024 		if (copy_to_user((void __user *) arg, &req_task,
2025 							sizeof(req_task)))
2026 			return -EFAULT;
2027 
2028 		return ret;
2029 	}
2030 
2031 	default:
2032 		return -EINVAL;
2033 	}
2034 	return 0;
2035 }
2036 
2037 /*
2038  * Submit an IO to the hw
2039  *
2040  * This function is called by the block layer to issue an io
2041  * to the device. Upon completion, the callback function will
2042  * be called with the data parameter passed as the callback data.
2043  *
2044  * @dd       Pointer to the driver data structure.
2045  * @start    First sector to read.
2046  * @nsect    Number of sectors to read.
2047  * @tag      The tag of this read command.
2048  * @callback Pointer to the function that should be called
2049  *	     when the read completes.
2050  * @data     Callback data passed to the callback function
2051  *	     when the read completes.
2052  * @dir      Direction (read or write)
2053  *
2054  * return value
2055  *	None
2056  */
2057 static void mtip_hw_submit_io(struct driver_data *dd, struct request *rq,
2058 			      struct mtip_cmd *command,
2059 			      struct blk_mq_hw_ctx *hctx)
2060 {
2061 	struct mtip_cmd_hdr *hdr =
2062 		dd->port->command_list + sizeof(struct mtip_cmd_hdr) * rq->tag;
2063 	struct host_to_dev_fis	*fis;
2064 	struct mtip_port *port = dd->port;
2065 	int dma_dir = rq_data_dir(rq) == READ ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2066 	u64 start = blk_rq_pos(rq);
2067 	unsigned int nsect = blk_rq_sectors(rq);
2068 	unsigned int nents;
2069 
2070 	/* Map the scatter list for DMA access */
2071 	nents = blk_rq_map_sg(hctx->queue, rq, command->sg);
2072 	nents = dma_map_sg(&dd->pdev->dev, command->sg, nents, dma_dir);
2073 
2074 	prefetch(&port->flags);
2075 
2076 	command->scatter_ents = nents;
2077 
2078 	/*
2079 	 * The number of retries for this command before it is
2080 	 * reported as a failure to the upper layers.
2081 	 */
2082 	command->retries = MTIP_MAX_RETRIES;
2083 
2084 	/* Fill out fis */
2085 	fis = command->command;
2086 	fis->type        = 0x27;
2087 	fis->opts        = 1 << 7;
2088 	if (dma_dir == DMA_FROM_DEVICE)
2089 		fis->command = ATA_CMD_FPDMA_READ;
2090 	else
2091 		fis->command = ATA_CMD_FPDMA_WRITE;
2092 	fis->lba_low     = start & 0xFF;
2093 	fis->lba_mid     = (start >> 8) & 0xFF;
2094 	fis->lba_hi      = (start >> 16) & 0xFF;
2095 	fis->lba_low_ex  = (start >> 24) & 0xFF;
2096 	fis->lba_mid_ex  = (start >> 32) & 0xFF;
2097 	fis->lba_hi_ex   = (start >> 40) & 0xFF;
2098 	fis->device	 = 1 << 6;
2099 	fis->features    = nsect & 0xFF;
2100 	fis->features_ex = (nsect >> 8) & 0xFF;
2101 	fis->sect_count  = ((rq->tag << 3) | (rq->tag >> 5));
2102 	fis->sect_cnt_ex = 0;
2103 	fis->control     = 0;
2104 	fis->res2        = 0;
2105 	fis->res3        = 0;
2106 	fill_command_sg(dd, command, nents);
2107 
2108 	if (unlikely(command->unaligned))
2109 		fis->device |= 1 << 7;
2110 
2111 	/* Populate the command header */
2112 	hdr->ctba = cpu_to_le32(command->command_dma & 0xFFFFFFFF);
2113 	if (test_bit(MTIP_PF_HOST_CAP_64, &dd->port->flags))
2114 		hdr->ctbau = cpu_to_le32((command->command_dma >> 16) >> 16);
2115 	hdr->opts = cpu_to_le32((nents << 16) | 5 | AHCI_CMD_PREFETCH);
2116 	hdr->byte_count = 0;
2117 
2118 	command->direction = dma_dir;
2119 
2120 	/*
2121 	 * To prevent this command from being issued
2122 	 * if an internal command is in progress or error handling is active.
2123 	 */
2124 	if (unlikely(port->flags & MTIP_PF_PAUSE_IO)) {
2125 		set_bit(rq->tag, port->cmds_to_issue);
2126 		set_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags);
2127 		return;
2128 	}
2129 
2130 	/* Issue the command to the hardware */
2131 	mtip_issue_ncq_command(port, rq->tag);
2132 }
2133 
2134 /*
2135  * Sysfs status dump.
2136  *
2137  * @dev  Pointer to the device structure, passed by the kernrel.
2138  * @attr Pointer to the device_attribute structure passed by the kernel.
2139  * @buf  Pointer to the char buffer that will receive the stats info.
2140  *
2141  * return value
2142  *	The size, in bytes, of the data copied into buf.
2143  */
2144 static ssize_t mtip_hw_show_status(struct device *dev,
2145 				struct device_attribute *attr,
2146 				char *buf)
2147 {
2148 	struct driver_data *dd = dev_to_disk(dev)->private_data;
2149 	int size = 0;
2150 
2151 	if (test_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag))
2152 		size += sprintf(buf, "%s", "thermal_shutdown\n");
2153 	else if (test_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag))
2154 		size += sprintf(buf, "%s", "write_protect\n");
2155 	else
2156 		size += sprintf(buf, "%s", "online\n");
2157 
2158 	return size;
2159 }
2160 
2161 static DEVICE_ATTR(status, 0444, mtip_hw_show_status, NULL);
2162 
2163 static struct attribute *mtip_disk_attrs[] = {
2164 	&dev_attr_status.attr,
2165 	NULL,
2166 };
2167 
2168 static const struct attribute_group mtip_disk_attr_group = {
2169 	.attrs = mtip_disk_attrs,
2170 };
2171 
2172 static const struct attribute_group *mtip_disk_attr_groups[] = {
2173 	&mtip_disk_attr_group,
2174 	NULL,
2175 };
2176 
2177 /* debugsfs entries */
2178 
2179 static ssize_t show_device_status(struct device_driver *drv, char *buf)
2180 {
2181 	int size = 0;
2182 	struct driver_data *dd, *tmp;
2183 	unsigned long flags;
2184 	char id_buf[42];
2185 	u16 status = 0;
2186 
2187 	spin_lock_irqsave(&dev_lock, flags);
2188 	size += sprintf(&buf[size], "Devices Present:\n");
2189 	list_for_each_entry_safe(dd, tmp, &online_list, online_list) {
2190 		if (dd->pdev) {
2191 			if (dd->port &&
2192 			    dd->port->identify &&
2193 			    dd->port->identify_valid) {
2194 				strlcpy(id_buf,
2195 					(char *) (dd->port->identify + 10), 21);
2196 				status = *(dd->port->identify + 141);
2197 			} else {
2198 				memset(id_buf, 0, 42);
2199 				status = 0;
2200 			}
2201 
2202 			if (dd->port &&
2203 			    test_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags)) {
2204 				size += sprintf(&buf[size],
2205 					" device %s %s (ftl rebuild %d %%)\n",
2206 					dev_name(&dd->pdev->dev),
2207 					id_buf,
2208 					status);
2209 			} else {
2210 				size += sprintf(&buf[size],
2211 					" device %s %s\n",
2212 					dev_name(&dd->pdev->dev),
2213 					id_buf);
2214 			}
2215 		}
2216 	}
2217 
2218 	size += sprintf(&buf[size], "Devices Being Removed:\n");
2219 	list_for_each_entry_safe(dd, tmp, &removing_list, remove_list) {
2220 		if (dd->pdev) {
2221 			if (dd->port &&
2222 			    dd->port->identify &&
2223 			    dd->port->identify_valid) {
2224 				strlcpy(id_buf,
2225 					(char *) (dd->port->identify+10), 21);
2226 				status = *(dd->port->identify + 141);
2227 			} else {
2228 				memset(id_buf, 0, 42);
2229 				status = 0;
2230 			}
2231 
2232 			if (dd->port &&
2233 			    test_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags)) {
2234 				size += sprintf(&buf[size],
2235 					" device %s %s (ftl rebuild %d %%)\n",
2236 					dev_name(&dd->pdev->dev),
2237 					id_buf,
2238 					status);
2239 			} else {
2240 				size += sprintf(&buf[size],
2241 					" device %s %s\n",
2242 					dev_name(&dd->pdev->dev),
2243 					id_buf);
2244 			}
2245 		}
2246 	}
2247 	spin_unlock_irqrestore(&dev_lock, flags);
2248 
2249 	return size;
2250 }
2251 
2252 static ssize_t mtip_hw_read_device_status(struct file *f, char __user *ubuf,
2253 						size_t len, loff_t *offset)
2254 {
2255 	int size = *offset;
2256 	char *buf;
2257 	int rv = 0;
2258 
2259 	if (!len || *offset)
2260 		return 0;
2261 
2262 	buf = kzalloc(MTIP_DFS_MAX_BUF_SIZE, GFP_KERNEL);
2263 	if (!buf)
2264 		return -ENOMEM;
2265 
2266 	size += show_device_status(NULL, buf);
2267 
2268 	*offset = size <= len ? size : len;
2269 	size = copy_to_user(ubuf, buf, *offset);
2270 	if (size)
2271 		rv = -EFAULT;
2272 
2273 	kfree(buf);
2274 	return rv ? rv : *offset;
2275 }
2276 
2277 static ssize_t mtip_hw_read_registers(struct file *f, char __user *ubuf,
2278 				  size_t len, loff_t *offset)
2279 {
2280 	struct driver_data *dd =  (struct driver_data *)f->private_data;
2281 	char *buf;
2282 	u32 group_allocated;
2283 	int size = *offset;
2284 	int n, rv = 0;
2285 
2286 	if (!len || size)
2287 		return 0;
2288 
2289 	buf = kzalloc(MTIP_DFS_MAX_BUF_SIZE, GFP_KERNEL);
2290 	if (!buf)
2291 		return -ENOMEM;
2292 
2293 	size += sprintf(&buf[size], "H/ S ACTive      : [ 0x");
2294 
2295 	for (n = dd->slot_groups-1; n >= 0; n--)
2296 		size += sprintf(&buf[size], "%08X ",
2297 					 readl(dd->port->s_active[n]));
2298 
2299 	size += sprintf(&buf[size], "]\n");
2300 	size += sprintf(&buf[size], "H/ Command Issue : [ 0x");
2301 
2302 	for (n = dd->slot_groups-1; n >= 0; n--)
2303 		size += sprintf(&buf[size], "%08X ",
2304 					readl(dd->port->cmd_issue[n]));
2305 
2306 	size += sprintf(&buf[size], "]\n");
2307 	size += sprintf(&buf[size], "H/ Completed     : [ 0x");
2308 
2309 	for (n = dd->slot_groups-1; n >= 0; n--)
2310 		size += sprintf(&buf[size], "%08X ",
2311 				readl(dd->port->completed[n]));
2312 
2313 	size += sprintf(&buf[size], "]\n");
2314 	size += sprintf(&buf[size], "H/ PORT IRQ STAT : [ 0x%08X ]\n",
2315 				readl(dd->port->mmio + PORT_IRQ_STAT));
2316 	size += sprintf(&buf[size], "H/ HOST IRQ STAT : [ 0x%08X ]\n",
2317 				readl(dd->mmio + HOST_IRQ_STAT));
2318 	size += sprintf(&buf[size], "\n");
2319 
2320 	size += sprintf(&buf[size], "L/ Commands in Q : [ 0x");
2321 
2322 	for (n = dd->slot_groups-1; n >= 0; n--) {
2323 		if (sizeof(long) > sizeof(u32))
2324 			group_allocated =
2325 				dd->port->cmds_to_issue[n/2] >> (32*(n&1));
2326 		else
2327 			group_allocated = dd->port->cmds_to_issue[n];
2328 		size += sprintf(&buf[size], "%08X ", group_allocated);
2329 	}
2330 	size += sprintf(&buf[size], "]\n");
2331 
2332 	*offset = size <= len ? size : len;
2333 	size = copy_to_user(ubuf, buf, *offset);
2334 	if (size)
2335 		rv = -EFAULT;
2336 
2337 	kfree(buf);
2338 	return rv ? rv : *offset;
2339 }
2340 
2341 static ssize_t mtip_hw_read_flags(struct file *f, char __user *ubuf,
2342 				  size_t len, loff_t *offset)
2343 {
2344 	struct driver_data *dd =  (struct driver_data *)f->private_data;
2345 	char *buf;
2346 	int size = *offset;
2347 	int rv = 0;
2348 
2349 	if (!len || size)
2350 		return 0;
2351 
2352 	buf = kzalloc(MTIP_DFS_MAX_BUF_SIZE, GFP_KERNEL);
2353 	if (!buf)
2354 		return -ENOMEM;
2355 
2356 	size += sprintf(&buf[size], "Flag-port : [ %08lX ]\n",
2357 							dd->port->flags);
2358 	size += sprintf(&buf[size], "Flag-dd   : [ %08lX ]\n",
2359 							dd->dd_flag);
2360 
2361 	*offset = size <= len ? size : len;
2362 	size = copy_to_user(ubuf, buf, *offset);
2363 	if (size)
2364 		rv = -EFAULT;
2365 
2366 	kfree(buf);
2367 	return rv ? rv : *offset;
2368 }
2369 
2370 static const struct file_operations mtip_device_status_fops = {
2371 	.owner  = THIS_MODULE,
2372 	.open   = simple_open,
2373 	.read   = mtip_hw_read_device_status,
2374 	.llseek = no_llseek,
2375 };
2376 
2377 static const struct file_operations mtip_regs_fops = {
2378 	.owner  = THIS_MODULE,
2379 	.open   = simple_open,
2380 	.read   = mtip_hw_read_registers,
2381 	.llseek = no_llseek,
2382 };
2383 
2384 static const struct file_operations mtip_flags_fops = {
2385 	.owner  = THIS_MODULE,
2386 	.open   = simple_open,
2387 	.read   = mtip_hw_read_flags,
2388 	.llseek = no_llseek,
2389 };
2390 
2391 static int mtip_hw_debugfs_init(struct driver_data *dd)
2392 {
2393 	if (!dfs_parent)
2394 		return -1;
2395 
2396 	dd->dfs_node = debugfs_create_dir(dd->disk->disk_name, dfs_parent);
2397 	if (IS_ERR_OR_NULL(dd->dfs_node)) {
2398 		dev_warn(&dd->pdev->dev,
2399 			"Error creating node %s under debugfs\n",
2400 						dd->disk->disk_name);
2401 		dd->dfs_node = NULL;
2402 		return -1;
2403 	}
2404 
2405 	debugfs_create_file("flags", 0444, dd->dfs_node, dd, &mtip_flags_fops);
2406 	debugfs_create_file("registers", 0444, dd->dfs_node, dd,
2407 			    &mtip_regs_fops);
2408 
2409 	return 0;
2410 }
2411 
2412 static void mtip_hw_debugfs_exit(struct driver_data *dd)
2413 {
2414 	debugfs_remove_recursive(dd->dfs_node);
2415 }
2416 
2417 /*
2418  * Perform any init/resume time hardware setup
2419  *
2420  * @dd Pointer to the driver data structure.
2421  *
2422  * return value
2423  *	None
2424  */
2425 static inline void hba_setup(struct driver_data *dd)
2426 {
2427 	u32 hwdata;
2428 	hwdata = readl(dd->mmio + HOST_HSORG);
2429 
2430 	/* interrupt bug workaround: use only 1 IS bit.*/
2431 	writel(hwdata |
2432 		HSORG_DISABLE_SLOTGRP_INTR |
2433 		HSORG_DISABLE_SLOTGRP_PXIS,
2434 		dd->mmio + HOST_HSORG);
2435 }
2436 
2437 static int mtip_device_unaligned_constrained(struct driver_data *dd)
2438 {
2439 	return (dd->pdev->device == P420M_DEVICE_ID ? 1 : 0);
2440 }
2441 
2442 /*
2443  * Detect the details of the product, and store anything needed
2444  * into the driver data structure.  This includes product type and
2445  * version and number of slot groups.
2446  *
2447  * @dd Pointer to the driver data structure.
2448  *
2449  * return value
2450  *	None
2451  */
2452 static void mtip_detect_product(struct driver_data *dd)
2453 {
2454 	u32 hwdata;
2455 	unsigned int rev, slotgroups;
2456 
2457 	/*
2458 	 * HBA base + 0xFC [15:0] - vendor-specific hardware interface
2459 	 * info register:
2460 	 * [15:8] hardware/software interface rev#
2461 	 * [   3] asic-style interface
2462 	 * [ 2:0] number of slot groups, minus 1 (only valid for asic-style).
2463 	 */
2464 	hwdata = readl(dd->mmio + HOST_HSORG);
2465 
2466 	dd->product_type = MTIP_PRODUCT_UNKNOWN;
2467 	dd->slot_groups = 1;
2468 
2469 	if (hwdata & 0x8) {
2470 		dd->product_type = MTIP_PRODUCT_ASICFPGA;
2471 		rev = (hwdata & HSORG_HWREV) >> 8;
2472 		slotgroups = (hwdata & HSORG_SLOTGROUPS) + 1;
2473 		dev_info(&dd->pdev->dev,
2474 			"ASIC-FPGA design, HS rev 0x%x, "
2475 			"%i slot groups [%i slots]\n",
2476 			 rev,
2477 			 slotgroups,
2478 			 slotgroups * 32);
2479 
2480 		if (slotgroups > MTIP_MAX_SLOT_GROUPS) {
2481 			dev_warn(&dd->pdev->dev,
2482 				"Warning: driver only supports "
2483 				"%i slot groups.\n", MTIP_MAX_SLOT_GROUPS);
2484 			slotgroups = MTIP_MAX_SLOT_GROUPS;
2485 		}
2486 		dd->slot_groups = slotgroups;
2487 		return;
2488 	}
2489 
2490 	dev_warn(&dd->pdev->dev, "Unrecognized product id\n");
2491 }
2492 
2493 /*
2494  * Blocking wait for FTL rebuild to complete
2495  *
2496  * @dd Pointer to the DRIVER_DATA structure.
2497  *
2498  * return value
2499  *	0	FTL rebuild completed successfully
2500  *	-EFAULT FTL rebuild error/timeout/interruption
2501  */
2502 static int mtip_ftl_rebuild_poll(struct driver_data *dd)
2503 {
2504 	unsigned long timeout, cnt = 0, start;
2505 
2506 	dev_warn(&dd->pdev->dev,
2507 		"FTL rebuild in progress. Polling for completion.\n");
2508 
2509 	start = jiffies;
2510 	timeout = jiffies + msecs_to_jiffies(MTIP_FTL_REBUILD_TIMEOUT_MS);
2511 
2512 	do {
2513 		if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
2514 				&dd->dd_flag)))
2515 			return -EFAULT;
2516 		if (mtip_check_surprise_removal(dd->pdev))
2517 			return -EFAULT;
2518 
2519 		if (mtip_get_identify(dd->port, NULL) < 0)
2520 			return -EFAULT;
2521 
2522 		if (*(dd->port->identify + MTIP_FTL_REBUILD_OFFSET) ==
2523 			MTIP_FTL_REBUILD_MAGIC) {
2524 			ssleep(1);
2525 			/* Print message every 3 minutes */
2526 			if (cnt++ >= 180) {
2527 				dev_warn(&dd->pdev->dev,
2528 				"FTL rebuild in progress (%d secs).\n",
2529 				jiffies_to_msecs(jiffies - start) / 1000);
2530 				cnt = 0;
2531 			}
2532 		} else {
2533 			dev_warn(&dd->pdev->dev,
2534 				"FTL rebuild complete (%d secs).\n",
2535 			jiffies_to_msecs(jiffies - start) / 1000);
2536 			mtip_block_initialize(dd);
2537 			return 0;
2538 		}
2539 	} while (time_before(jiffies, timeout));
2540 
2541 	/* Check for timeout */
2542 	dev_err(&dd->pdev->dev,
2543 		"Timed out waiting for FTL rebuild to complete (%d secs).\n",
2544 		jiffies_to_msecs(jiffies - start) / 1000);
2545 	return -EFAULT;
2546 }
2547 
2548 static void mtip_softirq_done_fn(struct request *rq)
2549 {
2550 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
2551 	struct driver_data *dd = rq->q->queuedata;
2552 
2553 	/* Unmap the DMA scatter list entries */
2554 	dma_unmap_sg(&dd->pdev->dev, cmd->sg, cmd->scatter_ents,
2555 							cmd->direction);
2556 
2557 	if (unlikely(cmd->unaligned))
2558 		atomic_inc(&dd->port->cmd_slot_unal);
2559 
2560 	blk_mq_end_request(rq, cmd->status);
2561 }
2562 
2563 static bool mtip_abort_cmd(struct request *req, void *data, bool reserved)
2564 {
2565 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(req);
2566 	struct driver_data *dd = data;
2567 
2568 	dbg_printk(MTIP_DRV_NAME " Aborting request, tag = %d\n", req->tag);
2569 
2570 	clear_bit(req->tag, dd->port->cmds_to_issue);
2571 	cmd->status = BLK_STS_IOERR;
2572 	mtip_softirq_done_fn(req);
2573 	return true;
2574 }
2575 
2576 static bool mtip_queue_cmd(struct request *req, void *data, bool reserved)
2577 {
2578 	struct driver_data *dd = data;
2579 
2580 	set_bit(req->tag, dd->port->cmds_to_issue);
2581 	blk_abort_request(req);
2582 	return true;
2583 }
2584 
2585 /*
2586  * service thread to issue queued commands
2587  *
2588  * @data Pointer to the driver data structure.
2589  *
2590  * return value
2591  *	0
2592  */
2593 
2594 static int mtip_service_thread(void *data)
2595 {
2596 	struct driver_data *dd = (struct driver_data *)data;
2597 	unsigned long slot, slot_start, slot_wrap, to;
2598 	unsigned int num_cmd_slots = dd->slot_groups * 32;
2599 	struct mtip_port *port = dd->port;
2600 
2601 	while (1) {
2602 		if (kthread_should_stop() ||
2603 			test_bit(MTIP_PF_SVC_THD_STOP_BIT, &port->flags))
2604 			goto st_out;
2605 		clear_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags);
2606 
2607 		/*
2608 		 * the condition is to check neither an internal command is
2609 		 * is in progress nor error handling is active
2610 		 */
2611 		wait_event_interruptible(port->svc_wait, (port->flags) &&
2612 			(port->flags & MTIP_PF_SVC_THD_WORK));
2613 
2614 		if (kthread_should_stop() ||
2615 			test_bit(MTIP_PF_SVC_THD_STOP_BIT, &port->flags))
2616 			goto st_out;
2617 
2618 		if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
2619 				&dd->dd_flag)))
2620 			goto st_out;
2621 
2622 		set_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags);
2623 
2624 restart_eh:
2625 		/* Demux bits: start with error handling */
2626 		if (test_bit(MTIP_PF_EH_ACTIVE_BIT, &port->flags)) {
2627 			mtip_handle_tfe(dd);
2628 			clear_bit(MTIP_PF_EH_ACTIVE_BIT, &port->flags);
2629 		}
2630 
2631 		if (test_bit(MTIP_PF_EH_ACTIVE_BIT, &port->flags))
2632 			goto restart_eh;
2633 
2634 		if (test_bit(MTIP_PF_TO_ACTIVE_BIT, &port->flags)) {
2635 			to = jiffies + msecs_to_jiffies(5000);
2636 
2637 			do {
2638 				mdelay(100);
2639 			} while (atomic_read(&dd->irq_workers_active) != 0 &&
2640 				time_before(jiffies, to));
2641 
2642 			if (atomic_read(&dd->irq_workers_active) != 0)
2643 				dev_warn(&dd->pdev->dev,
2644 					"Completion workers still active!");
2645 
2646 			blk_mq_quiesce_queue(dd->queue);
2647 
2648 			blk_mq_tagset_busy_iter(&dd->tags, mtip_queue_cmd, dd);
2649 
2650 			set_bit(MTIP_PF_ISSUE_CMDS_BIT, &dd->port->flags);
2651 
2652 			if (mtip_device_reset(dd))
2653 				blk_mq_tagset_busy_iter(&dd->tags,
2654 							mtip_abort_cmd, dd);
2655 
2656 			clear_bit(MTIP_PF_TO_ACTIVE_BIT, &dd->port->flags);
2657 
2658 			blk_mq_unquiesce_queue(dd->queue);
2659 		}
2660 
2661 		if (test_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags)) {
2662 			slot = 1;
2663 			/* used to restrict the loop to one iteration */
2664 			slot_start = num_cmd_slots;
2665 			slot_wrap = 0;
2666 			while (1) {
2667 				slot = find_next_bit(port->cmds_to_issue,
2668 						num_cmd_slots, slot);
2669 				if (slot_wrap == 1) {
2670 					if ((slot_start >= slot) ||
2671 						(slot >= num_cmd_slots))
2672 						break;
2673 				}
2674 				if (unlikely(slot_start == num_cmd_slots))
2675 					slot_start = slot;
2676 
2677 				if (unlikely(slot == num_cmd_slots)) {
2678 					slot = 1;
2679 					slot_wrap = 1;
2680 					continue;
2681 				}
2682 
2683 				/* Issue the command to the hardware */
2684 				mtip_issue_ncq_command(port, slot);
2685 
2686 				clear_bit(slot, port->cmds_to_issue);
2687 			}
2688 
2689 			clear_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags);
2690 		}
2691 
2692 		if (test_bit(MTIP_PF_REBUILD_BIT, &port->flags)) {
2693 			if (mtip_ftl_rebuild_poll(dd) == 0)
2694 				clear_bit(MTIP_PF_REBUILD_BIT, &port->flags);
2695 		}
2696 	}
2697 
2698 st_out:
2699 	return 0;
2700 }
2701 
2702 /*
2703  * DMA region teardown
2704  *
2705  * @dd Pointer to driver_data structure
2706  *
2707  * return value
2708  *      None
2709  */
2710 static void mtip_dma_free(struct driver_data *dd)
2711 {
2712 	struct mtip_port *port = dd->port;
2713 
2714 	if (port->block1)
2715 		dma_free_coherent(&dd->pdev->dev, BLOCK_DMA_ALLOC_SZ,
2716 					port->block1, port->block1_dma);
2717 
2718 	if (port->command_list) {
2719 		dma_free_coherent(&dd->pdev->dev, AHCI_CMD_TBL_SZ,
2720 				port->command_list, port->command_list_dma);
2721 	}
2722 }
2723 
2724 /*
2725  * DMA region setup
2726  *
2727  * @dd Pointer to driver_data structure
2728  *
2729  * return value
2730  *      -ENOMEM Not enough free DMA region space to initialize driver
2731  */
2732 static int mtip_dma_alloc(struct driver_data *dd)
2733 {
2734 	struct mtip_port *port = dd->port;
2735 
2736 	/* Allocate dma memory for RX Fis, Identify, and Sector Bufffer */
2737 	port->block1 =
2738 		dma_alloc_coherent(&dd->pdev->dev, BLOCK_DMA_ALLOC_SZ,
2739 					&port->block1_dma, GFP_KERNEL);
2740 	if (!port->block1)
2741 		return -ENOMEM;
2742 
2743 	/* Allocate dma memory for command list */
2744 	port->command_list =
2745 		dma_alloc_coherent(&dd->pdev->dev, AHCI_CMD_TBL_SZ,
2746 					&port->command_list_dma, GFP_KERNEL);
2747 	if (!port->command_list) {
2748 		dma_free_coherent(&dd->pdev->dev, BLOCK_DMA_ALLOC_SZ,
2749 					port->block1, port->block1_dma);
2750 		port->block1 = NULL;
2751 		port->block1_dma = 0;
2752 		return -ENOMEM;
2753 	}
2754 
2755 	/* Setup all pointers into first DMA region */
2756 	port->rxfis         = port->block1 + AHCI_RX_FIS_OFFSET;
2757 	port->rxfis_dma     = port->block1_dma + AHCI_RX_FIS_OFFSET;
2758 	port->identify      = port->block1 + AHCI_IDFY_OFFSET;
2759 	port->identify_dma  = port->block1_dma + AHCI_IDFY_OFFSET;
2760 	port->log_buf       = port->block1 + AHCI_SECTBUF_OFFSET;
2761 	port->log_buf_dma   = port->block1_dma + AHCI_SECTBUF_OFFSET;
2762 	port->smart_buf     = port->block1 + AHCI_SMARTBUF_OFFSET;
2763 	port->smart_buf_dma = port->block1_dma + AHCI_SMARTBUF_OFFSET;
2764 
2765 	return 0;
2766 }
2767 
2768 static int mtip_hw_get_identify(struct driver_data *dd)
2769 {
2770 	struct smart_attr attr242;
2771 	unsigned char *buf;
2772 	int rv;
2773 
2774 	if (mtip_get_identify(dd->port, NULL) < 0)
2775 		return -EFAULT;
2776 
2777 	if (*(dd->port->identify + MTIP_FTL_REBUILD_OFFSET) ==
2778 		MTIP_FTL_REBUILD_MAGIC) {
2779 		set_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags);
2780 		return MTIP_FTL_REBUILD_MAGIC;
2781 	}
2782 	mtip_dump_identify(dd->port);
2783 
2784 	/* check write protect, over temp and rebuild statuses */
2785 	rv = mtip_read_log_page(dd->port, ATA_LOG_SATA_NCQ,
2786 				dd->port->log_buf,
2787 				dd->port->log_buf_dma, 1);
2788 	if (rv) {
2789 		dev_warn(&dd->pdev->dev,
2790 			"Error in READ LOG EXT (10h) command\n");
2791 		/* non-critical error, don't fail the load */
2792 	} else {
2793 		buf = (unsigned char *)dd->port->log_buf;
2794 		if (buf[259] & 0x1) {
2795 			dev_info(&dd->pdev->dev,
2796 				"Write protect bit is set.\n");
2797 			set_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag);
2798 		}
2799 		if (buf[288] == 0xF7) {
2800 			dev_info(&dd->pdev->dev,
2801 				"Exceeded Tmax, drive in thermal shutdown.\n");
2802 			set_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag);
2803 		}
2804 		if (buf[288] == 0xBF) {
2805 			dev_info(&dd->pdev->dev,
2806 				"Drive indicates rebuild has failed.\n");
2807 			set_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag);
2808 		}
2809 	}
2810 
2811 	/* get write protect progess */
2812 	memset(&attr242, 0, sizeof(struct smart_attr));
2813 	if (mtip_get_smart_attr(dd->port, 242, &attr242))
2814 		dev_warn(&dd->pdev->dev,
2815 				"Unable to check write protect progress\n");
2816 	else
2817 		dev_info(&dd->pdev->dev,
2818 				"Write protect progress: %u%% (%u blocks)\n",
2819 				attr242.cur, le32_to_cpu(attr242.data));
2820 
2821 	return rv;
2822 }
2823 
2824 /*
2825  * Called once for each card.
2826  *
2827  * @dd Pointer to the driver data structure.
2828  *
2829  * return value
2830  *	0 on success, else an error code.
2831  */
2832 static int mtip_hw_init(struct driver_data *dd)
2833 {
2834 	int i;
2835 	int rv;
2836 	unsigned long timeout, timetaken;
2837 
2838 	dd->mmio = pcim_iomap_table(dd->pdev)[MTIP_ABAR];
2839 
2840 	mtip_detect_product(dd);
2841 	if (dd->product_type == MTIP_PRODUCT_UNKNOWN) {
2842 		rv = -EIO;
2843 		goto out1;
2844 	}
2845 
2846 	hba_setup(dd);
2847 
2848 	dd->port = kzalloc_node(sizeof(struct mtip_port), GFP_KERNEL,
2849 				dd->numa_node);
2850 	if (!dd->port)
2851 		return -ENOMEM;
2852 
2853 	/* Continue workqueue setup */
2854 	for (i = 0; i < MTIP_MAX_SLOT_GROUPS; i++)
2855 		dd->work[i].port = dd->port;
2856 
2857 	/* Enable unaligned IO constraints for some devices */
2858 	if (mtip_device_unaligned_constrained(dd))
2859 		dd->unal_qdepth = MTIP_MAX_UNALIGNED_SLOTS;
2860 	else
2861 		dd->unal_qdepth = 0;
2862 
2863 	atomic_set(&dd->port->cmd_slot_unal, dd->unal_qdepth);
2864 
2865 	/* Spinlock to prevent concurrent issue */
2866 	for (i = 0; i < MTIP_MAX_SLOT_GROUPS; i++)
2867 		spin_lock_init(&dd->port->cmd_issue_lock[i]);
2868 
2869 	/* Set the port mmio base address. */
2870 	dd->port->mmio	= dd->mmio + PORT_OFFSET;
2871 	dd->port->dd	= dd;
2872 
2873 	/* DMA allocations */
2874 	rv = mtip_dma_alloc(dd);
2875 	if (rv < 0)
2876 		goto out1;
2877 
2878 	/* Setup the pointers to the extended s_active and CI registers. */
2879 	for (i = 0; i < dd->slot_groups; i++) {
2880 		dd->port->s_active[i] =
2881 			dd->port->mmio + i*0x80 + PORT_SCR_ACT;
2882 		dd->port->cmd_issue[i] =
2883 			dd->port->mmio + i*0x80 + PORT_COMMAND_ISSUE;
2884 		dd->port->completed[i] =
2885 			dd->port->mmio + i*0x80 + PORT_SDBV;
2886 	}
2887 
2888 	timetaken = jiffies;
2889 	timeout = jiffies + msecs_to_jiffies(30000);
2890 	while (((readl(dd->port->mmio + PORT_SCR_STAT) & 0x0F) != 0x03) &&
2891 		 time_before(jiffies, timeout)) {
2892 		mdelay(100);
2893 	}
2894 	if (unlikely(mtip_check_surprise_removal(dd->pdev))) {
2895 		timetaken = jiffies - timetaken;
2896 		dev_warn(&dd->pdev->dev,
2897 			"Surprise removal detected at %u ms\n",
2898 			jiffies_to_msecs(timetaken));
2899 		rv = -ENODEV;
2900 		goto out2 ;
2901 	}
2902 	if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag))) {
2903 		timetaken = jiffies - timetaken;
2904 		dev_warn(&dd->pdev->dev,
2905 			"Removal detected at %u ms\n",
2906 			jiffies_to_msecs(timetaken));
2907 		rv = -EFAULT;
2908 		goto out2;
2909 	}
2910 
2911 	/* Conditionally reset the HBA. */
2912 	if (!(readl(dd->mmio + HOST_CAP) & HOST_CAP_NZDMA)) {
2913 		if (mtip_hba_reset(dd) < 0) {
2914 			dev_err(&dd->pdev->dev,
2915 				"Card did not reset within timeout\n");
2916 			rv = -EIO;
2917 			goto out2;
2918 		}
2919 	} else {
2920 		/* Clear any pending interrupts on the HBA */
2921 		writel(readl(dd->mmio + HOST_IRQ_STAT),
2922 			dd->mmio + HOST_IRQ_STAT);
2923 	}
2924 
2925 	mtip_init_port(dd->port);
2926 	mtip_start_port(dd->port);
2927 
2928 	/* Setup the ISR and enable interrupts. */
2929 	rv = request_irq(dd->pdev->irq, mtip_irq_handler, IRQF_SHARED,
2930 			 dev_driver_string(&dd->pdev->dev), dd);
2931 	if (rv) {
2932 		dev_err(&dd->pdev->dev,
2933 			"Unable to allocate IRQ %d\n", dd->pdev->irq);
2934 		goto out2;
2935 	}
2936 	irq_set_affinity_hint(dd->pdev->irq, get_cpu_mask(dd->isr_binding));
2937 
2938 	/* Enable interrupts on the HBA. */
2939 	writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN,
2940 					dd->mmio + HOST_CTL);
2941 
2942 	init_waitqueue_head(&dd->port->svc_wait);
2943 
2944 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)) {
2945 		rv = -EFAULT;
2946 		goto out3;
2947 	}
2948 
2949 	return rv;
2950 
2951 out3:
2952 	/* Disable interrupts on the HBA. */
2953 	writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN,
2954 			dd->mmio + HOST_CTL);
2955 
2956 	/* Release the IRQ. */
2957 	irq_set_affinity_hint(dd->pdev->irq, NULL);
2958 	free_irq(dd->pdev->irq, dd);
2959 
2960 out2:
2961 	mtip_deinit_port(dd->port);
2962 	mtip_dma_free(dd);
2963 
2964 out1:
2965 	/* Free the memory allocated for the for structure. */
2966 	kfree(dd->port);
2967 
2968 	return rv;
2969 }
2970 
2971 static int mtip_standby_drive(struct driver_data *dd)
2972 {
2973 	int rv = 0;
2974 
2975 	if (dd->sr || !dd->port)
2976 		return -ENODEV;
2977 	/*
2978 	 * Send standby immediate (E0h) to the drive so that it
2979 	 * saves its state.
2980 	 */
2981 	if (!test_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags) &&
2982 	    !test_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag) &&
2983 	    !test_bit(MTIP_DDF_SEC_LOCK_BIT, &dd->dd_flag)) {
2984 		rv = mtip_standby_immediate(dd->port);
2985 		if (rv)
2986 			dev_warn(&dd->pdev->dev,
2987 				"STANDBY IMMEDIATE failed\n");
2988 	}
2989 	return rv;
2990 }
2991 
2992 /*
2993  * Called to deinitialize an interface.
2994  *
2995  * @dd Pointer to the driver data structure.
2996  *
2997  * return value
2998  *	0
2999  */
3000 static int mtip_hw_exit(struct driver_data *dd)
3001 {
3002 	if (!dd->sr) {
3003 		/* de-initialize the port. */
3004 		mtip_deinit_port(dd->port);
3005 
3006 		/* Disable interrupts on the HBA. */
3007 		writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN,
3008 				dd->mmio + HOST_CTL);
3009 	}
3010 
3011 	/* Release the IRQ. */
3012 	irq_set_affinity_hint(dd->pdev->irq, NULL);
3013 	free_irq(dd->pdev->irq, dd);
3014 	msleep(1000);
3015 
3016 	/* Free dma regions */
3017 	mtip_dma_free(dd);
3018 
3019 	/* Free the memory allocated for the for structure. */
3020 	kfree(dd->port);
3021 	dd->port = NULL;
3022 
3023 	return 0;
3024 }
3025 
3026 /*
3027  * Issue a Standby Immediate command to the device.
3028  *
3029  * This function is called by the Block Layer just before the
3030  * system powers off during a shutdown.
3031  *
3032  * @dd Pointer to the driver data structure.
3033  *
3034  * return value
3035  *	0
3036  */
3037 static int mtip_hw_shutdown(struct driver_data *dd)
3038 {
3039 	/*
3040 	 * Send standby immediate (E0h) to the drive so that it
3041 	 * saves its state.
3042 	 */
3043 	mtip_standby_drive(dd);
3044 
3045 	return 0;
3046 }
3047 
3048 /*
3049  * Suspend function
3050  *
3051  * This function is called by the Block Layer just before the
3052  * system hibernates.
3053  *
3054  * @dd Pointer to the driver data structure.
3055  *
3056  * return value
3057  *	0	Suspend was successful
3058  *	-EFAULT Suspend was not successful
3059  */
3060 static int mtip_hw_suspend(struct driver_data *dd)
3061 {
3062 	/*
3063 	 * Send standby immediate (E0h) to the drive
3064 	 * so that it saves its state.
3065 	 */
3066 	if (mtip_standby_drive(dd) != 0) {
3067 		dev_err(&dd->pdev->dev,
3068 			"Failed standby-immediate command\n");
3069 		return -EFAULT;
3070 	}
3071 
3072 	/* Disable interrupts on the HBA.*/
3073 	writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN,
3074 			dd->mmio + HOST_CTL);
3075 	mtip_deinit_port(dd->port);
3076 
3077 	return 0;
3078 }
3079 
3080 /*
3081  * Resume function
3082  *
3083  * This function is called by the Block Layer as the
3084  * system resumes.
3085  *
3086  * @dd Pointer to the driver data structure.
3087  *
3088  * return value
3089  *	0	Resume was successful
3090  *      -EFAULT Resume was not successful
3091  */
3092 static int mtip_hw_resume(struct driver_data *dd)
3093 {
3094 	/* Perform any needed hardware setup steps */
3095 	hba_setup(dd);
3096 
3097 	/* Reset the HBA */
3098 	if (mtip_hba_reset(dd) != 0) {
3099 		dev_err(&dd->pdev->dev,
3100 			"Unable to reset the HBA\n");
3101 		return -EFAULT;
3102 	}
3103 
3104 	/*
3105 	 * Enable the port, DMA engine, and FIS reception specific
3106 	 * h/w in controller.
3107 	 */
3108 	mtip_init_port(dd->port);
3109 	mtip_start_port(dd->port);
3110 
3111 	/* Enable interrupts on the HBA.*/
3112 	writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN,
3113 			dd->mmio + HOST_CTL);
3114 
3115 	return 0;
3116 }
3117 
3118 /*
3119  * Helper function for reusing disk name
3120  * upon hot insertion.
3121  */
3122 static int rssd_disk_name_format(char *prefix,
3123 				 int index,
3124 				 char *buf,
3125 				 int buflen)
3126 {
3127 	const int base = 'z' - 'a' + 1;
3128 	char *begin = buf + strlen(prefix);
3129 	char *end = buf + buflen;
3130 	char *p;
3131 	int unit;
3132 
3133 	p = end - 1;
3134 	*p = '\0';
3135 	unit = base;
3136 	do {
3137 		if (p == begin)
3138 			return -EINVAL;
3139 		*--p = 'a' + (index % unit);
3140 		index = (index / unit) - 1;
3141 	} while (index >= 0);
3142 
3143 	memmove(begin, p, end - p);
3144 	memcpy(buf, prefix, strlen(prefix));
3145 
3146 	return 0;
3147 }
3148 
3149 /*
3150  * Block layer IOCTL handler.
3151  *
3152  * @dev Pointer to the block_device structure.
3153  * @mode ignored
3154  * @cmd IOCTL command passed from the user application.
3155  * @arg Argument passed from the user application.
3156  *
3157  * return value
3158  *	0        IOCTL completed successfully.
3159  *	-ENOTTY  IOCTL not supported or invalid driver data
3160  *                 structure pointer.
3161  */
3162 static int mtip_block_ioctl(struct block_device *dev,
3163 			    fmode_t mode,
3164 			    unsigned cmd,
3165 			    unsigned long arg)
3166 {
3167 	struct driver_data *dd = dev->bd_disk->private_data;
3168 
3169 	if (!capable(CAP_SYS_ADMIN))
3170 		return -EACCES;
3171 
3172 	if (!dd)
3173 		return -ENOTTY;
3174 
3175 	if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)))
3176 		return -ENOTTY;
3177 
3178 	switch (cmd) {
3179 	case BLKFLSBUF:
3180 		return -ENOTTY;
3181 	default:
3182 		return mtip_hw_ioctl(dd, cmd, arg);
3183 	}
3184 }
3185 
3186 #ifdef CONFIG_COMPAT
3187 /*
3188  * Block layer compat IOCTL handler.
3189  *
3190  * @dev Pointer to the block_device structure.
3191  * @mode ignored
3192  * @cmd IOCTL command passed from the user application.
3193  * @arg Argument passed from the user application.
3194  *
3195  * return value
3196  *	0        IOCTL completed successfully.
3197  *	-ENOTTY  IOCTL not supported or invalid driver data
3198  *                 structure pointer.
3199  */
3200 static int mtip_block_compat_ioctl(struct block_device *dev,
3201 			    fmode_t mode,
3202 			    unsigned cmd,
3203 			    unsigned long arg)
3204 {
3205 	struct driver_data *dd = dev->bd_disk->private_data;
3206 
3207 	if (!capable(CAP_SYS_ADMIN))
3208 		return -EACCES;
3209 
3210 	if (!dd)
3211 		return -ENOTTY;
3212 
3213 	if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)))
3214 		return -ENOTTY;
3215 
3216 	switch (cmd) {
3217 	case BLKFLSBUF:
3218 		return -ENOTTY;
3219 	case HDIO_DRIVE_TASKFILE: {
3220 		struct mtip_compat_ide_task_request_s __user *compat_req_task;
3221 		ide_task_request_t req_task;
3222 		int compat_tasksize, outtotal, ret;
3223 
3224 		compat_tasksize =
3225 			sizeof(struct mtip_compat_ide_task_request_s);
3226 
3227 		compat_req_task =
3228 			(struct mtip_compat_ide_task_request_s __user *) arg;
3229 
3230 		if (copy_from_user(&req_task, (void __user *) arg,
3231 			compat_tasksize - (2 * sizeof(compat_long_t))))
3232 			return -EFAULT;
3233 
3234 		if (get_user(req_task.out_size, &compat_req_task->out_size))
3235 			return -EFAULT;
3236 
3237 		if (get_user(req_task.in_size, &compat_req_task->in_size))
3238 			return -EFAULT;
3239 
3240 		outtotal = sizeof(struct mtip_compat_ide_task_request_s);
3241 
3242 		ret = exec_drive_taskfile(dd, (void __user *) arg,
3243 						&req_task, outtotal);
3244 
3245 		if (copy_to_user((void __user *) arg, &req_task,
3246 				compat_tasksize -
3247 				(2 * sizeof(compat_long_t))))
3248 			return -EFAULT;
3249 
3250 		if (put_user(req_task.out_size, &compat_req_task->out_size))
3251 			return -EFAULT;
3252 
3253 		if (put_user(req_task.in_size, &compat_req_task->in_size))
3254 			return -EFAULT;
3255 
3256 		return ret;
3257 	}
3258 	default:
3259 		return mtip_hw_ioctl(dd, cmd, arg);
3260 	}
3261 }
3262 #endif
3263 
3264 /*
3265  * Obtain the geometry of the device.
3266  *
3267  * You may think that this function is obsolete, but some applications,
3268  * fdisk for example still used CHS values. This function describes the
3269  * device as having 224 heads and 56 sectors per cylinder. These values are
3270  * chosen so that each cylinder is aligned on a 4KB boundary. Since a
3271  * partition is described in terms of a start and end cylinder this means
3272  * that each partition is also 4KB aligned. Non-aligned partitions adversely
3273  * affects performance.
3274  *
3275  * @dev Pointer to the block_device strucutre.
3276  * @geo Pointer to a hd_geometry structure.
3277  *
3278  * return value
3279  *	0       Operation completed successfully.
3280  *	-ENOTTY An error occurred while reading the drive capacity.
3281  */
3282 static int mtip_block_getgeo(struct block_device *dev,
3283 				struct hd_geometry *geo)
3284 {
3285 	struct driver_data *dd = dev->bd_disk->private_data;
3286 	sector_t capacity;
3287 
3288 	if (!dd)
3289 		return -ENOTTY;
3290 
3291 	if (!(mtip_hw_get_capacity(dd, &capacity))) {
3292 		dev_warn(&dd->pdev->dev,
3293 			"Could not get drive capacity.\n");
3294 		return -ENOTTY;
3295 	}
3296 
3297 	geo->heads = 224;
3298 	geo->sectors = 56;
3299 	sector_div(capacity, (geo->heads * geo->sectors));
3300 	geo->cylinders = capacity;
3301 	return 0;
3302 }
3303 
3304 static int mtip_block_open(struct block_device *dev, fmode_t mode)
3305 {
3306 	struct driver_data *dd;
3307 
3308 	if (dev && dev->bd_disk) {
3309 		dd = (struct driver_data *) dev->bd_disk->private_data;
3310 
3311 		if (dd) {
3312 			if (test_bit(MTIP_DDF_REMOVAL_BIT,
3313 							&dd->dd_flag)) {
3314 				return -ENODEV;
3315 			}
3316 			return 0;
3317 		}
3318 	}
3319 	return -ENODEV;
3320 }
3321 
3322 static void mtip_block_release(struct gendisk *disk, fmode_t mode)
3323 {
3324 }
3325 
3326 /*
3327  * Block device operation function.
3328  *
3329  * This structure contains pointers to the functions required by the block
3330  * layer.
3331  */
3332 static const struct block_device_operations mtip_block_ops = {
3333 	.open		= mtip_block_open,
3334 	.release	= mtip_block_release,
3335 	.ioctl		= mtip_block_ioctl,
3336 #ifdef CONFIG_COMPAT
3337 	.compat_ioctl	= mtip_block_compat_ioctl,
3338 #endif
3339 	.getgeo		= mtip_block_getgeo,
3340 	.owner		= THIS_MODULE
3341 };
3342 
3343 static inline bool is_se_active(struct driver_data *dd)
3344 {
3345 	if (unlikely(test_bit(MTIP_PF_SE_ACTIVE_BIT, &dd->port->flags))) {
3346 		if (dd->port->ic_pause_timer) {
3347 			unsigned long to = dd->port->ic_pause_timer +
3348 							msecs_to_jiffies(1000);
3349 			if (time_after(jiffies, to)) {
3350 				clear_bit(MTIP_PF_SE_ACTIVE_BIT,
3351 							&dd->port->flags);
3352 				clear_bit(MTIP_DDF_SEC_LOCK_BIT, &dd->dd_flag);
3353 				dd->port->ic_pause_timer = 0;
3354 				wake_up_interruptible(&dd->port->svc_wait);
3355 				return false;
3356 			}
3357 		}
3358 		return true;
3359 	}
3360 	return false;
3361 }
3362 
3363 static inline bool is_stopped(struct driver_data *dd, struct request *rq)
3364 {
3365 	if (likely(!(dd->dd_flag & MTIP_DDF_STOP_IO)))
3366 		return false;
3367 
3368 	if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag))
3369 		return true;
3370 	if (test_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag))
3371 		return true;
3372 	if (test_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag) &&
3373 	    rq_data_dir(rq))
3374 		return true;
3375 	if (test_bit(MTIP_DDF_SEC_LOCK_BIT, &dd->dd_flag))
3376 		return true;
3377 	if (test_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag))
3378 		return true;
3379 
3380 	return false;
3381 }
3382 
3383 static bool mtip_check_unal_depth(struct blk_mq_hw_ctx *hctx,
3384 				  struct request *rq)
3385 {
3386 	struct driver_data *dd = hctx->queue->queuedata;
3387 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3388 
3389 	if (rq_data_dir(rq) == READ || !dd->unal_qdepth)
3390 		return false;
3391 
3392 	/*
3393 	 * If unaligned depth must be limited on this controller, mark it
3394 	 * as unaligned if the IO isn't on a 4k boundary (start of length).
3395 	 */
3396 	if (blk_rq_sectors(rq) <= 64) {
3397 		if ((blk_rq_pos(rq) & 7) || (blk_rq_sectors(rq) & 7))
3398 			cmd->unaligned = 1;
3399 	}
3400 
3401 	if (cmd->unaligned && atomic_dec_if_positive(&dd->port->cmd_slot_unal) >= 0)
3402 		return true;
3403 
3404 	return false;
3405 }
3406 
3407 static blk_status_t mtip_issue_reserved_cmd(struct blk_mq_hw_ctx *hctx,
3408 		struct request *rq)
3409 {
3410 	struct driver_data *dd = hctx->queue->queuedata;
3411 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3412 	struct mtip_int_cmd *icmd = cmd->icmd;
3413 	struct mtip_cmd_hdr *hdr =
3414 		dd->port->command_list + sizeof(struct mtip_cmd_hdr) * rq->tag;
3415 	struct mtip_cmd_sg *command_sg;
3416 
3417 	if (mtip_commands_active(dd->port))
3418 		return BLK_STS_DEV_RESOURCE;
3419 
3420 	hdr->ctba = cpu_to_le32(cmd->command_dma & 0xFFFFFFFF);
3421 	if (test_bit(MTIP_PF_HOST_CAP_64, &dd->port->flags))
3422 		hdr->ctbau = cpu_to_le32((cmd->command_dma >> 16) >> 16);
3423 	/* Populate the SG list */
3424 	hdr->opts = cpu_to_le32(icmd->opts | icmd->fis_len);
3425 	if (icmd->buf_len) {
3426 		command_sg = cmd->command + AHCI_CMD_TBL_HDR_SZ;
3427 
3428 		command_sg->info = cpu_to_le32((icmd->buf_len-1) & 0x3FFFFF);
3429 		command_sg->dba	= cpu_to_le32(icmd->buffer & 0xFFFFFFFF);
3430 		command_sg->dba_upper =
3431 			cpu_to_le32((icmd->buffer >> 16) >> 16);
3432 
3433 		hdr->opts |= cpu_to_le32((1 << 16));
3434 	}
3435 
3436 	/* Populate the command header */
3437 	hdr->byte_count = 0;
3438 
3439 	blk_mq_start_request(rq);
3440 	mtip_issue_non_ncq_command(dd->port, rq->tag);
3441 	return 0;
3442 }
3443 
3444 static blk_status_t mtip_queue_rq(struct blk_mq_hw_ctx *hctx,
3445 			 const struct blk_mq_queue_data *bd)
3446 {
3447 	struct driver_data *dd = hctx->queue->queuedata;
3448 	struct request *rq = bd->rq;
3449 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3450 
3451 	if (blk_rq_is_passthrough(rq))
3452 		return mtip_issue_reserved_cmd(hctx, rq);
3453 
3454 	if (unlikely(mtip_check_unal_depth(hctx, rq)))
3455 		return BLK_STS_DEV_RESOURCE;
3456 
3457 	if (is_se_active(dd) || is_stopped(dd, rq))
3458 		return BLK_STS_IOERR;
3459 
3460 	blk_mq_start_request(rq);
3461 
3462 	mtip_hw_submit_io(dd, rq, cmd, hctx);
3463 	return BLK_STS_OK;
3464 }
3465 
3466 static void mtip_free_cmd(struct blk_mq_tag_set *set, struct request *rq,
3467 			  unsigned int hctx_idx)
3468 {
3469 	struct driver_data *dd = set->driver_data;
3470 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3471 
3472 	if (!cmd->command)
3473 		return;
3474 
3475 	dma_free_coherent(&dd->pdev->dev, CMD_DMA_ALLOC_SZ, cmd->command,
3476 			  cmd->command_dma);
3477 }
3478 
3479 static int mtip_init_cmd(struct blk_mq_tag_set *set, struct request *rq,
3480 			 unsigned int hctx_idx, unsigned int numa_node)
3481 {
3482 	struct driver_data *dd = set->driver_data;
3483 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3484 
3485 	cmd->command = dma_alloc_coherent(&dd->pdev->dev, CMD_DMA_ALLOC_SZ,
3486 			&cmd->command_dma, GFP_KERNEL);
3487 	if (!cmd->command)
3488 		return -ENOMEM;
3489 
3490 	sg_init_table(cmd->sg, MTIP_MAX_SG);
3491 	return 0;
3492 }
3493 
3494 static enum blk_eh_timer_return mtip_cmd_timeout(struct request *req,
3495 								bool reserved)
3496 {
3497 	struct driver_data *dd = req->q->queuedata;
3498 
3499 	if (reserved) {
3500 		struct mtip_cmd *cmd = blk_mq_rq_to_pdu(req);
3501 
3502 		cmd->status = BLK_STS_TIMEOUT;
3503 		blk_mq_complete_request(req);
3504 		return BLK_EH_DONE;
3505 	}
3506 
3507 	if (test_bit(req->tag, dd->port->cmds_to_issue))
3508 		goto exit_handler;
3509 
3510 	if (test_and_set_bit(MTIP_PF_TO_ACTIVE_BIT, &dd->port->flags))
3511 		goto exit_handler;
3512 
3513 	wake_up_interruptible(&dd->port->svc_wait);
3514 exit_handler:
3515 	return BLK_EH_RESET_TIMER;
3516 }
3517 
3518 static const struct blk_mq_ops mtip_mq_ops = {
3519 	.queue_rq	= mtip_queue_rq,
3520 	.init_request	= mtip_init_cmd,
3521 	.exit_request	= mtip_free_cmd,
3522 	.complete	= mtip_softirq_done_fn,
3523 	.timeout        = mtip_cmd_timeout,
3524 };
3525 
3526 /*
3527  * Block layer initialization function.
3528  *
3529  * This function is called once by the PCI layer for each P320
3530  * device that is connected to the system.
3531  *
3532  * @dd Pointer to the driver data structure.
3533  *
3534  * return value
3535  *	0 on success else an error code.
3536  */
3537 static int mtip_block_initialize(struct driver_data *dd)
3538 {
3539 	int rv = 0, wait_for_rebuild = 0;
3540 	sector_t capacity;
3541 	unsigned int index = 0;
3542 
3543 	if (dd->disk)
3544 		goto skip_create_disk; /* hw init done, before rebuild */
3545 
3546 	if (mtip_hw_init(dd)) {
3547 		rv = -EINVAL;
3548 		goto protocol_init_error;
3549 	}
3550 
3551 	memset(&dd->tags, 0, sizeof(dd->tags));
3552 	dd->tags.ops = &mtip_mq_ops;
3553 	dd->tags.nr_hw_queues = 1;
3554 	dd->tags.queue_depth = MTIP_MAX_COMMAND_SLOTS;
3555 	dd->tags.reserved_tags = 1;
3556 	dd->tags.cmd_size = sizeof(struct mtip_cmd);
3557 	dd->tags.numa_node = dd->numa_node;
3558 	dd->tags.flags = BLK_MQ_F_SHOULD_MERGE;
3559 	dd->tags.driver_data = dd;
3560 	dd->tags.timeout = MTIP_NCQ_CMD_TIMEOUT_MS;
3561 
3562 	rv = blk_mq_alloc_tag_set(&dd->tags);
3563 	if (rv) {
3564 		dev_err(&dd->pdev->dev,
3565 			"Unable to allocate request queue\n");
3566 		goto block_queue_alloc_tag_error;
3567 	}
3568 
3569 	dd->disk = blk_mq_alloc_disk(&dd->tags, dd);
3570 	if (IS_ERR(dd->disk)) {
3571 		dev_err(&dd->pdev->dev,
3572 			"Unable to allocate request queue\n");
3573 		rv = -ENOMEM;
3574 		goto block_queue_alloc_init_error;
3575 	}
3576 	dd->queue		= dd->disk->queue;
3577 
3578 	rv = ida_alloc(&rssd_index_ida, GFP_KERNEL);
3579 	if (rv < 0)
3580 		goto ida_get_error;
3581 	index = rv;
3582 
3583 	rv = rssd_disk_name_format("rssd",
3584 				index,
3585 				dd->disk->disk_name,
3586 				DISK_NAME_LEN);
3587 	if (rv)
3588 		goto disk_index_error;
3589 
3590 	dd->disk->major		= dd->major;
3591 	dd->disk->first_minor	= index * MTIP_MAX_MINORS;
3592 	dd->disk->minors 	= MTIP_MAX_MINORS;
3593 	dd->disk->fops		= &mtip_block_ops;
3594 	dd->disk->private_data	= dd;
3595 	dd->index		= index;
3596 
3597 	mtip_hw_debugfs_init(dd);
3598 
3599 skip_create_disk:
3600 	/* Initialize the protocol layer. */
3601 	wait_for_rebuild = mtip_hw_get_identify(dd);
3602 	if (wait_for_rebuild < 0) {
3603 		dev_err(&dd->pdev->dev,
3604 			"Protocol layer initialization failed\n");
3605 		rv = -EINVAL;
3606 		goto init_hw_cmds_error;
3607 	}
3608 
3609 	/*
3610 	 * if rebuild pending, start the service thread, and delay the block
3611 	 * queue creation and device_add_disk()
3612 	 */
3613 	if (wait_for_rebuild == MTIP_FTL_REBUILD_MAGIC)
3614 		goto start_service_thread;
3615 
3616 	/* Set device limits. */
3617 	blk_queue_flag_set(QUEUE_FLAG_NONROT, dd->queue);
3618 	blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, dd->queue);
3619 	blk_queue_max_segments(dd->queue, MTIP_MAX_SG);
3620 	blk_queue_physical_block_size(dd->queue, 4096);
3621 	blk_queue_max_hw_sectors(dd->queue, 0xffff);
3622 	blk_queue_max_segment_size(dd->queue, 0x400000);
3623 	dma_set_max_seg_size(&dd->pdev->dev, 0x400000);
3624 	blk_queue_io_min(dd->queue, 4096);
3625 
3626 	/* Set the capacity of the device in 512 byte sectors. */
3627 	if (!(mtip_hw_get_capacity(dd, &capacity))) {
3628 		dev_warn(&dd->pdev->dev,
3629 			"Could not read drive capacity\n");
3630 		rv = -EIO;
3631 		goto read_capacity_error;
3632 	}
3633 	set_capacity(dd->disk, capacity);
3634 
3635 	/* Enable the block device and add it to /dev */
3636 	rv = device_add_disk(&dd->pdev->dev, dd->disk, mtip_disk_attr_groups);
3637 	if (rv)
3638 		goto read_capacity_error;
3639 
3640 	if (dd->mtip_svc_handler) {
3641 		set_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag);
3642 		return rv; /* service thread created for handling rebuild */
3643 	}
3644 
3645 start_service_thread:
3646 	dd->mtip_svc_handler = kthread_create_on_node(mtip_service_thread,
3647 						dd, dd->numa_node,
3648 						"mtip_svc_thd_%02d", index);
3649 
3650 	if (IS_ERR(dd->mtip_svc_handler)) {
3651 		dev_err(&dd->pdev->dev, "service thread failed to start\n");
3652 		dd->mtip_svc_handler = NULL;
3653 		rv = -EFAULT;
3654 		goto kthread_run_error;
3655 	}
3656 	wake_up_process(dd->mtip_svc_handler);
3657 	if (wait_for_rebuild == MTIP_FTL_REBUILD_MAGIC)
3658 		rv = wait_for_rebuild;
3659 
3660 	return rv;
3661 
3662 kthread_run_error:
3663 	/* Delete our gendisk. This also removes the device from /dev */
3664 	del_gendisk(dd->disk);
3665 read_capacity_error:
3666 init_hw_cmds_error:
3667 	mtip_hw_debugfs_exit(dd);
3668 disk_index_error:
3669 	ida_free(&rssd_index_ida, index);
3670 ida_get_error:
3671 	blk_cleanup_disk(dd->disk);
3672 block_queue_alloc_init_error:
3673 	blk_mq_free_tag_set(&dd->tags);
3674 block_queue_alloc_tag_error:
3675 	mtip_hw_exit(dd); /* De-initialize the protocol layer. */
3676 protocol_init_error:
3677 	return rv;
3678 }
3679 
3680 static bool mtip_no_dev_cleanup(struct request *rq, void *data, bool reserv)
3681 {
3682 	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
3683 
3684 	cmd->status = BLK_STS_IOERR;
3685 	blk_mq_complete_request(rq);
3686 	return true;
3687 }
3688 
3689 /*
3690  * Block layer deinitialization function.
3691  *
3692  * Called by the PCI layer as each P320 device is removed.
3693  *
3694  * @dd Pointer to the driver data structure.
3695  *
3696  * return value
3697  *	0
3698  */
3699 static int mtip_block_remove(struct driver_data *dd)
3700 {
3701 	mtip_hw_debugfs_exit(dd);
3702 
3703 	if (dd->mtip_svc_handler) {
3704 		set_bit(MTIP_PF_SVC_THD_STOP_BIT, &dd->port->flags);
3705 		wake_up_interruptible(&dd->port->svc_wait);
3706 		kthread_stop(dd->mtip_svc_handler);
3707 	}
3708 
3709 	if (!dd->sr) {
3710 		/*
3711 		 * Explicitly wait here for IOs to quiesce,
3712 		 * as mtip_standby_drive usually won't wait for IOs.
3713 		 */
3714 		if (!mtip_quiesce_io(dd->port, MTIP_QUIESCE_IO_TIMEOUT_MS))
3715 			mtip_standby_drive(dd);
3716 	}
3717 	else
3718 		dev_info(&dd->pdev->dev, "device %s surprise removal\n",
3719 						dd->disk->disk_name);
3720 
3721 	blk_freeze_queue_start(dd->queue);
3722 	blk_mq_quiesce_queue(dd->queue);
3723 	blk_mq_tagset_busy_iter(&dd->tags, mtip_no_dev_cleanup, dd);
3724 	blk_mq_unquiesce_queue(dd->queue);
3725 
3726 	if (dd->disk) {
3727 		if (test_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag))
3728 			del_gendisk(dd->disk);
3729 		if (dd->disk->queue) {
3730 			blk_cleanup_queue(dd->queue);
3731 			blk_mq_free_tag_set(&dd->tags);
3732 			dd->queue = NULL;
3733 		}
3734 		put_disk(dd->disk);
3735 	}
3736 	dd->disk  = NULL;
3737 
3738 	ida_free(&rssd_index_ida, dd->index);
3739 
3740 	/* De-initialize the protocol layer. */
3741 	mtip_hw_exit(dd);
3742 
3743 	return 0;
3744 }
3745 
3746 /*
3747  * Function called by the PCI layer when just before the
3748  * machine shuts down.
3749  *
3750  * If a protocol layer shutdown function is present it will be called
3751  * by this function.
3752  *
3753  * @dd Pointer to the driver data structure.
3754  *
3755  * return value
3756  *	0
3757  */
3758 static int mtip_block_shutdown(struct driver_data *dd)
3759 {
3760 	mtip_hw_shutdown(dd);
3761 
3762 	/* Delete our gendisk structure, and cleanup the blk queue. */
3763 	if (dd->disk) {
3764 		dev_info(&dd->pdev->dev,
3765 			"Shutting down %s ...\n", dd->disk->disk_name);
3766 
3767 		if (test_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag))
3768 			del_gendisk(dd->disk);
3769 		if (dd->disk->queue) {
3770 			blk_cleanup_queue(dd->queue);
3771 			blk_mq_free_tag_set(&dd->tags);
3772 		}
3773 		put_disk(dd->disk);
3774 		dd->disk  = NULL;
3775 		dd->queue = NULL;
3776 	}
3777 
3778 	ida_free(&rssd_index_ida, dd->index);
3779 	return 0;
3780 }
3781 
3782 static int mtip_block_suspend(struct driver_data *dd)
3783 {
3784 	dev_info(&dd->pdev->dev,
3785 		"Suspending %s ...\n", dd->disk->disk_name);
3786 	mtip_hw_suspend(dd);
3787 	return 0;
3788 }
3789 
3790 static int mtip_block_resume(struct driver_data *dd)
3791 {
3792 	dev_info(&dd->pdev->dev, "Resuming %s ...\n",
3793 		dd->disk->disk_name);
3794 	mtip_hw_resume(dd);
3795 	return 0;
3796 }
3797 
3798 static void drop_cpu(int cpu)
3799 {
3800 	cpu_use[cpu]--;
3801 }
3802 
3803 static int get_least_used_cpu_on_node(int node)
3804 {
3805 	int cpu, least_used_cpu, least_cnt;
3806 	const struct cpumask *node_mask;
3807 
3808 	node_mask = cpumask_of_node(node);
3809 	least_used_cpu = cpumask_first(node_mask);
3810 	least_cnt = cpu_use[least_used_cpu];
3811 	cpu = least_used_cpu;
3812 
3813 	for_each_cpu(cpu, node_mask) {
3814 		if (cpu_use[cpu] < least_cnt) {
3815 			least_used_cpu = cpu;
3816 			least_cnt = cpu_use[cpu];
3817 		}
3818 	}
3819 	cpu_use[least_used_cpu]++;
3820 	return least_used_cpu;
3821 }
3822 
3823 /* Helper for selecting a node in round robin mode */
3824 static inline int mtip_get_next_rr_node(void)
3825 {
3826 	static int next_node = NUMA_NO_NODE;
3827 
3828 	if (next_node == NUMA_NO_NODE) {
3829 		next_node = first_online_node;
3830 		return next_node;
3831 	}
3832 
3833 	next_node = next_online_node(next_node);
3834 	if (next_node == MAX_NUMNODES)
3835 		next_node = first_online_node;
3836 	return next_node;
3837 }
3838 
3839 static DEFINE_HANDLER(0);
3840 static DEFINE_HANDLER(1);
3841 static DEFINE_HANDLER(2);
3842 static DEFINE_HANDLER(3);
3843 static DEFINE_HANDLER(4);
3844 static DEFINE_HANDLER(5);
3845 static DEFINE_HANDLER(6);
3846 static DEFINE_HANDLER(7);
3847 
3848 static void mtip_disable_link_opts(struct driver_data *dd, struct pci_dev *pdev)
3849 {
3850 	unsigned short pcie_dev_ctrl;
3851 
3852 	if (pci_is_pcie(pdev)) {
3853 		pcie_capability_read_word(pdev, PCI_EXP_DEVCTL, &pcie_dev_ctrl);
3854 		if (pcie_dev_ctrl & PCI_EXP_DEVCTL_NOSNOOP_EN ||
3855 		    pcie_dev_ctrl & PCI_EXP_DEVCTL_RELAX_EN) {
3856 			dev_info(&dd->pdev->dev,
3857 				"Disabling ERO/No-Snoop on bridge device %04x:%04x\n",
3858 					pdev->vendor, pdev->device);
3859 			pcie_dev_ctrl &= ~(PCI_EXP_DEVCTL_NOSNOOP_EN |
3860 						PCI_EXP_DEVCTL_RELAX_EN);
3861 			pcie_capability_write_word(pdev, PCI_EXP_DEVCTL,
3862 				pcie_dev_ctrl);
3863 		}
3864 	}
3865 }
3866 
3867 static void mtip_fix_ero_nosnoop(struct driver_data *dd, struct pci_dev *pdev)
3868 {
3869 	/*
3870 	 * This workaround is specific to AMD/ATI chipset with a PCI upstream
3871 	 * device with device id 0x5aXX
3872 	 */
3873 	if (pdev->bus && pdev->bus->self) {
3874 		if (pdev->bus->self->vendor == PCI_VENDOR_ID_ATI &&
3875 		    ((pdev->bus->self->device & 0xff00) == 0x5a00)) {
3876 			mtip_disable_link_opts(dd, pdev->bus->self);
3877 		} else {
3878 			/* Check further up the topology */
3879 			struct pci_dev *parent_dev = pdev->bus->self;
3880 			if (parent_dev->bus &&
3881 				parent_dev->bus->parent &&
3882 				parent_dev->bus->parent->self &&
3883 				parent_dev->bus->parent->self->vendor ==
3884 					 PCI_VENDOR_ID_ATI &&
3885 				(parent_dev->bus->parent->self->device &
3886 					0xff00) == 0x5a00) {
3887 				mtip_disable_link_opts(dd,
3888 					parent_dev->bus->parent->self);
3889 			}
3890 		}
3891 	}
3892 }
3893 
3894 /*
3895  * Called for each supported PCI device detected.
3896  *
3897  * This function allocates the private data structure, enables the
3898  * PCI device and then calls the block layer initialization function.
3899  *
3900  * return value
3901  *	0 on success else an error code.
3902  */
3903 static int mtip_pci_probe(struct pci_dev *pdev,
3904 			const struct pci_device_id *ent)
3905 {
3906 	int rv = 0;
3907 	struct driver_data *dd = NULL;
3908 	char cpu_list[256];
3909 	const struct cpumask *node_mask;
3910 	int cpu, i = 0, j = 0;
3911 	int my_node = NUMA_NO_NODE;
3912 	unsigned long flags;
3913 
3914 	/* Allocate memory for this devices private data. */
3915 	my_node = pcibus_to_node(pdev->bus);
3916 	if (my_node != NUMA_NO_NODE) {
3917 		if (!node_online(my_node))
3918 			my_node = mtip_get_next_rr_node();
3919 	} else {
3920 		dev_info(&pdev->dev, "Kernel not reporting proximity, choosing a node\n");
3921 		my_node = mtip_get_next_rr_node();
3922 	}
3923 	dev_info(&pdev->dev, "NUMA node %d (closest: %d,%d, probe on %d:%d)\n",
3924 		my_node, pcibus_to_node(pdev->bus), dev_to_node(&pdev->dev),
3925 		cpu_to_node(raw_smp_processor_id()), raw_smp_processor_id());
3926 
3927 	dd = kzalloc_node(sizeof(struct driver_data), GFP_KERNEL, my_node);
3928 	if (!dd)
3929 		return -ENOMEM;
3930 
3931 	/* Attach the private data to this PCI device.  */
3932 	pci_set_drvdata(pdev, dd);
3933 
3934 	rv = pcim_enable_device(pdev);
3935 	if (rv < 0) {
3936 		dev_err(&pdev->dev, "Unable to enable device\n");
3937 		goto iomap_err;
3938 	}
3939 
3940 	/* Map BAR5 to memory. */
3941 	rv = pcim_iomap_regions(pdev, 1 << MTIP_ABAR, MTIP_DRV_NAME);
3942 	if (rv < 0) {
3943 		dev_err(&pdev->dev, "Unable to map regions\n");
3944 		goto iomap_err;
3945 	}
3946 
3947 	rv = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3948 	if (rv) {
3949 		dev_warn(&pdev->dev, "64-bit DMA enable failed\n");
3950 		goto setmask_err;
3951 	}
3952 
3953 	/* Copy the info we may need later into the private data structure. */
3954 	dd->major	= mtip_major;
3955 	dd->instance	= instance;
3956 	dd->pdev	= pdev;
3957 	dd->numa_node	= my_node;
3958 
3959 	INIT_LIST_HEAD(&dd->online_list);
3960 	INIT_LIST_HEAD(&dd->remove_list);
3961 
3962 	memset(dd->workq_name, 0, 32);
3963 	snprintf(dd->workq_name, 31, "mtipq%d", dd->instance);
3964 
3965 	dd->isr_workq = create_workqueue(dd->workq_name);
3966 	if (!dd->isr_workq) {
3967 		dev_warn(&pdev->dev, "Can't create wq %d\n", dd->instance);
3968 		rv = -ENOMEM;
3969 		goto setmask_err;
3970 	}
3971 
3972 	memset(cpu_list, 0, sizeof(cpu_list));
3973 
3974 	node_mask = cpumask_of_node(dd->numa_node);
3975 	if (!cpumask_empty(node_mask)) {
3976 		for_each_cpu(cpu, node_mask)
3977 		{
3978 			snprintf(&cpu_list[j], 256 - j, "%d ", cpu);
3979 			j = strlen(cpu_list);
3980 		}
3981 
3982 		dev_info(&pdev->dev, "Node %d on package %d has %d cpu(s): %s\n",
3983 			dd->numa_node,
3984 			topology_physical_package_id(cpumask_first(node_mask)),
3985 			nr_cpus_node(dd->numa_node),
3986 			cpu_list);
3987 	} else
3988 		dev_dbg(&pdev->dev, "mtip32xx: node_mask empty\n");
3989 
3990 	dd->isr_binding = get_least_used_cpu_on_node(dd->numa_node);
3991 	dev_info(&pdev->dev, "Initial IRQ binding node:cpu %d:%d\n",
3992 		cpu_to_node(dd->isr_binding), dd->isr_binding);
3993 
3994 	/* first worker context always runs in ISR */
3995 	dd->work[0].cpu_binding = dd->isr_binding;
3996 	dd->work[1].cpu_binding = get_least_used_cpu_on_node(dd->numa_node);
3997 	dd->work[2].cpu_binding = get_least_used_cpu_on_node(dd->numa_node);
3998 	dd->work[3].cpu_binding = dd->work[0].cpu_binding;
3999 	dd->work[4].cpu_binding = dd->work[1].cpu_binding;
4000 	dd->work[5].cpu_binding = dd->work[2].cpu_binding;
4001 	dd->work[6].cpu_binding = dd->work[2].cpu_binding;
4002 	dd->work[7].cpu_binding = dd->work[1].cpu_binding;
4003 
4004 	/* Log the bindings */
4005 	for_each_present_cpu(cpu) {
4006 		memset(cpu_list, 0, sizeof(cpu_list));
4007 		for (i = 0, j = 0; i < MTIP_MAX_SLOT_GROUPS; i++) {
4008 			if (dd->work[i].cpu_binding == cpu) {
4009 				snprintf(&cpu_list[j], 256 - j, "%d ", i);
4010 				j = strlen(cpu_list);
4011 			}
4012 		}
4013 		if (j)
4014 			dev_info(&pdev->dev, "CPU %d: WQs %s\n", cpu, cpu_list);
4015 	}
4016 
4017 	INIT_WORK(&dd->work[0].work, mtip_workq_sdbf0);
4018 	INIT_WORK(&dd->work[1].work, mtip_workq_sdbf1);
4019 	INIT_WORK(&dd->work[2].work, mtip_workq_sdbf2);
4020 	INIT_WORK(&dd->work[3].work, mtip_workq_sdbf3);
4021 	INIT_WORK(&dd->work[4].work, mtip_workq_sdbf4);
4022 	INIT_WORK(&dd->work[5].work, mtip_workq_sdbf5);
4023 	INIT_WORK(&dd->work[6].work, mtip_workq_sdbf6);
4024 	INIT_WORK(&dd->work[7].work, mtip_workq_sdbf7);
4025 
4026 	pci_set_master(pdev);
4027 	rv = pci_enable_msi(pdev);
4028 	if (rv) {
4029 		dev_warn(&pdev->dev,
4030 			"Unable to enable MSI interrupt.\n");
4031 		goto msi_initialize_err;
4032 	}
4033 
4034 	mtip_fix_ero_nosnoop(dd, pdev);
4035 
4036 	/* Initialize the block layer. */
4037 	rv = mtip_block_initialize(dd);
4038 	if (rv < 0) {
4039 		dev_err(&pdev->dev,
4040 			"Unable to initialize block layer\n");
4041 		goto block_initialize_err;
4042 	}
4043 
4044 	/*
4045 	 * Increment the instance count so that each device has a unique
4046 	 * instance number.
4047 	 */
4048 	instance++;
4049 	if (rv != MTIP_FTL_REBUILD_MAGIC)
4050 		set_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag);
4051 	else
4052 		rv = 0; /* device in rebuild state, return 0 from probe */
4053 
4054 	/* Add to online list even if in ftl rebuild */
4055 	spin_lock_irqsave(&dev_lock, flags);
4056 	list_add(&dd->online_list, &online_list);
4057 	spin_unlock_irqrestore(&dev_lock, flags);
4058 
4059 	goto done;
4060 
4061 block_initialize_err:
4062 	pci_disable_msi(pdev);
4063 
4064 msi_initialize_err:
4065 	if (dd->isr_workq) {
4066 		destroy_workqueue(dd->isr_workq);
4067 		drop_cpu(dd->work[0].cpu_binding);
4068 		drop_cpu(dd->work[1].cpu_binding);
4069 		drop_cpu(dd->work[2].cpu_binding);
4070 	}
4071 setmask_err:
4072 	pcim_iounmap_regions(pdev, 1 << MTIP_ABAR);
4073 
4074 iomap_err:
4075 	kfree(dd);
4076 	pci_set_drvdata(pdev, NULL);
4077 	return rv;
4078 done:
4079 	return rv;
4080 }
4081 
4082 /*
4083  * Called for each probed device when the device is removed or the
4084  * driver is unloaded.
4085  *
4086  * return value
4087  *	None
4088  */
4089 static void mtip_pci_remove(struct pci_dev *pdev)
4090 {
4091 	struct driver_data *dd = pci_get_drvdata(pdev);
4092 	unsigned long flags, to;
4093 
4094 	set_bit(MTIP_DDF_REMOVAL_BIT, &dd->dd_flag);
4095 
4096 	spin_lock_irqsave(&dev_lock, flags);
4097 	list_del_init(&dd->online_list);
4098 	list_add(&dd->remove_list, &removing_list);
4099 	spin_unlock_irqrestore(&dev_lock, flags);
4100 
4101 	mtip_check_surprise_removal(pdev);
4102 	synchronize_irq(dd->pdev->irq);
4103 
4104 	/* Spin until workers are done */
4105 	to = jiffies + msecs_to_jiffies(4000);
4106 	do {
4107 		msleep(20);
4108 	} while (atomic_read(&dd->irq_workers_active) != 0 &&
4109 		time_before(jiffies, to));
4110 
4111 	if (atomic_read(&dd->irq_workers_active) != 0) {
4112 		dev_warn(&dd->pdev->dev,
4113 			"Completion workers still active!\n");
4114 	}
4115 
4116 	blk_set_queue_dying(dd->queue);
4117 	set_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag);
4118 
4119 	/* Clean up the block layer. */
4120 	mtip_block_remove(dd);
4121 
4122 	if (dd->isr_workq) {
4123 		destroy_workqueue(dd->isr_workq);
4124 		drop_cpu(dd->work[0].cpu_binding);
4125 		drop_cpu(dd->work[1].cpu_binding);
4126 		drop_cpu(dd->work[2].cpu_binding);
4127 	}
4128 
4129 	pci_disable_msi(pdev);
4130 
4131 	spin_lock_irqsave(&dev_lock, flags);
4132 	list_del_init(&dd->remove_list);
4133 	spin_unlock_irqrestore(&dev_lock, flags);
4134 
4135 	kfree(dd);
4136 
4137 	pcim_iounmap_regions(pdev, 1 << MTIP_ABAR);
4138 	pci_set_drvdata(pdev, NULL);
4139 }
4140 
4141 /*
4142  * Called for each probed device when the device is suspended.
4143  *
4144  * return value
4145  *	0  Success
4146  *	<0 Error
4147  */
4148 static int mtip_pci_suspend(struct pci_dev *pdev, pm_message_t mesg)
4149 {
4150 	int rv = 0;
4151 	struct driver_data *dd = pci_get_drvdata(pdev);
4152 
4153 	if (!dd) {
4154 		dev_err(&pdev->dev,
4155 			"Driver private datastructure is NULL\n");
4156 		return -EFAULT;
4157 	}
4158 
4159 	set_bit(MTIP_DDF_RESUME_BIT, &dd->dd_flag);
4160 
4161 	/* Disable ports & interrupts then send standby immediate */
4162 	rv = mtip_block_suspend(dd);
4163 	if (rv < 0) {
4164 		dev_err(&pdev->dev,
4165 			"Failed to suspend controller\n");
4166 		return rv;
4167 	}
4168 
4169 	/*
4170 	 * Save the pci config space to pdev structure &
4171 	 * disable the device
4172 	 */
4173 	pci_save_state(pdev);
4174 	pci_disable_device(pdev);
4175 
4176 	/* Move to Low power state*/
4177 	pci_set_power_state(pdev, PCI_D3hot);
4178 
4179 	return rv;
4180 }
4181 
4182 /*
4183  * Called for each probed device when the device is resumed.
4184  *
4185  * return value
4186  *      0  Success
4187  *      <0 Error
4188  */
4189 static int mtip_pci_resume(struct pci_dev *pdev)
4190 {
4191 	int rv = 0;
4192 	struct driver_data *dd;
4193 
4194 	dd = pci_get_drvdata(pdev);
4195 	if (!dd) {
4196 		dev_err(&pdev->dev,
4197 			"Driver private datastructure is NULL\n");
4198 		return -EFAULT;
4199 	}
4200 
4201 	/* Move the device to active State */
4202 	pci_set_power_state(pdev, PCI_D0);
4203 
4204 	/* Restore PCI configuration space */
4205 	pci_restore_state(pdev);
4206 
4207 	/* Enable the PCI device*/
4208 	rv = pcim_enable_device(pdev);
4209 	if (rv < 0) {
4210 		dev_err(&pdev->dev,
4211 			"Failed to enable card during resume\n");
4212 		goto err;
4213 	}
4214 	pci_set_master(pdev);
4215 
4216 	/*
4217 	 * Calls hbaReset, initPort, & startPort function
4218 	 * then enables interrupts
4219 	 */
4220 	rv = mtip_block_resume(dd);
4221 	if (rv < 0)
4222 		dev_err(&pdev->dev, "Unable to resume\n");
4223 
4224 err:
4225 	clear_bit(MTIP_DDF_RESUME_BIT, &dd->dd_flag);
4226 
4227 	return rv;
4228 }
4229 
4230 /*
4231  * Shutdown routine
4232  *
4233  * return value
4234  *      None
4235  */
4236 static void mtip_pci_shutdown(struct pci_dev *pdev)
4237 {
4238 	struct driver_data *dd = pci_get_drvdata(pdev);
4239 	if (dd)
4240 		mtip_block_shutdown(dd);
4241 }
4242 
4243 /* Table of device ids supported by this driver. */
4244 static const struct pci_device_id mtip_pci_tbl[] = {
4245 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P320H_DEVICE_ID) },
4246 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P320M_DEVICE_ID) },
4247 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P320S_DEVICE_ID) },
4248 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P325M_DEVICE_ID) },
4249 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P420H_DEVICE_ID) },
4250 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P420M_DEVICE_ID) },
4251 	{ PCI_DEVICE(PCI_VENDOR_ID_MICRON, P425M_DEVICE_ID) },
4252 	{ 0 }
4253 };
4254 
4255 /* Structure that describes the PCI driver functions. */
4256 static struct pci_driver mtip_pci_driver = {
4257 	.name			= MTIP_DRV_NAME,
4258 	.id_table		= mtip_pci_tbl,
4259 	.probe			= mtip_pci_probe,
4260 	.remove			= mtip_pci_remove,
4261 	.suspend		= mtip_pci_suspend,
4262 	.resume			= mtip_pci_resume,
4263 	.shutdown		= mtip_pci_shutdown,
4264 };
4265 
4266 MODULE_DEVICE_TABLE(pci, mtip_pci_tbl);
4267 
4268 /*
4269  * Module initialization function.
4270  *
4271  * Called once when the module is loaded. This function allocates a major
4272  * block device number to the Cyclone devices and registers the PCI layer
4273  * of the driver.
4274  *
4275  * Return value
4276  *      0 on success else error code.
4277  */
4278 static int __init mtip_init(void)
4279 {
4280 	int error;
4281 
4282 	pr_info(MTIP_DRV_NAME " Version " MTIP_DRV_VERSION "\n");
4283 
4284 	/* Allocate a major block device number to use with this driver. */
4285 	error = register_blkdev(0, MTIP_DRV_NAME);
4286 	if (error <= 0) {
4287 		pr_err("Unable to register block device (%d)\n",
4288 		error);
4289 		return -EBUSY;
4290 	}
4291 	mtip_major = error;
4292 
4293 	dfs_parent = debugfs_create_dir("rssd", NULL);
4294 	if (IS_ERR_OR_NULL(dfs_parent)) {
4295 		pr_warn("Error creating debugfs parent\n");
4296 		dfs_parent = NULL;
4297 	}
4298 	if (dfs_parent) {
4299 		dfs_device_status = debugfs_create_file("device_status",
4300 					0444, dfs_parent, NULL,
4301 					&mtip_device_status_fops);
4302 		if (IS_ERR_OR_NULL(dfs_device_status)) {
4303 			pr_err("Error creating device_status node\n");
4304 			dfs_device_status = NULL;
4305 		}
4306 	}
4307 
4308 	/* Register our PCI operations. */
4309 	error = pci_register_driver(&mtip_pci_driver);
4310 	if (error) {
4311 		debugfs_remove(dfs_parent);
4312 		unregister_blkdev(mtip_major, MTIP_DRV_NAME);
4313 	}
4314 
4315 	return error;
4316 }
4317 
4318 /*
4319  * Module de-initialization function.
4320  *
4321  * Called once when the module is unloaded. This function deallocates
4322  * the major block device number allocated by mtip_init() and
4323  * unregisters the PCI layer of the driver.
4324  *
4325  * Return value
4326  *      none
4327  */
4328 static void __exit mtip_exit(void)
4329 {
4330 	/* Release the allocated major block device number. */
4331 	unregister_blkdev(mtip_major, MTIP_DRV_NAME);
4332 
4333 	/* Unregister the PCI driver. */
4334 	pci_unregister_driver(&mtip_pci_driver);
4335 
4336 	debugfs_remove_recursive(dfs_parent);
4337 }
4338 
4339 MODULE_AUTHOR("Micron Technology, Inc");
4340 MODULE_DESCRIPTION("Micron RealSSD PCIe Block Driver");
4341 MODULE_LICENSE("GPL");
4342 MODULE_VERSION(MTIP_DRV_VERSION);
4343 
4344 module_init(mtip_init);
4345 module_exit(mtip_exit);
4346