aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2012-01-15 15:48:41 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2012-01-15 15:48:41 -0500
commit16008d641670571ff4cd750b416c7caf2d89f467 (patch)
tree6dc32821bd4576106cee50d52a90a8106a9d481f
parentb3c9dd182ed3bdcdaf0e42625a35924b0497afdc (diff)
parent85a0f7b220777cd3c232fd2f079db5c4a874c4ee (diff)
Merge branch 'for-3.3/drivers' of git://git.kernel.dk/linux-block
* 'for-3.3/drivers' of git://git.kernel.dk/linux-block: mtip32xx: do rebuild monitoring asynchronously xen-blkfront: Use kcalloc instead of kzalloc to allocate array mtip32xx: uninitialized variable in mtip_quiesce_io() mtip32xx: updates based on feedback xen-blkback: convert hole punching to discard request on loop devices xen/blkback: Move processing of BLKIF_OP_DISCARD from dispatch_rw_block_io xen/blk[front|back]: Enhance discard support with secure erasing support. xen/blk[front|back]: Squash blkif_request_rw and blkif_request_discard together mtip32xx: update to new ->make_request() API mtip32xx: add module.h include to avoid conflict with moduleh tree mtip32xx: mark a few more items static mtip32xx: ensure that all local functions are static mtip32xx: cleanup compat ioctl handling mtip32xx: fix warnings/errors on 32-bit compiles block: Add driver for Micron RealSSD pcie flash cards
Diffstat
-rw-r--r--drivers/block/Kconfig2
-rw-r--r--drivers/block/Makefile1
-rw-r--r--drivers/block/mtip32xx/Kconfig9
-rw-r--r--drivers/block/mtip32xx/Makefile5
-rw-r--r--drivers/block/mtip32xx/mtip32xx.c3651
-rw-r--r--drivers/block/mtip32xx/mtip32xx.h423
-rw-r--r--drivers/block/xen-blkback/blkback.c84
-rw-r--r--drivers/block/xen-blkback/common.h67
-rw-r--r--drivers/block/xen-blkback/xenbus.c12
-rw-r--r--drivers/block/xen-blkfront.c79
-rw-r--r--include/xen/interface/io/blkif.h40
11 files changed, 4261 insertions, 112 deletions
diff --git a/drivers/block/Kconfig b/drivers/block/Kconfig
index 6f07ec1c2f58..a30aa103f95b 100644
--- a/drivers/block/Kconfig
+++ b/drivers/block/Kconfig
@@ -116,6 +116,8 @@ config PARIDE
116 116
117source "drivers/block/paride/Kconfig" 117source "drivers/block/paride/Kconfig"
118 118
119source "drivers/block/mtip32xx/Kconfig"
120
119config BLK_CPQ_DA 121config BLK_CPQ_DA
120 tristate "Compaq SMART2 support" 122 tristate "Compaq SMART2 support"
121 depends on PCI && VIRT_TO_BUS 123 depends on PCI && VIRT_TO_BUS
diff --git a/drivers/block/Makefile b/drivers/block/Makefile
index 76646e9a1c91..ad7b74a44ef3 100644
--- a/drivers/block/Makefile
+++ b/drivers/block/Makefile
@@ -39,5 +39,6 @@ obj-$(CONFIG_XEN_BLKDEV_FRONTEND) += xen-blkfront.o
39obj-$(CONFIG_XEN_BLKDEV_BACKEND) += xen-blkback/ 39obj-$(CONFIG_XEN_BLKDEV_BACKEND) += xen-blkback/
40obj-$(CONFIG_BLK_DEV_DRBD) += drbd/ 40obj-$(CONFIG_BLK_DEV_DRBD) += drbd/
41obj-$(CONFIG_BLK_DEV_RBD) += rbd.o 41obj-$(CONFIG_BLK_DEV_RBD) += rbd.o
42obj-$(CONFIG_BLK_DEV_PCIESSD_MTIP32XX) += mtip32xx/
42 43
43swim_mod-y := swim.o swim_asm.o 44swim_mod-y := swim.o swim_asm.o
diff --git a/drivers/block/mtip32xx/Kconfig b/drivers/block/mtip32xx/Kconfig
new file mode 100644
index 000000000000..b5dd14e072f2
--- /dev/null
+++ b/drivers/block/mtip32xx/Kconfig
@@ -0,0 +1,9 @@
1#
2# mtip32xx device driver configuration
3#
4
5config BLK_DEV_PCIESSD_MTIP32XX
6 tristate "Block Device Driver for Micron PCIe SSDs"
7 depends on HOTPLUG_PCI_PCIE
8 help
9 This enables the block driver for Micron PCIe SSDs.
diff --git a/drivers/block/mtip32xx/Makefile b/drivers/block/mtip32xx/Makefile
new file mode 100644
index 000000000000..4fbef8c8329b
--- /dev/null
+++ b/drivers/block/mtip32xx/Makefile
@@ -0,0 +1,5 @@
1#
2# Makefile for Block device driver for Micron PCIe SSD
3#
4
5obj-$(CONFIG_BLK_DEV_PCIESSD_MTIP32XX) += mtip32xx.o
diff --git a/drivers/block/mtip32xx/mtip32xx.c b/drivers/block/mtip32xx/mtip32xx.c
new file mode 100644
index 000000000000..b74eab70c3d0
--- /dev/null
+++ b/drivers/block/mtip32xx/mtip32xx.c
@@ -0,0 +1,3651 @@
1/*
2 * Driver for the Micron P320 SSD
3 * Copyright (C) 2011 Micron Technology, Inc.
4 *
5 * Portions of this code were derived from works subjected to the
6 * following copyright:
7 * Copyright (C) 2009 Integrated Device Technology, Inc.
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
13 *
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
18 *
19 */
20
21#include <linux/pci.h>
22#include <linux/interrupt.h>
23#include <linux/ata.h>
24#include <linux/delay.h>
25#include <linux/hdreg.h>
26#include <linux/uaccess.h>
27#include <linux/random.h>
28#include <linux/smp.h>
29#include <linux/compat.h>
30#include <linux/fs.h>
31#include <linux/module.h>
32#include <linux/genhd.h>
33#include <linux/blkdev.h>
34#include <linux/bio.h>
35#include <linux/dma-mapping.h>
36#include <linux/idr.h>
37#include <linux/kthread.h>
38#include <../drivers/ata/ahci.h>
39#include "mtip32xx.h"
40
41#define HW_CMD_SLOT_SZ (MTIP_MAX_COMMAND_SLOTS * 32)
42#define HW_CMD_TBL_SZ (AHCI_CMD_TBL_HDR_SZ + (MTIP_MAX_SG * 16))
43#define HW_CMD_TBL_AR_SZ (HW_CMD_TBL_SZ * MTIP_MAX_COMMAND_SLOTS)
44#define HW_PORT_PRIV_DMA_SZ \
45 (HW_CMD_SLOT_SZ + HW_CMD_TBL_AR_SZ + AHCI_RX_FIS_SZ)
46
47#define HOST_HSORG 0xFC
48#define HSORG_DISABLE_SLOTGRP_INTR (1<<24)
49#define HSORG_DISABLE_SLOTGRP_PXIS (1<<16)
50#define HSORG_HWREV 0xFF00
51#define HSORG_STYLE 0x8
52#define HSORG_SLOTGROUPS 0x7
53
54#define PORT_COMMAND_ISSUE 0x38
55#define PORT_SDBV 0x7C
56
57#define PORT_OFFSET 0x100
58#define PORT_MEM_SIZE 0x80
59
60#define PORT_IRQ_ERR \
61 (PORT_IRQ_HBUS_ERR | PORT_IRQ_IF_ERR | PORT_IRQ_CONNECT | \
62 PORT_IRQ_PHYRDY | PORT_IRQ_UNK_FIS | PORT_IRQ_BAD_PMP | \
63 PORT_IRQ_TF_ERR | PORT_IRQ_HBUS_DATA_ERR | PORT_IRQ_IF_NONFATAL | \
64 PORT_IRQ_OVERFLOW)
65#define PORT_IRQ_LEGACY \
66 (PORT_IRQ_PIOS_FIS | PORT_IRQ_D2H_REG_FIS)
67#define PORT_IRQ_HANDLED \
68 (PORT_IRQ_SDB_FIS | PORT_IRQ_LEGACY | \
69 PORT_IRQ_TF_ERR | PORT_IRQ_IF_ERR | \
70 PORT_IRQ_CONNECT | PORT_IRQ_PHYRDY)
71#define DEF_PORT_IRQ \
72 (PORT_IRQ_ERR | PORT_IRQ_LEGACY | PORT_IRQ_SDB_FIS)
73
74/* product numbers */
75#define MTIP_PRODUCT_UNKNOWN 0x00
76#define MTIP_PRODUCT_ASICFPGA 0x11
77
78/* Device instance number, incremented each time a device is probed. */
79static int instance;
80
81/*
82 * Global variable used to hold the major block device number
83 * allocated in mtip_init().
84 */
85static int mtip_major;
86
87static DEFINE_SPINLOCK(rssd_index_lock);
88static DEFINE_IDA(rssd_index_ida);
89
90static int mtip_block_initialize(struct driver_data *dd);
91
92#ifdef CONFIG_COMPAT
93struct mtip_compat_ide_task_request_s {
94 __u8 io_ports[8];
95 __u8 hob_ports[8];
96 ide_reg_valid_t out_flags;
97 ide_reg_valid_t in_flags;
98 int data_phase;
99 int req_cmd;
100 compat_ulong_t out_size;
101 compat_ulong_t in_size;
102};
103#endif
104
105/*
106 * This function check_for_surprise_removal is called
107 * while card is removed from the system and it will
108 * read the vendor id from the configration space
109 *
110 * @pdev Pointer to the pci_dev structure.
111 *
112 * return value
113 * true if device removed, else false
114 */
115static bool mtip_check_surprise_removal(struct pci_dev *pdev)
116{
117 u16 vendor_id = 0;
118
119 /* Read the vendorID from the configuration space */
120 pci_read_config_word(pdev, 0x00, &vendor_id);
121 if (vendor_id == 0xFFFF)
122 return true; /* device removed */
123
124 return false; /* device present */
125}
126
127/*
128 * This function is called for clean the pending command in the
129 * command slot during the surprise removal of device and return
130 * error to the upper layer.
131 *
132 * @dd Pointer to the DRIVER_DATA structure.
133 *
134 * return value
135 * None
136 */
137static void mtip_command_cleanup(struct driver_data *dd)
138{
139 int group = 0, commandslot = 0, commandindex = 0;
140 struct mtip_cmd *command;
141 struct mtip_port *port = dd->port;
142
143 for (group = 0; group < 4; group++) {
144 for (commandslot = 0; commandslot < 32; commandslot++) {
145 if (!(port->allocated[group] & (1 << commandslot)))
146 continue;
147
148 commandindex = group << 5 | commandslot;
149 command = &port->commands[commandindex];
150
151 if (atomic_read(&command->active)
152 && (command->async_callback)) {
153 command->async_callback(command->async_data,
154 -ENODEV);
155 command->async_callback = NULL;
156 command->async_data = NULL;
157 }
158
159 dma_unmap_sg(&port->dd->pdev->dev,
160 command->sg,
161 command->scatter_ents,
162 command->direction);
163 }
164 }
165
166 up(&port->cmd_slot);
167
168 atomic_set(&dd->drv_cleanup_done, true);
169}
170
171/*
172 * Obtain an empty command slot.
173 *
174 * This function needs to be reentrant since it could be called
175 * at the same time on multiple CPUs. The allocation of the
176 * command slot must be atomic.
177 *
178 * @port Pointer to the port data structure.
179 *
180 * return value
181 * >= 0 Index of command slot obtained.
182 * -1 No command slots available.
183 */
184static int get_slot(struct mtip_port *port)
185{
186 int slot, i;
187 unsigned int num_command_slots = port->dd->slot_groups * 32;
188
189 /*
190 * Try 10 times, because there is a small race here.
191 * that's ok, because it's still cheaper than a lock.
192 *
193 * Race: Since this section is not protected by lock, same bit
194 * could be chosen by different process contexts running in
195 * different processor. So instead of costly lock, we are going
196 * with loop.
197 */
198 for (i = 0; i < 10; i++) {
199 slot = find_next_zero_bit(port->allocated,
200 num_command_slots, 1);
201 if ((slot < num_command_slots) &&
202 (!test_and_set_bit(slot, port->allocated)))
203 return slot;
204 }
205 dev_warn(&port->dd->pdev->dev, "Failed to get a tag.\n");
206
207 if (mtip_check_surprise_removal(port->dd->pdev)) {
208 /* Device not present, clean outstanding commands */
209 mtip_command_cleanup(port->dd);
210 }
211 return -1;
212}
213
214/*
215 * Release a command slot.
216 *
217 * @port Pointer to the port data structure.
218 * @tag Tag of command to release
219 *
220 * return value
221 * None
222 */
223static inline void release_slot(struct mtip_port *port, int tag)
224{
225 smp_mb__before_clear_bit();
226 clear_bit(tag, port->allocated);
227 smp_mb__after_clear_bit();
228}
229
230/*
231 * Reset the HBA (without sleeping)
232 *
233 * Just like hba_reset, except does not call sleep, so can be
234 * run from interrupt/tasklet context.
235 *
236 * @dd Pointer to the driver data structure.
237 *
238 * return value
239 * 0 The reset was successful.
240 * -1 The HBA Reset bit did not clear.
241 */
242static int hba_reset_nosleep(struct driver_data *dd)
243{
244 unsigned long timeout;
245
246 /* Chip quirk: quiesce any chip function */
247 mdelay(10);
248
249 /* Set the reset bit */
250 writel(HOST_RESET, dd->mmio + HOST_CTL);
251
252 /* Flush */
253 readl(dd->mmio + HOST_CTL);
254
255 /*
256 * Wait 10ms then spin for up to 1 second
257 * waiting for reset acknowledgement
258 */
259 timeout = jiffies + msecs_to_jiffies(1000);
260 mdelay(10);
261 while ((readl(dd->mmio + HOST_CTL) & HOST_RESET)
262 && time_before(jiffies, timeout))
263 mdelay(1);
264
265 if (readl(dd->mmio + HOST_CTL) & HOST_RESET)
266 return -1;
267
268 return 0;
269}
270
271/*
272 * Issue a command to the hardware.
273 *
274 * Set the appropriate bit in the s_active and Command Issue hardware
275 * registers, causing hardware command processing to begin.
276 *
277 * @port Pointer to the port structure.
278 * @tag The tag of the command to be issued.
279 *
280 * return value
281 * None
282 */
283static inline void mtip_issue_ncq_command(struct mtip_port *port, int tag)
284{
285 unsigned long flags = 0;
286
287 atomic_set(&port->commands[tag].active, 1);
288
289 spin_lock_irqsave(&port->cmd_issue_lock, flags);
290
291 writel((1 << MTIP_TAG_BIT(tag)),
292 port->s_active[MTIP_TAG_INDEX(tag)]);
293 writel((1 << MTIP_TAG_BIT(tag)),
294 port->cmd_issue[MTIP_TAG_INDEX(tag)]);
295
296 spin_unlock_irqrestore(&port->cmd_issue_lock, flags);
297}
298
299/*
300 * Enable/disable the reception of FIS
301 *
302 * @port Pointer to the port data structure
303 * @enable 1 to enable, 0 to disable
304 *
305 * return value
306 * Previous state: 1 enabled, 0 disabled
307 */
308static int mtip_enable_fis(struct mtip_port *port, int enable)
309{
310 u32 tmp;
311
312 /* enable FIS reception */
313 tmp = readl(port->mmio + PORT_CMD);
314 if (enable)
315 writel(tmp | PORT_CMD_FIS_RX, port->mmio + PORT_CMD);
316 else
317 writel(tmp & ~PORT_CMD_FIS_RX, port->mmio + PORT_CMD);
318
319 /* Flush */
320 readl(port->mmio + PORT_CMD);
321
322 return (((tmp & PORT_CMD_FIS_RX) == PORT_CMD_FIS_RX));
323}
324
325/*
326 * Enable/disable the DMA engine
327 *
328 * @port Pointer to the port data structure
329 * @enable 1 to enable, 0 to disable
330 *
331 * return value
332 * Previous state: 1 enabled, 0 disabled.
333 */
334static int mtip_enable_engine(struct mtip_port *port, int enable)
335{
336 u32 tmp;
337
338 /* enable FIS reception */
339 tmp = readl(port->mmio + PORT_CMD);
340 if (enable)
341 writel(tmp | PORT_CMD_START, port->mmio + PORT_CMD);
342 else
343 writel(tmp & ~PORT_CMD_START, port->mmio + PORT_CMD);
344
345 readl(port->mmio + PORT_CMD);
346 return (((tmp & PORT_CMD_START) == PORT_CMD_START));
347}
348
349/*
350 * Enables the port DMA engine and FIS reception.
351 *
352 * return value
353 * None
354 */
355static inline void mtip_start_port(struct mtip_port *port)
356{
357 /* Enable FIS reception */
358 mtip_enable_fis(port, 1);
359
360 /* Enable the DMA engine */
361 mtip_enable_engine(port, 1);
362}
363
364/*
365 * Deinitialize a port by disabling port interrupts, the DMA engine,
366 * and FIS reception.
367 *
368 * @port Pointer to the port structure
369 *
370 * return value
371 * None
372 */
373static inline void mtip_deinit_port(struct mtip_port *port)
374{
375 /* Disable interrupts on this port */
376 writel(0, port->mmio + PORT_IRQ_MASK);
377
378 /* Disable the DMA engine */
379 mtip_enable_engine(port, 0);
380
381 /* Disable FIS reception */
382 mtip_enable_fis(port, 0);
383}
384
385/*
386 * Initialize a port.
387 *
388 * This function deinitializes the port by calling mtip_deinit_port() and
389 * then initializes it by setting the command header and RX FIS addresses,
390 * clearing the SError register and any pending port interrupts before
391 * re-enabling the default set of port interrupts.
392 *
393 * @port Pointer to the port structure.
394 *
395 * return value
396 * None
397 */
398static void mtip_init_port(struct mtip_port *port)
399{
400 int i;
401 mtip_deinit_port(port);
402
403 /* Program the command list base and FIS base addresses */
404 if (readl(port->dd->mmio + HOST_CAP) & HOST_CAP_64) {
405 writel((port->command_list_dma >> 16) >> 16,
406 port->mmio + PORT_LST_ADDR_HI);
407 writel((port->rxfis_dma >> 16) >> 16,
408 port->mmio + PORT_FIS_ADDR_HI);
409 }
410
411 writel(port->command_list_dma & 0xFFFFFFFF,
412 port->mmio + PORT_LST_ADDR);
413 writel(port->rxfis_dma & 0xFFFFFFFF, port->mmio + PORT_FIS_ADDR);
414
415 /* Clear SError */
416 writel(readl(port->mmio + PORT_SCR_ERR), port->mmio + PORT_SCR_ERR);
417
418 /* reset the completed registers.*/
419 for (i = 0; i < port->dd->slot_groups; i++)
420 writel(0xFFFFFFFF, port->completed[i]);
421
422 /* Clear any pending interrupts for this port */
423 writel(readl(port->mmio + PORT_IRQ_STAT), port->mmio + PORT_IRQ_STAT);
424
425 /* Enable port interrupts */
426 writel(DEF_PORT_IRQ, port->mmio + PORT_IRQ_MASK);
427}
428
429/*
430 * Restart a port
431 *
432 * @port Pointer to the port data structure.
433 *
434 * return value
435 * None
436 */
437static void mtip_restart_port(struct mtip_port *port)
438{
439 unsigned long timeout;
440
441 /* Disable the DMA engine */
442 mtip_enable_engine(port, 0);
443
444 /* Chip quirk: wait up to 500ms for PxCMD.CR == 0 */
445 timeout = jiffies + msecs_to_jiffies(500);
446 while ((readl(port->mmio + PORT_CMD) & PORT_CMD_LIST_ON)
447 && time_before(jiffies, timeout))
448 ;
449
450 /*
451 * Chip quirk: escalate to hba reset if
452 * PxCMD.CR not clear after 500 ms
453 */
454 if (readl(port->mmio + PORT_CMD) & PORT_CMD_LIST_ON) {
455 dev_warn(&port->dd->pdev->dev,
456 "PxCMD.CR not clear, escalating reset\n");
457
458 if (hba_reset_nosleep(port->dd))
459 dev_err(&port->dd->pdev->dev,
460 "HBA reset escalation failed.\n");
461
462 /* 30 ms delay before com reset to quiesce chip */
463 mdelay(30);
464 }
465
466 dev_warn(&port->dd->pdev->dev, "Issuing COM reset\n");
467
468 /* Set PxSCTL.DET */
469 writel(readl(port->mmio + PORT_SCR_CTL) |
470 1, port->mmio + PORT_SCR_CTL);
471 readl(port->mmio + PORT_SCR_CTL);
472
473 /* Wait 1 ms to quiesce chip function */
474 timeout = jiffies + msecs_to_jiffies(1);
475 while (time_before(jiffies, timeout))
476 ;
477
478 /* Clear PxSCTL.DET */
479 writel(readl(port->mmio + PORT_SCR_CTL) & ~1,
480 port->mmio + PORT_SCR_CTL);
481 readl(port->mmio + PORT_SCR_CTL);
482
483 /* Wait 500 ms for bit 0 of PORT_SCR_STS to be set */
484 timeout = jiffies + msecs_to_jiffies(500);
485 while (((readl(port->mmio + PORT_SCR_STAT) & 0x01) == 0)
486 && time_before(jiffies, timeout))
487 ;
488
489 if ((readl(port->mmio + PORT_SCR_STAT) & 0x01) == 0)
490 dev_warn(&port->dd->pdev->dev,
491 "COM reset failed\n");
492
493 /* Clear SError, the PxSERR.DIAG.x should be set so clear it */
494 writel(readl(port->mmio + PORT_SCR_ERR), port->mmio + PORT_SCR_ERR);
495
496 /* Enable the DMA engine */
497 mtip_enable_engine(port, 1);
498}
499
500/*
501 * Called periodically to see if any read/write commands are
502 * taking too long to complete.
503 *
504 * @data Pointer to the PORT data structure.
505 *
506 * return value
507 * None
508 */
509static void mtip_timeout_function(unsigned long int data)
510{
511 struct mtip_port *port = (struct mtip_port *) data;
512 struct host_to_dev_fis *fis;
513 struct mtip_cmd *command;
514 int tag, cmdto_cnt = 0;
515 unsigned int bit, group;
516 unsigned int num_command_slots = port->dd->slot_groups * 32;
517
518 if (unlikely(!port))
519 return;
520
521 if (atomic_read(&port->dd->resumeflag) == true) {
522 mod_timer(&port->cmd_timer,
523 jiffies + msecs_to_jiffies(30000));
524 return;
525 }
526
527 for (tag = 0; tag < num_command_slots; tag++) {
528 /*
529 * Skip internal command slot as it has
530 * its own timeout mechanism
531 */
532 if (tag == MTIP_TAG_INTERNAL)
533 continue;
534
535 if (atomic_read(&port->commands[tag].active) &&
536 (time_after(jiffies, port->commands[tag].comp_time))) {
537 group = tag >> 5;
538 bit = tag & 0x1F;
539
540 command = &port->commands[tag];
541 fis = (struct host_to_dev_fis *) command->command;
542
543 dev_warn(&port->dd->pdev->dev,
544 "Timeout for command tag %d\n", tag);
545
546 cmdto_cnt++;
547 if (cmdto_cnt == 1)
548 set_bit(MTIP_FLAG_EH_ACTIVE_BIT, &port->flags);
549
550 /*
551 * Clear the completed bit. This should prevent
552 * any interrupt handlers from trying to retire
553 * the command.
554 */
555 writel(1 << bit, port->completed[group]);
556
557 /* Call the async completion callback. */
558 if (likely(command->async_callback))
559 command->async_callback(command->async_data,
560 -EIO);
561 command->async_callback = NULL;
562 command->comp_func = NULL;
563
564 /* Unmap the DMA scatter list entries */
565 dma_unmap_sg(&port->dd->pdev->dev,
566 command->sg,
567 command->scatter_ents,
568 command->direction);
569
570 /*
571 * Clear the allocated bit and active tag for the
572 * command.
573 */
574 atomic_set(&port->commands[tag].active, 0);
575 release_slot(port, tag);
576
577 up(&port->cmd_slot);
578 }
579 }
580
581 if (cmdto_cnt) {
582 dev_warn(&port->dd->pdev->dev,
583 "%d commands timed out: restarting port",
584 cmdto_cnt);
585 mtip_restart_port(port);
586 clear_bit(MTIP_FLAG_EH_ACTIVE_BIT, &port->flags);
587 wake_up_interruptible(&port->svc_wait);
588 }
589
590 /* Restart the timer */
591 mod_timer(&port->cmd_timer,
592 jiffies + msecs_to_jiffies(MTIP_TIMEOUT_CHECK_PERIOD));
593}
594
595/*
596 * IO completion function.
597 *
598 * This completion function is called by the driver ISR when a
599 * command that was issued by the kernel completes. It first calls the
600 * asynchronous completion function which normally calls back into the block
601 * layer passing the asynchronous callback data, then unmaps the
602 * scatter list associated with the completed command, and finally
603 * clears the allocated bit associated with the completed command.
604 *
605 * @port Pointer to the port data structure.
606 * @tag Tag of the command.
607 * @data Pointer to driver_data.
608 * @status Completion status.
609 *
610 * return value
611 * None
612 */
613static void mtip_async_complete(struct mtip_port *port,
614 int tag,
615 void *data,
616 int status)
617{
618 struct mtip_cmd *command;
619 struct driver_data *dd = data;
620 int cb_status = status ? -EIO : 0;
621
622 if (unlikely(!dd) || unlikely(!port))
623 return;
624
625 command = &port->commands[tag];
626
627 if (unlikely(status == PORT_IRQ_TF_ERR)) {
628 dev_warn(&port->dd->pdev->dev,
629 "Command tag %d failed due to TFE\n", tag);
630 }
631
632 /* Upper layer callback */
633 if (likely(command->async_callback))
634 command->async_callback(command->async_data, cb_status);
635
636 command->async_callback = NULL;
637 command->comp_func = NULL;
638
639 /* Unmap the DMA scatter list entries */
640 dma_unmap_sg(&dd->pdev->dev,
641 command->sg,
642 command->scatter_ents,
643 command->direction);
644
645 /* Clear the allocated and active bits for the command */
646 atomic_set(&port->commands[tag].active, 0);
647 release_slot(port, tag);
648
649 up(&port->cmd_slot);
650}
651
652/*
653 * Internal command completion callback function.
654 *
655 * This function is normally called by the driver ISR when an internal
656 * command completed. This function signals the command completion by
657 * calling complete().
658 *
659 * @port Pointer to the port data structure.
660 * @tag Tag of the command that has completed.
661 * @data Pointer to a completion structure.
662 * @status Completion status.
663 *
664 * return value
665 * None
666 */
667static void mtip_completion(struct mtip_port *port,
668 int tag,
669 void *data,
670 int status)
671{
672 struct mtip_cmd *command = &port->commands[tag];
673 struct completion *waiting = data;
674 if (unlikely(status == PORT_IRQ_TF_ERR))
675 dev_warn(&port->dd->pdev->dev,
676 "Internal command %d completed with TFE\n", tag);
677
678 command->async_callback = NULL;
679 command->comp_func = NULL;
680
681 complete(waiting);
682}
683
684/*
685 * Helper function for tag logging
686 */
687static void print_tags(struct driver_data *dd,
688 char *msg,
689 unsigned long *tagbits)
690{
691 unsigned int tag, count = 0;
692
693 for (tag = 0; tag < (dd->slot_groups) * 32; tag++) {
694 if (test_bit(tag, tagbits))
695 count++;
696 }
697 if (count)
698 dev_info(&dd->pdev->dev, "%s [%i tags]\n", msg, count);
699}
700
701/*
702 * Handle an error.
703 *
704 * @dd Pointer to the DRIVER_DATA structure.
705 *
706 * return value
707 * None
708 */
709static void mtip_handle_tfe(struct driver_data *dd)
710{
711 int group, tag, bit, reissue;
712 struct mtip_port *port;
713 struct mtip_cmd *command;
714 u32 completed;
715 struct host_to_dev_fis *fis;
716 unsigned long tagaccum[SLOTBITS_IN_LONGS];
717
718 dev_warn(&dd->pdev->dev, "Taskfile error\n");
719
720 port = dd->port;
721
722 /* Stop the timer to prevent command timeouts. */
723 del_timer(&port->cmd_timer);
724
725 /* Set eh_active */
726 set_bit(MTIP_FLAG_EH_ACTIVE_BIT, &port->flags);
727
728 /* Loop through all the groups */
729 for (group = 0; group < dd->slot_groups; group++) {
730 completed = readl(port->completed[group]);
731
732 /* clear completed status register in the hardware.*/
733 writel(completed, port->completed[group]);
734
735 /* clear the tag accumulator */
736 memset(tagaccum, 0, SLOTBITS_IN_LONGS * sizeof(long));
737
738 /* Process successfully completed commands */
739 for (bit = 0; bit < 32 && completed; bit++) {
740 if (!(completed & (1<<bit)))
741 continue;
742 tag = (group << 5) + bit;
743
744 /* Skip the internal command slot */
745 if (tag == MTIP_TAG_INTERNAL)
746 continue;
747
748 command = &port->commands[tag];
749 if (likely(command->comp_func)) {
750 set_bit(tag, tagaccum);
751 atomic_set(&port->commands[tag].active, 0);
752 command->comp_func(port,
753 tag,
754 command->comp_data,
755 0);
756 } else {
757 dev_err(&port->dd->pdev->dev,
758 "Missing completion func for tag %d",
759 tag);
760 if (mtip_check_surprise_removal(dd->pdev)) {
761 mtip_command_cleanup(dd);
762 /* don't proceed further */
763 return;
764 }
765 }
766 }
767 }
768 print_tags(dd, "TFE tags completed:", tagaccum);
769
770 /* Restart the port */
771 mdelay(20);
772 mtip_restart_port(port);
773
774 /* clear the tag accumulator */
775 memset(tagaccum, 0, SLOTBITS_IN_LONGS * sizeof(long));
776
777 /* Loop through all the groups */
778 for (group = 0; group < dd->slot_groups; group++) {
779 for (bit = 0; bit < 32; bit++) {
780 reissue = 1;
781 tag = (group << 5) + bit;
782
783 /* If the active bit is set re-issue the command */
784 if (atomic_read(&port->commands[tag].active) == 0)
785 continue;
786
787 fis = (struct host_to_dev_fis *)
788 port->commands[tag].command;
789
790 /* Should re-issue? */
791 if (tag == MTIP_TAG_INTERNAL ||
792 fis->command == ATA_CMD_SET_FEATURES)
793 reissue = 0;
794
795 /*
796 * First check if this command has
797 * exceeded its retries.
798 */
799 if (reissue &&
800 (port->commands[tag].retries-- > 0)) {
801
802 set_bit(tag, tagaccum);
803
804 /* Update the timeout value. */
805 port->commands[tag].comp_time =
806 jiffies + msecs_to_jiffies(
807 MTIP_NCQ_COMMAND_TIMEOUT_MS);
808 /* Re-issue the command. */
809 mtip_issue_ncq_command(port, tag);
810
811 continue;
812 }
813
814 /* Retire a command that will not be reissued */
815 dev_warn(&port->dd->pdev->dev,
816 "retiring tag %d\n", tag);
817 atomic_set(&port->commands[tag].active, 0);
818
819 if (port->commands[tag].comp_func)
820 port->commands[tag].comp_func(
821 port,
822 tag,
823 port->commands[tag].comp_data,
824 PORT_IRQ_TF_ERR);
825 else
826 dev_warn(&port->dd->pdev->dev,
827 "Bad completion for tag %d\n",
828 tag);
829 }
830 }
831 print_tags(dd, "TFE tags reissued:", tagaccum);
832
833 /* clear eh_active */
834 clear_bit(MTIP_FLAG_EH_ACTIVE_BIT, &port->flags);
835 wake_up_interruptible(&port->svc_wait);
836
837 mod_timer(&port->cmd_timer,
838 jiffies + msecs_to_jiffies(MTIP_TIMEOUT_CHECK_PERIOD));
839}
840
841/*
842 * Handle a set device bits interrupt
843 */
844static inline void mtip_process_sdbf(struct driver_data *dd)
845{
846 struct mtip_port *port = dd->port;
847 int group, tag, bit;
848 u32 completed;
849 struct mtip_cmd *command;
850
851 /* walk all bits in all slot groups */
852 for (group = 0; group < dd->slot_groups; group++) {
853 completed = readl(port->completed[group]);
854
855 /* clear completed status register in the hardware.*/
856 writel(completed, port->completed[group]);
857
858 /* Process completed commands. */
859 for (bit = 0;
860 (bit < 32) && completed;
861 bit++, completed >>= 1) {
862 if (completed & 0x01) {
863 tag = (group << 5) | bit;
864
865 /* skip internal command slot. */
866 if (unlikely(tag == MTIP_TAG_INTERNAL))
867 continue;
868
869 command = &port->commands[tag];
870 /* make internal callback */
871 if (likely(command->comp_func)) {
872 command->comp_func(
873 port,
874 tag,
875 command->comp_data,
876 0);
877 } else {
878 dev_warn(&dd->pdev->dev,
879 "Null completion "
880 "for tag %d",
881 tag);
882
883 if (mtip_check_surprise_removal(
884 dd->pdev)) {
885 mtip_command_cleanup(dd);
886 return;
887 }
888 }
889 }
890 }
891 }
892}
893
894/*
895 * Process legacy pio and d2h interrupts
896 */
897static inline void mtip_process_legacy(struct driver_data *dd, u32 port_stat)
898{
899 struct mtip_port *port = dd->port;
900 struct mtip_cmd *cmd = &port->commands[MTIP_TAG_INTERNAL];
901
902 if (test_bit(MTIP_FLAG_IC_ACTIVE_BIT, &port->flags) &&
903 (cmd != NULL) && !(readl(port->cmd_issue[MTIP_TAG_INTERNAL])
904 & (1 << MTIP_TAG_INTERNAL))) {
905 if (cmd->comp_func) {
906 cmd->comp_func(port,
907 MTIP_TAG_INTERNAL,
908 cmd->comp_data,
909 0);
910 return;
911 }
912 }
913
914 dev_warn(&dd->pdev->dev, "IRQ status 0x%x ignored.\n", port_stat);
915
916 return;
917}
918
919/*
920 * Demux and handle errors
921 */
922static inline void mtip_process_errors(struct driver_data *dd, u32 port_stat)
923{
924 if (likely(port_stat & (PORT_IRQ_TF_ERR | PORT_IRQ_IF_ERR)))
925 mtip_handle_tfe(dd);
926
927 if (unlikely(port_stat & PORT_IRQ_CONNECT)) {
928 dev_warn(&dd->pdev->dev,
929 "Clearing PxSERR.DIAG.x\n");
930 writel((1 << 26), dd->port->mmio + PORT_SCR_ERR);
931 }
932
933 if (unlikely(port_stat & PORT_IRQ_PHYRDY)) {
934 dev_warn(&dd->pdev->dev,
935 "Clearing PxSERR.DIAG.n\n");
936 writel((1 << 16), dd->port->mmio + PORT_SCR_ERR);
937 }
938
939 if (unlikely(port_stat & ~PORT_IRQ_HANDLED)) {
940 dev_warn(&dd->pdev->dev,
941 "Port stat errors %x unhandled\n",
942 (port_stat & ~PORT_IRQ_HANDLED));
943 }
944}
945
946static inline irqreturn_t mtip_handle_irq(struct driver_data *data)
947{
948 struct driver_data *dd = (struct driver_data *) data;
949 struct mtip_port *port = dd->port;
950 u32 hba_stat, port_stat;
951 int rv = IRQ_NONE;
952
953 hba_stat = readl(dd->mmio + HOST_IRQ_STAT);
954 if (hba_stat) {
955 rv = IRQ_HANDLED;
956
957 /* Acknowledge the interrupt status on the port.*/
958 port_stat = readl(port->mmio + PORT_IRQ_STAT);
959 writel(port_stat, port->mmio + PORT_IRQ_STAT);
960
961 /* Demux port status */
962 if (likely(port_stat & PORT_IRQ_SDB_FIS))
963 mtip_process_sdbf(dd);
964
965 if (unlikely(port_stat & PORT_IRQ_ERR)) {
966 if (unlikely(mtip_check_surprise_removal(dd->pdev))) {
967 mtip_command_cleanup(dd);
968 /* don't proceed further */
969 return IRQ_HANDLED;
970 }
971
972 mtip_process_errors(dd, port_stat & PORT_IRQ_ERR);
973 }
974
975 if (unlikely(port_stat & PORT_IRQ_LEGACY))
976 mtip_process_legacy(dd, port_stat & PORT_IRQ_LEGACY);
977 }
978
979 /* acknowledge interrupt */
980 writel(hba_stat, dd->mmio + HOST_IRQ_STAT);
981
982 return rv;
983}
984
985/*
986 * Wrapper for mtip_handle_irq
987 * (ignores return code)
988 */
989static void mtip_tasklet(unsigned long data)
990{
991 mtip_handle_irq((struct driver_data *) data);
992}
993
994/*
995 * HBA interrupt subroutine.
996 *
997 * @irq IRQ number.
998 * @instance Pointer to the driver data structure.
999 *
1000 * return value
1001 * IRQ_HANDLED A HBA interrupt was pending and handled.
1002 * IRQ_NONE This interrupt was not for the HBA.
1003 */
1004static irqreturn_t mtip_irq_handler(int irq, void *instance)
1005{
1006 struct driver_data *dd = instance;
1007 tasklet_schedule(&dd->tasklet);
1008 return IRQ_HANDLED;
1009}
1010
1011static void mtip_issue_non_ncq_command(struct mtip_port *port, int tag)
1012{
1013 atomic_set(&port->commands[tag].active, 1);
1014 writel(1 << MTIP_TAG_BIT(tag),
1015 port->cmd_issue[MTIP_TAG_INDEX(tag)]);
1016}
1017
1018/*
1019 * Wait for port to quiesce
1020 *
1021 * @port Pointer to port data structure
1022 * @timeout Max duration to wait (ms)
1023 *
1024 * return value
1025 * 0 Success
1026 * -EBUSY Commands still active
1027 */
1028static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout)
1029{
1030 unsigned long to;
1031 unsigned int n;
1032 unsigned int active = 1;
1033
1034 to = jiffies + msecs_to_jiffies(timeout);
1035 do {
1036 if (test_bit(MTIP_FLAG_SVC_THD_ACTIVE_BIT, &port->flags) &&
1037 test_bit(MTIP_FLAG_ISSUE_CMDS_BIT, &port->flags)) {
1038 msleep(20);
1039 continue; /* svc thd is actively issuing commands */
1040 }
1041 /*
1042 * Ignore s_active bit 0 of array element 0.
1043 * This bit will always be set
1044 */
1045 active = readl(port->s_active[0]) & 0xFFFFFFFE;
1046 for (n = 1; n < port->dd->slot_groups; n++)
1047 active |= readl(port->s_active[n]);
1048
1049 if (!active)
1050 break;
1051
1052 msleep(20);
1053 } while (time_before(jiffies, to));
1054
1055 return active ? -EBUSY : 0;
1056}
1057
1058/*
1059 * Execute an internal command and wait for the completion.
1060 *
1061 * @port Pointer to the port data structure.
1062 * @fis Pointer to the FIS that describes the command.
1063 * @fis_len Length in WORDS of the FIS.
1064 * @buffer DMA accessible for command data.
1065 * @buf_len Length, in bytes, of the data buffer.
1066 * @opts Command header options, excluding the FIS length
1067 * and the number of PRD entries.
1068 * @timeout Time in ms to wait for the command to complete.
1069 *
1070 * return value
1071 * 0 Command completed successfully.
1072 * -EFAULT The buffer address is not correctly aligned.
1073 * -EBUSY Internal command or other IO in progress.
1074 * -EAGAIN Time out waiting for command to complete.
1075 */
1076static int mtip_exec_internal_command(struct mtip_port *port,
1077 void *fis,
1078 int fis_len,
1079 dma_addr_t buffer,
1080 int buf_len,
1081 u32 opts,
1082 gfp_t atomic,
1083 unsigned long timeout)
1084{
1085 struct mtip_cmd_sg *command_sg;
1086 DECLARE_COMPLETION_ONSTACK(wait);
1087 int rv = 0;
1088 struct mtip_cmd *int_cmd = &port->commands[MTIP_TAG_INTERNAL];
1089
1090 /* Make sure the buffer is 8 byte aligned. This is asic specific. */
1091 if (buffer & 0x00000007) {
1092 dev_err(&port->dd->pdev->dev,
1093 "SG buffer is not 8 byte aligned\n");
1094 return -EFAULT;
1095 }
1096
1097 /* Only one internal command should be running at a time */
1098 if (test_and_set_bit(MTIP_TAG_INTERNAL, port->allocated)) {
1099 dev_warn(&port->dd->pdev->dev,
1100 "Internal command already active\n");
1101 return -EBUSY;
1102 }
1103 set_bit(MTIP_FLAG_IC_ACTIVE_BIT, &port->flags);
1104
1105 if (atomic == GFP_KERNEL) {
1106 /* wait for io to complete if non atomic */
1107 if (mtip_quiesce_io(port, 5000) < 0) {
1108 dev_warn(&port->dd->pdev->dev,
1109 "Failed to quiesce IO\n");
1110 release_slot(port, MTIP_TAG_INTERNAL);
1111 clear_bit(MTIP_FLAG_IC_ACTIVE_BIT, &port->flags);
1112 wake_up_interruptible(&port->svc_wait);
1113 return -EBUSY;
1114 }
1115
1116 /* Set the completion function and data for the command. */
1117 int_cmd->comp_data = &wait;
1118 int_cmd->comp_func = mtip_completion;
1119
1120 } else {
1121 /* Clear completion - we're going to poll */
1122 int_cmd->comp_data = NULL;
1123 int_cmd->comp_func = NULL;
1124 }
1125
1126 /* Copy the command to the command table */
1127 memcpy(int_cmd->command, fis, fis_len*4);
1128
1129 /* Populate the SG list */
1130 int_cmd->command_header->opts =
1131 __force_bit2int cpu_to_le32(opts | fis_len);
1132 if (buf_len) {
1133 command_sg = int_cmd->command + AHCI_CMD_TBL_HDR_SZ;
1134
1135 command_sg->info =
1136 __force_bit2int cpu_to_le32((buf_len-1) & 0x3FFFFF);
1137 command_sg->dba =
1138 __force_bit2int cpu_to_le32(buffer & 0xFFFFFFFF);
1139 command_sg->dba_upper =
1140 __force_bit2int cpu_to_le32((buffer >> 16) >> 16);
1141
1142 int_cmd->command_header->opts |=
1143 __force_bit2int cpu_to_le32((1 << 16));
1144 }
1145
1146 /* Populate the command header */
1147 int_cmd->command_header->byte_count = 0;
1148
1149 /* Issue the command to the hardware */
1150 mtip_issue_non_ncq_command(port, MTIP_TAG_INTERNAL);
1151
1152 /* Poll if atomic, wait_for_completion otherwise */
1153 if (atomic == GFP_KERNEL) {
1154 /* Wait for the command to complete or timeout. */
1155 if (wait_for_completion_timeout(
1156 &wait,
1157 msecs_to_jiffies(timeout)) == 0) {
1158 dev_err(&port->dd->pdev->dev,
1159 "Internal command did not complete [%d] "
1160 "within timeout of %lu ms\n",
1161 atomic, timeout);
1162 rv = -EAGAIN;
1163 }
1164
1165 if (readl(port->cmd_issue[MTIP_TAG_INTERNAL])
1166 & (1 << MTIP_TAG_INTERNAL)) {
1167 dev_warn(&port->dd->pdev->dev,
1168 "Retiring internal command but CI is 1.\n");
1169 }
1170
1171 } else {
1172 /* Spin for <timeout> checking if command still outstanding */
1173 timeout = jiffies + msecs_to_jiffies(timeout);
1174
1175 while ((readl(
1176 port->cmd_issue[MTIP_TAG_INTERNAL])
1177 & (1 << MTIP_TAG_INTERNAL))
1178 && time_before(jiffies, timeout))
1179 ;
1180
1181 if (readl(port->cmd_issue[MTIP_TAG_INTERNAL])
1182 & (1 << MTIP_TAG_INTERNAL)) {
1183 dev_err(&port->dd->pdev->dev,
1184 "Internal command did not complete [%d]\n",
1185 atomic);
1186 rv = -EAGAIN;
1187 }
1188 }
1189
1190 /* Clear the allocated and active bits for the internal command. */
1191 atomic_set(&int_cmd->active, 0);
1192 release_slot(port, MTIP_TAG_INTERNAL);
1193 clear_bit(MTIP_FLAG_IC_ACTIVE_BIT, &port->flags);
1194 wake_up_interruptible(&port->svc_wait);
1195
1196 return rv;
1197}
1198
1199/*
1200 * Byte-swap ATA ID strings.
1201 *
1202 * ATA identify data contains strings in byte-swapped 16-bit words.
1203 * They must be swapped (on all architectures) to be usable as C strings.
1204 * This function swaps bytes in-place.
1205 *
1206 * @buf The buffer location of the string
1207 * @len The number of bytes to swap
1208 *
1209 * return value
1210 * None
1211 */
1212static inline void ata_swap_string(u16 *buf, unsigned int len)
1213{
1214 int i;
1215 for (i = 0; i < (len/2); i++)
1216 be16_to_cpus(&buf[i]);
1217}
1218
1219/*
1220 * Request the device identity information.
1221 *
1222 * If a user space buffer is not specified, i.e. is NULL, the
1223 * identify information is still read from the drive and placed
1224 * into the identify data buffer (@e port->identify) in the
1225 * port data structure.
1226 * When the identify buffer contains valid identify information @e
1227 * port->identify_valid is non-zero.
1228 *
1229 * @port Pointer to the port structure.
1230 * @user_buffer A user space buffer where the identify data should be
1231 * copied.
1232 *
1233 * return value
1234 * 0 Command completed successfully.
1235 * -EFAULT An error occurred while coping data to the user buffer.
1236 * -1 Command failed.
1237 */
1238static int mtip_get_identify(struct mtip_port *port, void __user *user_buffer)
1239{
1240 int rv = 0;
1241 struct host_to_dev_fis fis;
1242
1243 /* Build the FIS. */
1244 memset(&fis, 0, sizeof(struct host_to_dev_fis));
1245 fis.type = 0x27;
1246 fis.opts = 1 << 7;
1247 fis.command = ATA_CMD_ID_ATA;
1248
1249 /* Set the identify information as invalid. */
1250 port->identify_valid = 0;
1251
1252 /* Clear the identify information. */
1253 memset(port->identify, 0, sizeof(u16) * ATA_ID_WORDS);
1254
1255 /* Execute the command. */
1256 if (mtip_exec_internal_command(port,
1257 &fis,
1258 5,
1259 port->identify_dma,
1260 sizeof(u16) * ATA_ID_WORDS,
1261 0,
1262 GFP_KERNEL,
1263 MTIP_INTERNAL_COMMAND_TIMEOUT_MS)
1264 < 0) {
1265 rv = -1;
1266 goto out;
1267 }
1268
1269 /*
1270 * Perform any necessary byte-swapping. Yes, the kernel does in fact
1271 * perform field-sensitive swapping on the string fields.
1272 * See the kernel use of ata_id_string() for proof of this.
1273 */
1274#ifdef __LITTLE_ENDIAN
1275 ata_swap_string(port->identify + 27, 40); /* model string*/
1276 ata_swap_string(port->identify + 23, 8); /* firmware string*/
1277 ata_swap_string(port->identify + 10, 20); /* serial# string*/
1278#else
1279 {
1280 int i;
1281 for (i = 0; i < ATA_ID_WORDS; i++)
1282 port->identify[i] = le16_to_cpu(port->identify[i]);
1283 }
1284#endif
1285
1286 /* Set the identify buffer as valid. */
1287 port->identify_valid = 1;
1288
1289 if (user_buffer) {
1290 if (copy_to_user(
1291 user_buffer,
1292 port->identify,
1293 ATA_ID_WORDS * sizeof(u16))) {
1294 rv = -EFAULT;
1295 goto out;
1296 }
1297 }
1298
1299out:
1300 return rv;
1301}
1302
1303/*
1304 * Issue a standby immediate command to the device.
1305 *
1306 * @port Pointer to the port structure.
1307 *
1308 * return value
1309 * 0 Command was executed successfully.
1310 * -1 An error occurred while executing the command.
1311 */
1312static int mtip_standby_immediate(struct mtip_port *port)
1313{
1314 int rv;
1315 struct host_to_dev_fis fis;
1316
1317 /* Build the FIS. */
1318 memset(&fis, 0, sizeof(struct host_to_dev_fis));
1319 fis.type = 0x27;
1320 fis.opts = 1 << 7;
1321 fis.command = ATA_CMD_STANDBYNOW1;
1322
1323 /* Execute the command. Use a 15-second timeout for large drives. */
1324 rv = mtip_exec_internal_command(port,
1325 &fis,
1326 5,
1327 0,
1328 0,
1329 0,
1330 GFP_KERNEL,
1331 15000);
1332
1333 return rv;
1334}
1335
1336/*
1337 * Get the drive capacity.
1338 *
1339 * @dd Pointer to the device data structure.
1340 * @sectors Pointer to the variable that will receive the sector count.
1341 *
1342 * return value
1343 * 1 Capacity was returned successfully.
1344 * 0 The identify information is invalid.
1345 */
1346static bool mtip_hw_get_capacity(struct driver_data *dd, sector_t *sectors)
1347{
1348 struct mtip_port *port = dd->port;
1349 u64 total, raw0, raw1, raw2, raw3;
1350 raw0 = port->identify[100];
1351 raw1 = port->identify[101];
1352 raw2 = port->identify[102];
1353 raw3 = port->identify[103];
1354 total = raw0 | raw1<<16 | raw2<<32 | raw3<<48;
1355 *sectors = total;
1356 return (bool) !!port->identify_valid;
1357}
1358
1359/*
1360 * Reset the HBA.
1361 *
1362 * Resets the HBA by setting the HBA Reset bit in the Global
1363 * HBA Control register. After setting the HBA Reset bit the
1364 * function waits for 1 second before reading the HBA Reset
1365 * bit to make sure it has cleared. If HBA Reset is not clear
1366 * an error is returned. Cannot be used in non-blockable
1367 * context.
1368 *
1369 * @dd Pointer to the driver data structure.
1370 *
1371 * return value
1372 * 0 The reset was successful.
1373 * -1 The HBA Reset bit did not clear.
1374 */
1375static int mtip_hba_reset(struct driver_data *dd)
1376{
1377 mtip_deinit_port(dd->port);
1378
1379 /* Set the reset bit */
1380 writel(HOST_RESET, dd->mmio + HOST_CTL);
1381
1382 /* Flush */
1383 readl(dd->mmio + HOST_CTL);
1384
1385 /* Wait for reset to clear */
1386 ssleep(1);
1387
1388 /* Check the bit has cleared */
1389 if (readl(dd->mmio + HOST_CTL) & HOST_RESET) {
1390 dev_err(&dd->pdev->dev,
1391 "Reset bit did not clear.\n");
1392 return -1;
1393 }
1394
1395 return 0;
1396}
1397
1398/*
1399 * Display the identify command data.
1400 *
1401 * @port Pointer to the port data structure.
1402 *
1403 * return value
1404 * None
1405 */
1406static void mtip_dump_identify(struct mtip_port *port)
1407{
1408 sector_t sectors;
1409 unsigned short revid;
1410 char cbuf[42];
1411
1412 if (!port->identify_valid)
1413 return;
1414
1415 strlcpy(cbuf, (char *)(port->identify+10), 21);
1416 dev_info(&port->dd->pdev->dev,
1417 "Serial No.: %s\n", cbuf);
1418
1419 strlcpy(cbuf, (char *)(port->identify+23), 9);
1420 dev_info(&port->dd->pdev->dev,
1421 "Firmware Ver.: %s\n", cbuf);
1422
1423 strlcpy(cbuf, (char *)(port->identify+27), 41);
1424 dev_info(&port->dd->pdev->dev, "Model: %s\n", cbuf);
1425
1426 if (mtip_hw_get_capacity(port->dd, &sectors))
1427 dev_info(&port->dd->pdev->dev,
1428 "Capacity: %llu sectors (%llu MB)\n",
1429 (u64)sectors,
1430 ((u64)sectors) * ATA_SECT_SIZE >> 20);
1431
1432 pci_read_config_word(port->dd->pdev, PCI_REVISION_ID, &revid);
1433 switch (revid & 0xFF) {
1434 case 0x1:
1435 strlcpy(cbuf, "A0", 3);
1436 break;
1437 case 0x3:
1438 strlcpy(cbuf, "A2", 3);
1439 break;
1440 default:
1441 strlcpy(cbuf, "?", 2);
1442 break;
1443 }
1444 dev_info(&port->dd->pdev->dev,
1445 "Card Type: %s\n", cbuf);
1446}
1447
1448/*
1449 * Map the commands scatter list into the command table.
1450 *
1451 * @command Pointer to the command.
1452 * @nents Number of scatter list entries.
1453 *
1454 * return value
1455 * None
1456 */
1457static inline void fill_command_sg(struct driver_data *dd,
1458 struct mtip_cmd *command,
1459 int nents)
1460{
1461 int n;
1462 unsigned int dma_len;
1463 struct mtip_cmd_sg *command_sg;
1464 struct scatterlist *sg = command->sg;
1465
1466 command_sg = command->command + AHCI_CMD_TBL_HDR_SZ;
1467
1468 for (n = 0; n < nents; n++) {
1469 dma_len = sg_dma_len(sg);
1470 if (dma_len > 0x400000)
1471 dev_err(&dd->pdev->dev,
1472 "DMA segment length truncated\n");
1473 command_sg->info = __force_bit2int
1474 cpu_to_le32((dma_len-1) & 0x3FFFFF);
1475 command_sg->dba = __force_bit2int
1476 cpu_to_le32(sg_dma_address(sg));
1477 command_sg->dba_upper = __force_bit2int
1478 cpu_to_le32((sg_dma_address(sg) >> 16) >> 16);
1479 command_sg++;
1480 sg++;
1481 }
1482}
1483
1484/*
1485 * @brief Execute a drive command.
1486 *
1487 * return value 0 The command completed successfully.
1488 * return value -1 An error occurred while executing the command.
1489 */
1490static int exec_drive_task(struct mtip_port *port, u8 *command)
1491{
1492 struct host_to_dev_fis fis;
1493 struct host_to_dev_fis *reply = (port->rxfis + RX_FIS_D2H_REG);
1494
1495 /* Build the FIS. */
1496 memset(&fis, 0, sizeof(struct host_to_dev_fis));
1497 fis.type = 0x27;
1498 fis.opts = 1 << 7;
1499 fis.command = command[0];
1500 fis.features = command[1];
1501 fis.sect_count = command[2];
1502 fis.sector = command[3];
1503 fis.cyl_low = command[4];
1504 fis.cyl_hi = command[5];
1505 fis.device = command[6] & ~0x10; /* Clear the dev bit*/
1506
1507
1508 dbg_printk(MTIP_DRV_NAME "%s: User Command: cmd %x, feat %x, "
1509 "nsect %x, sect %x, lcyl %x, "
1510 "hcyl %x, sel %x\n",
1511 __func__,
1512 command[0],
1513 command[1],
1514 command[2],
1515 command[3],
1516 command[4],
1517 command[5],
1518 command[6]);
1519
1520 /* Execute the command. */
1521 if (mtip_exec_internal_command(port,
1522 &fis,
1523 5,
1524 0,
1525 0,
1526 0,
1527 GFP_KERNEL,
1528 MTIP_IOCTL_COMMAND_TIMEOUT_MS) < 0) {
1529 return -1;
1530 }
1531
1532 command[0] = reply->command; /* Status*/
1533 command[1] = reply->features; /* Error*/
1534 command[4] = reply->cyl_low;
1535 command[5] = reply->cyl_hi;
1536
1537 dbg_printk(MTIP_DRV_NAME "%s: Completion Status: stat %x, "
1538 "err %x , cyl_lo %x cyl_hi %x\n",
1539 __func__,
1540 command[0],
1541 command[1],
1542 command[4],
1543 command[5]);
1544
1545 return 0;
1546}
1547
1548/*
1549 * @brief Execute a drive command.
1550 *
1551 * @param port Pointer to the port data structure.
1552 * @param command Pointer to the user specified command parameters.
1553 * @param user_buffer Pointer to the user space buffer where read sector
1554 * data should be copied.
1555 *
1556 * return value 0 The command completed successfully.
1557 * return value -EFAULT An error occurred while copying the completion
1558 * data to the user space buffer.
1559 * return value -1 An error occurred while executing the command.
1560 */
1561static int exec_drive_command(struct mtip_port *port, u8 *command,
1562 void __user *user_buffer)
1563{
1564 struct host_to_dev_fis fis;
1565 struct host_to_dev_fis *reply = (port->rxfis + RX_FIS_D2H_REG);
1566
1567 /* Build the FIS. */
1568 memset(&fis, 0, sizeof(struct host_to_dev_fis));
1569 fis.type = 0x27;
1570 fis.opts = 1 << 7;
1571 fis.command = command[0];
1572 fis.features = command[2];
1573 fis.sect_count = command[3];
1574 if (fis.command == ATA_CMD_SMART) {
1575 fis.sector = command[1];
1576 fis.cyl_low = 0x4F;
1577 fis.cyl_hi = 0xC2;
1578 }
1579
1580 dbg_printk(MTIP_DRV_NAME
1581 "%s: User Command: cmd %x, sect %x, "
1582 "feat %x, sectcnt %x\n",
1583 __func__,
1584 command[0],
1585 command[1],
1586 command[2],
1587 command[3]);
1588
1589 memset(port->sector_buffer, 0x00, ATA_SECT_SIZE);
1590
1591 /* Execute the command. */
1592 if (mtip_exec_internal_command(port,
1593 &fis,
1594 5,
1595 port->sector_buffer_dma,
1596 (command[3] != 0) ? ATA_SECT_SIZE : 0,
1597 0,
1598 GFP_KERNEL,
1599 MTIP_IOCTL_COMMAND_TIMEOUT_MS)
1600 < 0) {
1601 return -1;
1602 }
1603
1604 /* Collect the completion status. */
1605 command[0] = reply->command; /* Status*/
1606 command[1] = reply->features; /* Error*/
1607 command[2] = command[3];
1608
1609 dbg_printk(MTIP_DRV_NAME
1610 "%s: Completion Status: stat %x, "
1611 "err %x, cmd %x\n",
1612 __func__,
1613 command[0],
1614 command[1],
1615 command[2]);
1616
1617 if (user_buffer && command[3]) {
1618 if (copy_to_user(user_buffer,
1619 port->sector_buffer,
1620 ATA_SECT_SIZE * command[3])) {
1621 return -EFAULT;
1622 }
1623 }
1624
1625 return 0;
1626}
1627
1628/*
1629 * Indicates whether a command has a single sector payload.
1630 *
1631 * @command passed to the device to perform the certain event.
1632 * @features passed to the device to perform the certain event.
1633 *
1634 * return value
1635 * 1 command is one that always has a single sector payload,
1636 * regardless of the value in the Sector Count field.
1637 * 0 otherwise
1638 *
1639 */
1640static unsigned int implicit_sector(unsigned char command,
1641 unsigned char features)
1642{
1643 unsigned int rv = 0;
1644
1645 /* list of commands that have an implicit sector count of 1 */
1646 switch (command) {
1647 case ATA_CMD_SEC_SET_PASS:
1648 case ATA_CMD_SEC_UNLOCK:
1649 case ATA_CMD_SEC_ERASE_PREP:
1650 case ATA_CMD_SEC_ERASE_UNIT:
1651 case ATA_CMD_SEC_FREEZE_LOCK:
1652 case ATA_CMD_SEC_DISABLE_PASS:
1653 case ATA_CMD_PMP_READ:
1654 case ATA_CMD_PMP_WRITE:
1655 rv = 1;
1656 break;
1657 case ATA_CMD_SET_MAX:
1658 if (features == ATA_SET_MAX_UNLOCK)
1659 rv = 1;
1660 break;
1661 case ATA_CMD_SMART:
1662 if ((features == ATA_SMART_READ_VALUES) ||
1663 (features == ATA_SMART_READ_THRESHOLDS))
1664 rv = 1;
1665 break;
1666 case ATA_CMD_CONF_OVERLAY:
1667 if ((features == ATA_DCO_IDENTIFY) ||
1668 (features == ATA_DCO_SET))
1669 rv = 1;
1670 break;
1671 }
1672 return rv;
1673}
1674
1675/*
1676 * Executes a taskfile
1677 * See ide_taskfile_ioctl() for derivation
1678 */
1679static int exec_drive_taskfile(struct driver_data *dd,
1680 void __user *buf,
1681 ide_task_request_t *req_task,
1682 int outtotal)
1683{
1684 struct host_to_dev_fis fis;
1685 struct host_to_dev_fis *reply;
1686 u8 *outbuf = NULL;
1687 u8 *inbuf = NULL;
1688 dma_addr_t outbuf_dma = 0;
1689 dma_addr_t inbuf_dma = 0;
1690 dma_addr_t dma_buffer = 0;
1691 int err = 0;
1692 unsigned int taskin = 0;
1693 unsigned int taskout = 0;
1694 u8 nsect = 0;
1695 unsigned int timeout = MTIP_IOCTL_COMMAND_TIMEOUT_MS;
1696 unsigned int force_single_sector;
1697 unsigned int transfer_size;
1698 unsigned long task_file_data;
1699 int intotal = outtotal + req_task->out_size;
1700
1701 taskout = req_task->out_size;
1702 taskin = req_task->in_size;
1703 /* 130560 = 512 * 0xFF*/
1704 if (taskin > 130560 || taskout > 130560) {
1705 err = -EINVAL;
1706 goto abort;
1707 }
1708
1709 if (taskout) {
1710 outbuf = kzalloc(taskout, GFP_KERNEL);
1711 if (outbuf == NULL) {
1712 err = -ENOMEM;
1713 goto abort;
1714 }
1715 if (copy_from_user(outbuf, buf + outtotal, taskout)) {
1716 err = -EFAULT;
1717 goto abort;
1718 }
1719 outbuf_dma = pci_map_single(dd->pdev,
1720 outbuf,
1721 taskout,
1722 DMA_TO_DEVICE);
1723 if (outbuf_dma == 0) {
1724 err = -ENOMEM;
1725 goto abort;
1726 }
1727 dma_buffer = outbuf_dma;
1728 }
1729
1730 if (taskin) {
1731 inbuf = kzalloc(taskin, GFP_KERNEL);
1732 if (inbuf == NULL) {
1733 err = -ENOMEM;
1734 goto abort;
1735 }
1736
1737 if (copy_from_user(inbuf, buf + intotal, taskin)) {
1738 err = -EFAULT;
1739 goto abort;
1740 }
1741 inbuf_dma = pci_map_single(dd->pdev,
1742 inbuf,
1743 taskin, DMA_FROM_DEVICE);
1744 if (inbuf_dma == 0) {
1745 err = -ENOMEM;
1746 goto abort;
1747 }
1748 dma_buffer = inbuf_dma;
1749 }
1750
1751 /* only supports PIO and non-data commands from this ioctl. */
1752 switch (req_task->data_phase) {
1753 case TASKFILE_OUT:
1754 nsect = taskout / ATA_SECT_SIZE;
1755 reply = (dd->port->rxfis + RX_FIS_PIO_SETUP);
1756 break;
1757 case TASKFILE_IN:
1758 reply = (dd->port->rxfis + RX_FIS_PIO_SETUP);
1759 break;
1760 case TASKFILE_NO_DATA:
1761 reply = (dd->port->rxfis + RX_FIS_D2H_REG);
1762 break;
1763 default:
1764 err = -EINVAL;
1765 goto abort;
1766 }
1767
1768 /* Build the FIS. */
1769 memset(&fis, 0, sizeof(struct host_to_dev_fis));
1770
1771 fis.type = 0x27;
1772 fis.opts = 1 << 7;
1773 fis.command = req_task->io_ports[7];
1774 fis.features = req_task->io_ports[1];
1775 fis.sect_count = req_task->io_ports[2];
1776 fis.lba_low = req_task->io_ports[3];
1777 fis.lba_mid = req_task->io_ports[4];
1778 fis.lba_hi = req_task->io_ports[5];
1779 /* Clear the dev bit*/
1780 fis.device = req_task->io_ports[6] & ~0x10;
1781
1782 if ((req_task->in_flags.all == 0) && (req_task->out_flags.all & 1)) {
1783 req_task->in_flags.all =
1784 IDE_TASKFILE_STD_IN_FLAGS |
1785 (IDE_HOB_STD_IN_FLAGS << 8);
1786 fis.lba_low_ex = req_task->hob_ports[3];
1787 fis.lba_mid_ex = req_task->hob_ports[4];
1788 fis.lba_hi_ex = req_task->hob_ports[5];
1789 fis.features_ex = req_task->hob_ports[1];
1790 fis.sect_cnt_ex = req_task->hob_ports[2];
1791
1792 } else {
1793 req_task->in_flags.all = IDE_TASKFILE_STD_IN_FLAGS;
1794 }
1795
1796 force_single_sector = implicit_sector(fis.command, fis.features);
1797
1798 if ((taskin || taskout) && (!fis.sect_count)) {
1799 if (nsect)
1800 fis.sect_count = nsect;
1801 else {
1802 if (!force_single_sector) {
1803 dev_warn(&dd->pdev->dev,
1804 "data movement but "
1805 "sect_count is 0\n");
1806 err = -EINVAL;
1807 goto abort;
1808 }
1809 }
1810 }
1811
1812 dbg_printk(MTIP_DRV_NAME
1813 "taskfile: cmd %x, feat %x, nsect %x,"
1814 " sect/lbal %x, lcyl/lbam %x, hcyl/lbah %x,"
1815 " head/dev %x\n",
1816 fis.command,
1817 fis.features,
1818 fis.sect_count,
1819 fis.lba_low,
1820 fis.lba_mid,
1821 fis.lba_hi,
1822 fis.device);
1823
1824 switch (fis.command) {
1825 case ATA_CMD_DOWNLOAD_MICRO:
1826 /* Change timeout for Download Microcode to 60 seconds.*/
1827 timeout = 60000;
1828 break;
1829 case ATA_CMD_SEC_ERASE_UNIT:
1830 /* Change timeout for Security Erase Unit to 4 minutes.*/
1831 timeout = 240000;
1832 break;
1833 case ATA_CMD_STANDBYNOW1:
1834 /* Change timeout for standby immediate to 10 seconds.*/
1835 timeout = 10000;
1836 break;
1837 case 0xF7:
1838 case 0xFA:
1839 /* Change timeout for vendor unique command to 10 secs */
1840 timeout = 10000;
1841 break;
1842 case ATA_CMD_SMART:
1843 /* Change timeout for vendor unique command to 10 secs */
1844 timeout = 10000;
1845 break;
1846 default:
1847 timeout = MTIP_IOCTL_COMMAND_TIMEOUT_MS;
1848 break;
1849 }
1850
1851 /* Determine the correct transfer size.*/
1852 if (force_single_sector)
1853 transfer_size = ATA_SECT_SIZE;
1854 else
1855 transfer_size = ATA_SECT_SIZE * fis.sect_count;
1856
1857 /* Execute the command.*/
1858 if (mtip_exec_internal_command(dd->port,
1859 &fis,
1860 5,
1861 dma_buffer,
1862 transfer_size,
1863 0,
1864 GFP_KERNEL,
1865 timeout) < 0) {
1866 err = -EIO;
1867 goto abort;
1868 }
1869
1870 task_file_data = readl(dd->port->mmio+PORT_TFDATA);
1871
1872 if ((req_task->data_phase == TASKFILE_IN) && !(task_file_data & 1)) {
1873 reply = dd->port->rxfis + RX_FIS_PIO_SETUP;
1874 req_task->io_ports[7] = reply->control;
1875 } else {
1876 reply = dd->port->rxfis + RX_FIS_D2H_REG;
1877 req_task->io_ports[7] = reply->command;
1878 }
1879
1880 /* reclaim the DMA buffers.*/
1881 if (inbuf_dma)
1882 pci_unmap_single(dd->pdev, inbuf_dma,
1883 taskin, DMA_FROM_DEVICE);
1884 if (outbuf_dma)
1885 pci_unmap_single(dd->pdev, outbuf_dma,
1886 taskout, DMA_TO_DEVICE);
1887 inbuf_dma = 0;
1888 outbuf_dma = 0;
1889
1890 /* return the ATA registers to the caller.*/
1891 req_task->io_ports[1] = reply->features;
1892 req_task->io_ports[2] = reply->sect_count;
1893 req_task->io_ports[3] = reply->lba_low;
1894 req_task->io_ports[4] = reply->lba_mid;
1895 req_task->io_ports[5] = reply->lba_hi;
1896 req_task->io_ports[6] = reply->device;
1897
1898 if (req_task->out_flags.all & 1) {
1899
1900 req_task->hob_ports[3] = reply->lba_low_ex;
1901 req_task->hob_ports[4] = reply->lba_mid_ex;
1902 req_task->hob_ports[5] = reply->lba_hi_ex;
1903 req_task->hob_ports[1] = reply->features_ex;
1904 req_task->hob_ports[2] = reply->sect_cnt_ex;
1905 }
1906
1907 /* Com rest after secure erase or lowlevel format */
1908 if (((fis.command == ATA_CMD_SEC_ERASE_UNIT) ||
1909 ((fis.command == 0xFC) &&
1910 (fis.features == 0x27 || fis.features == 0x72 ||
1911 fis.features == 0x62 || fis.features == 0x26))) &&
1912 !(reply->command & 1)) {
1913 mtip_restart_port(dd->port);
1914 }
1915
1916 dbg_printk(MTIP_DRV_NAME
1917 "%s: Completion: stat %x,"
1918 "err %x, sect_cnt %x, lbalo %x,"
1919 "lbamid %x, lbahi %x, dev %x\n",
1920 __func__,
1921 req_task->io_ports[7],
1922 req_task->io_ports[1],
1923 req_task->io_ports[2],
1924 req_task->io_ports[3],
1925 req_task->io_ports[4],
1926 req_task->io_ports[5],
1927 req_task->io_ports[6]);
1928
1929 if (taskout) {
1930 if (copy_to_user(buf + outtotal, outbuf, taskout)) {
1931 err = -EFAULT;
1932 goto abort;
1933 }
1934 }
1935 if (taskin) {
1936 if (copy_to_user(buf + intotal, inbuf, taskin)) {
1937 err = -EFAULT;
1938 goto abort;
1939 }
1940 }
1941abort:
1942 if (inbuf_dma)
1943 pci_unmap_single(dd->pdev, inbuf_dma,
1944 taskin, DMA_FROM_DEVICE);
1945 if (outbuf_dma)
1946 pci_unmap_single(dd->pdev, outbuf_dma,
1947 taskout, DMA_TO_DEVICE);
1948 kfree(outbuf);
1949 kfree(inbuf);
1950
1951 return err;
1952}
1953
1954/*
1955 * Handle IOCTL calls from the Block Layer.
1956 *
1957 * This function is called by the Block Layer when it receives an IOCTL
1958 * command that it does not understand. If the IOCTL command is not supported
1959 * this function returns -ENOTTY.
1960 *
1961 * @dd Pointer to the driver data structure.
1962 * @cmd IOCTL command passed from the Block Layer.
1963 * @arg IOCTL argument passed from the Block Layer.
1964 *
1965 * return value
1966 * 0 The IOCTL completed successfully.
1967 * -ENOTTY The specified command is not supported.
1968 * -EFAULT An error occurred copying data to a user space buffer.
1969 * -EIO An error occurred while executing the command.
1970 */
1971static int mtip_hw_ioctl(struct driver_data *dd, unsigned int cmd,
1972 unsigned long arg)
1973{
1974 switch (cmd) {
1975 case HDIO_GET_IDENTITY:
1976 if (mtip_get_identify(dd->port, (void __user *) arg) < 0) {
1977 dev_warn(&dd->pdev->dev,
1978 "Unable to read identity\n");
1979 return -EIO;
1980 }
1981
1982 break;
1983 case HDIO_DRIVE_CMD:
1984 {
1985 u8 drive_command[4];
1986
1987 /* Copy the user command info to our buffer. */
1988 if (copy_from_user(drive_command,
1989 (void __user *) arg,
1990 sizeof(drive_command)))
1991 return -EFAULT;
1992
1993 /* Execute the drive command. */
1994 if (exec_drive_command(dd->port,
1995 drive_command,
1996 (void __user *) (arg+4)))
1997 return -EIO;
1998
1999 /* Copy the status back to the users buffer. */
2000 if (copy_to_user((void __user *) arg,
2001 drive_command,
2002 sizeof(drive_command)))
2003 return -EFAULT;
2004
2005 break;
2006 }
2007 case HDIO_DRIVE_TASK:
2008 {
2009 u8 drive_command[7];
2010
2011 /* Copy the user command info to our buffer. */
2012 if (copy_from_user(drive_command,
2013 (void __user *) arg,
2014 sizeof(drive_command)))
2015 return -EFAULT;
2016
2017 /* Execute the drive command. */
2018 if (exec_drive_task(dd->port, drive_command))
2019 return -EIO;
2020
2021 /* Copy the status back to the users buffer. */
2022 if (copy_to_user((void __user *) arg,
2023 drive_command,
2024 sizeof(drive_command)))
2025 return -EFAULT;
2026
2027 break;
2028 }
2029 case HDIO_DRIVE_TASKFILE: {
2030 ide_task_request_t req_task;
2031 int ret, outtotal;
2032
2033 if (copy_from_user(&req_task, (void __user *) arg,
2034 sizeof(req_task)))
2035 return -EFAULT;
2036
2037 outtotal = sizeof(req_task);
2038
2039 ret = exec_drive_taskfile(dd, (void __user *) arg,
2040 &req_task, outtotal);
2041
2042 if (copy_to_user((void __user *) arg, &req_task,
2043 sizeof(req_task)))
2044 return -EFAULT;
2045
2046 return ret;
2047 }
2048
2049 default:
2050 return -EINVAL;
2051 }
2052 return 0;
2053}
2054
2055/*
2056 * Submit an IO to the hw
2057 *
2058 * This function is called by the block layer to issue an io
2059 * to the device. Upon completion, the callback function will
2060 * be called with the data parameter passed as the callback data.
2061 *
2062 * @dd Pointer to the driver data structure.
2063 * @start First sector to read.
2064 * @nsect Number of sectors to read.
2065 * @nents Number of entries in scatter list for the read command.
2066 * @tag The tag of this read command.
2067 * @callback Pointer to the function that should be called
2068 * when the read completes.
2069 * @data Callback data passed to the callback function
2070 * when the read completes.
2071 * @barrier If non-zero, this command must be completed before
2072 * issuing any other commands.
2073 * @dir Direction (read or write)
2074 *
2075 * return value
2076 * None
2077 */
2078static void mtip_hw_submit_io(struct driver_data *dd, sector_t start,
2079 int nsect, int nents, int tag, void *callback,
2080 void *data, int barrier, int dir)
2081{
2082 struct host_to_dev_fis *fis;
2083 struct mtip_port *port = dd->port;
2084 struct mtip_cmd *command = &port->commands[tag];
2085
2086 /* Map the scatter list for DMA access */
2087 if (dir == READ)
2088 nents = dma_map_sg(&dd->pdev->dev, command->sg,
2089 nents, DMA_FROM_DEVICE);
2090 else
2091 nents = dma_map_sg(&dd->pdev->dev, command->sg,
2092 nents, DMA_TO_DEVICE);
2093
2094 command->scatter_ents = nents;
2095
2096 /*
2097 * The number of retries for this command before it is
2098 * reported as a failure to the upper layers.
2099 */
2100 command->retries = MTIP_MAX_RETRIES;
2101
2102 /* Fill out fis */
2103 fis = command->command;
2104 fis->type = 0x27;
2105 fis->opts = 1 << 7;
2106 fis->command =
2107 (dir == READ ? ATA_CMD_FPDMA_READ : ATA_CMD_FPDMA_WRITE);
2108 *((unsigned int *) &fis->lba_low) = (start & 0xFFFFFF);
2109 *((unsigned int *) &fis->lba_low_ex) = ((start >> 24) & 0xFFFFFF);
2110 fis->device = 1 << 6;
2111 if (barrier)
2112 fis->device |= FUA_BIT;
2113 fis->features = nsect & 0xFF;
2114 fis->features_ex = (nsect >> 8) & 0xFF;
2115 fis->sect_count = ((tag << 3) | (tag >> 5));
2116 fis->sect_cnt_ex = 0;
2117 fis->control = 0;
2118 fis->res2 = 0;
2119 fis->res3 = 0;
2120 fill_command_sg(dd, command, nents);
2121
2122 /* Populate the command header */
2123 command->command_header->opts =
2124 __force_bit2int cpu_to_le32(
2125 (nents << 16) | 5 | AHCI_CMD_PREFETCH);
2126 command->command_header->byte_count = 0;
2127
2128 /*
2129 * Set the completion function and data for the command
2130 * within this layer.
2131 */
2132 command->comp_data = dd;
2133 command->comp_func = mtip_async_complete;
2134 command->direction = (dir == READ ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
2135
2136 /*
2137 * Set the completion function and data for the command passed
2138 * from the upper layer.
2139 */
2140 command->async_data = data;
2141 command->async_callback = callback;
2142
2143 /*
2144 * To prevent this command from being issued
2145 * if an internal command is in progress or error handling is active.
2146 */
2147 if (unlikely(test_bit(MTIP_FLAG_IC_ACTIVE_BIT, &port->flags) ||
2148 test_bit(MTIP_FLAG_EH_ACTIVE_BIT, &port->flags))) {
2149 set_bit(tag, port->cmds_to_issue);
2150 set_bit(MTIP_FLAG_ISSUE_CMDS_BIT, &port->flags);
2151 return;
2152 }
2153
2154 /* Issue the command to the hardware */
2155 mtip_issue_ncq_command(port, tag);
2156
2157 /* Set the command's timeout value.*/
2158 port->commands[tag].comp_time = jiffies + msecs_to_jiffies(
2159 MTIP_NCQ_COMMAND_TIMEOUT_MS);
2160}
2161
2162/*
2163 * Release a command slot.
2164 *
2165 * @dd Pointer to the driver data structure.
2166 * @tag Slot tag
2167 *
2168 * return value
2169 * None
2170 */
2171static void mtip_hw_release_scatterlist(struct driver_data *dd, int tag)
2172{
2173 release_slot(dd->port, tag);
2174}
2175
2176/*
2177 * Obtain a command slot and return its associated scatter list.
2178 *
2179 * @dd Pointer to the driver data structure.
2180 * @tag Pointer to an int that will receive the allocated command
2181 * slot tag.
2182 *
2183 * return value
2184 * Pointer to the scatter list for the allocated command slot
2185 * or NULL if no command slots are available.
2186 */
2187static struct scatterlist *mtip_hw_get_scatterlist(struct driver_data *dd,
2188 int *tag)
2189{
2190 /*
2191 * It is possible that, even with this semaphore, a thread
2192 * may think that no command slots are available. Therefore, we
2193 * need to make an attempt to get_slot().
2194 */
2195 down(&dd->port->cmd_slot);
2196 *tag = get_slot(dd->port);
2197
2198 if (unlikely(*tag < 0))
2199 return NULL;
2200
2201 return dd->port->commands[*tag].sg;
2202}
2203
2204/*
2205 * Sysfs register/status dump.
2206 *
2207 * @dev Pointer to the device structure, passed by the kernrel.
2208 * @attr Pointer to the device_attribute structure passed by the kernel.
2209 * @buf Pointer to the char buffer that will receive the stats info.
2210 *
2211 * return value
2212 * The size, in bytes, of the data copied into buf.
2213 */
2214static ssize_t hw_show_registers(struct device *dev,
2215 struct device_attribute *attr,
2216 char *buf)
2217{
2218 u32 group_allocated;
2219 struct driver_data *dd = dev_to_disk(dev)->private_data;
2220 int size = 0;
2221 int n;
2222
2223 size += sprintf(&buf[size], "%s:\ns_active:\n", __func__);
2224
2225 for (n = 0; n < dd->slot_groups; n++)
2226 size += sprintf(&buf[size], "0x%08x\n",
2227 readl(dd->port->s_active[n]));
2228
2229 size += sprintf(&buf[size], "Command Issue:\n");
2230
2231 for (n = 0; n < dd->slot_groups; n++)
2232 size += sprintf(&buf[size], "0x%08x\n",
2233 readl(dd->port->cmd_issue[n]));
2234
2235 size += sprintf(&buf[size], "Allocated:\n");
2236
2237 for (n = 0; n < dd->slot_groups; n++) {
2238 if (sizeof(long) > sizeof(u32))
2239 group_allocated =
2240 dd->port->allocated[n/2] >> (32*(n&1));
2241 else
2242 group_allocated = dd->port->allocated[n];
2243 size += sprintf(&buf[size], "0x%08x\n",
2244 group_allocated);
2245 }
2246
2247 size += sprintf(&buf[size], "completed:\n");
2248
2249 for (n = 0; n < dd->slot_groups; n++)
2250 size += sprintf(&buf[size], "0x%08x\n",
2251 readl(dd->port->completed[n]));
2252
2253 size += sprintf(&buf[size], "PORT_IRQ_STAT 0x%08x\n",
2254 readl(dd->port->mmio + PORT_IRQ_STAT));
2255 size += sprintf(&buf[size], "HOST_IRQ_STAT 0x%08x\n",
2256 readl(dd->mmio + HOST_IRQ_STAT));
2257
2258 return size;
2259}
2260static DEVICE_ATTR(registers, S_IRUGO, hw_show_registers, NULL);
2261
2262/*
2263 * Create the sysfs related attributes.
2264 *
2265 * @dd Pointer to the driver data structure.
2266 * @kobj Pointer to the kobj for the block device.
2267 *
2268 * return value
2269 * 0 Operation completed successfully.
2270 * -EINVAL Invalid parameter.
2271 */
2272static int mtip_hw_sysfs_init(struct driver_data *dd, struct kobject *kobj)
2273{
2274 if (!kobj || !dd)
2275 return -EINVAL;
2276
2277 if (sysfs_create_file(kobj, &dev_attr_registers.attr))
2278 dev_warn(&dd->pdev->dev,
2279 "Error creating registers sysfs entry\n");
2280 return 0;
2281}
2282
2283/*
2284 * Remove the sysfs related attributes.
2285 *
2286 * @dd Pointer to the driver data structure.
2287 * @kobj Pointer to the kobj for the block device.
2288 *
2289 * return value
2290 * 0 Operation completed successfully.
2291 * -EINVAL Invalid parameter.
2292 */
2293static int mtip_hw_sysfs_exit(struct driver_data *dd, struct kobject *kobj)
2294{
2295 if (!kobj || !dd)
2296 return -EINVAL;
2297
2298 sysfs_remove_file(kobj, &dev_attr_registers.attr);
2299
2300 return 0;
2301}
2302
2303/*
2304 * Perform any init/resume time hardware setup
2305 *
2306 * @dd Pointer to the driver data structure.
2307 *
2308 * return value
2309 * None
2310 */
2311static inline void hba_setup(struct driver_data *dd)
2312{
2313 u32 hwdata;
2314 hwdata = readl(dd->mmio + HOST_HSORG);
2315
2316 /* interrupt bug workaround: use only 1 IS bit.*/
2317 writel(hwdata |
2318 HSORG_DISABLE_SLOTGRP_INTR |
2319 HSORG_DISABLE_SLOTGRP_PXIS,
2320 dd->mmio + HOST_HSORG);
2321}
2322
2323/*
2324 * Detect the details of the product, and store anything needed
2325 * into the driver data structure. This includes product type and
2326 * version and number of slot groups.
2327 *
2328 * @dd Pointer to the driver data structure.
2329 *
2330 * return value
2331 * None
2332 */
2333static void mtip_detect_product(struct driver_data *dd)
2334{
2335 u32 hwdata;
2336 unsigned int rev, slotgroups;
2337
2338 /*
2339 * HBA base + 0xFC [15:0] - vendor-specific hardware interface
2340 * info register:
2341 * [15:8] hardware/software interface rev#
2342 * [ 3] asic-style interface
2343 * [ 2:0] number of slot groups, minus 1 (only valid for asic-style).
2344 */
2345 hwdata = readl(dd->mmio + HOST_HSORG);
2346
2347 dd->product_type = MTIP_PRODUCT_UNKNOWN;
2348 dd->slot_groups = 1;
2349
2350 if (hwdata & 0x8) {
2351 dd->product_type = MTIP_PRODUCT_ASICFPGA;
2352 rev = (hwdata & HSORG_HWREV) >> 8;
2353 slotgroups = (hwdata & HSORG_SLOTGROUPS) + 1;
2354 dev_info(&dd->pdev->dev,
2355 "ASIC-FPGA design, HS rev 0x%x, "
2356 "%i slot groups [%i slots]\n",
2357 rev,
2358 slotgroups,
2359 slotgroups * 32);
2360
2361 if (slotgroups > MTIP_MAX_SLOT_GROUPS) {
2362 dev_warn(&dd->pdev->dev,
2363 "Warning: driver only supports "
2364 "%i slot groups.\n", MTIP_MAX_SLOT_GROUPS);
2365 slotgroups = MTIP_MAX_SLOT_GROUPS;
2366 }
2367 dd->slot_groups = slotgroups;
2368 return;
2369 }
2370
2371 dev_warn(&dd->pdev->dev, "Unrecognized product id\n");
2372}
2373
2374/*
2375 * Blocking wait for FTL rebuild to complete
2376 *
2377 * @dd Pointer to the DRIVER_DATA structure.
2378 *
2379 * return value
2380 * 0 FTL rebuild completed successfully
2381 * -EFAULT FTL rebuild error/timeout/interruption
2382 */
2383static int mtip_ftl_rebuild_poll(struct driver_data *dd)
2384{
2385 unsigned long timeout, cnt = 0, start;
2386
2387 dev_warn(&dd->pdev->dev,
2388 "FTL rebuild in progress. Polling for completion.\n");
2389
2390 start = jiffies;
2391 dd->ftlrebuildflag = 1;
2392 timeout = jiffies + msecs_to_jiffies(MTIP_FTL_REBUILD_TIMEOUT_MS);
2393
2394 do {
2395 if (mtip_check_surprise_removal(dd->pdev))
2396 return -EFAULT;
2397
2398 if (mtip_get_identify(dd->port, NULL) < 0)
2399 return -EFAULT;
2400
2401 if (*(dd->port->identify + MTIP_FTL_REBUILD_OFFSET) ==
2402 MTIP_FTL_REBUILD_MAGIC) {
2403 ssleep(1);
2404 /* Print message every 3 minutes */
2405 if (cnt++ >= 180) {
2406 dev_warn(&dd->pdev->dev,
2407 "FTL rebuild in progress (%d secs).\n",
2408 jiffies_to_msecs(jiffies - start) / 1000);
2409 cnt = 0;
2410 }
2411 } else {
2412 dev_warn(&dd->pdev->dev,
2413 "FTL rebuild complete (%d secs).\n",
2414 jiffies_to_msecs(jiffies - start) / 1000);
2415 dd->ftlrebuildflag = 0;
2416 mtip_block_initialize(dd);
2417 break;
2418 }
2419 ssleep(10);
2420 } while (time_before(jiffies, timeout));
2421
2422 /* Check for timeout */
2423 if (dd->ftlrebuildflag) {
2424 dev_err(&dd->pdev->dev,
2425 "Timed out waiting for FTL rebuild to complete (%d secs).\n",
2426 jiffies_to_msecs(jiffies - start) / 1000);
2427 return -EFAULT;
2428 }
2429
2430 return 0;
2431}
2432
2433/*
2434 * service thread to issue queued commands
2435 *
2436 * @data Pointer to the driver data structure.
2437 *
2438 * return value
2439 * 0
2440 */
2441
2442static int mtip_service_thread(void *data)
2443{
2444 struct driver_data *dd = (struct driver_data *)data;
2445 unsigned long slot, slot_start, slot_wrap;
2446 unsigned int num_cmd_slots = dd->slot_groups * 32;
2447 struct mtip_port *port = dd->port;
2448
2449 while (1) {
2450 /*
2451 * the condition is to check neither an internal command is
2452 * is in progress nor error handling is active
2453 */
2454 wait_event_interruptible(port->svc_wait, (port->flags) &&
2455 !test_bit(MTIP_FLAG_IC_ACTIVE_BIT, &port->flags) &&
2456 !test_bit(MTIP_FLAG_EH_ACTIVE_BIT, &port->flags));
2457
2458 if (kthread_should_stop())
2459 break;
2460
2461 set_bit(MTIP_FLAG_SVC_THD_ACTIVE_BIT, &port->flags);
2462 if (test_bit(MTIP_FLAG_ISSUE_CMDS_BIT, &port->flags)) {
2463 slot = 1;
2464 /* used to restrict the loop to one iteration */
2465 slot_start = num_cmd_slots;
2466 slot_wrap = 0;
2467 while (1) {
2468 slot = find_next_bit(port->cmds_to_issue,
2469 num_cmd_slots, slot);
2470 if (slot_wrap == 1) {
2471 if ((slot_start >= slot) ||
2472 (slot >= num_cmd_slots))
2473 break;
2474 }
2475 if (unlikely(slot_start == num_cmd_slots))
2476 slot_start = slot;
2477
2478 if (unlikely(slot == num_cmd_slots)) {
2479 slot = 1;
2480 slot_wrap = 1;
2481 continue;
2482 }
2483
2484 /* Issue the command to the hardware */
2485 mtip_issue_ncq_command(port, slot);
2486
2487 /* Set the command's timeout value.*/
2488 port->commands[slot].comp_time = jiffies +
2489 msecs_to_jiffies(MTIP_NCQ_COMMAND_TIMEOUT_MS);
2490
2491 clear_bit(slot, port->cmds_to_issue);
2492 }
2493
2494 clear_bit(MTIP_FLAG_ISSUE_CMDS_BIT, &port->flags);
2495 } else if (test_bit(MTIP_FLAG_REBUILD_BIT, &port->flags)) {
2496 mtip_ftl_rebuild_poll(dd);
2497 clear_bit(MTIP_FLAG_REBUILD_BIT, &port->flags);
2498 }
2499 clear_bit(MTIP_FLAG_SVC_THD_ACTIVE_BIT, &port->flags);
2500
2501 if (test_bit(MTIP_FLAG_SVC_THD_SHOULD_STOP_BIT, &port->flags))
2502 break;
2503 }
2504 return 0;
2505}
2506
2507/*
2508 * Called once for each card.
2509 *
2510 * @dd Pointer to the driver data structure.
2511 *
2512 * return value
2513 * 0 on success, else an error code.
2514 */
2515static int mtip_hw_init(struct driver_data *dd)
2516{
2517 int i;
2518 int rv;
2519 unsigned int num_command_slots;
2520
2521 dd->mmio = pcim_iomap_table(dd->pdev)[MTIP_ABAR];
2522
2523 mtip_detect_product(dd);
2524 if (dd->product_type == MTIP_PRODUCT_UNKNOWN) {
2525 rv = -EIO;
2526 goto out1;
2527 }
2528 num_command_slots = dd->slot_groups * 32;
2529
2530 hba_setup(dd);
2531
2532 tasklet_init(&dd->tasklet, mtip_tasklet, (unsigned long)dd);
2533
2534 dd->port = kzalloc(sizeof(struct mtip_port), GFP_KERNEL);
2535 if (!dd->port) {
2536 dev_err(&dd->pdev->dev,
2537 "Memory allocation: port structure\n");
2538 return -ENOMEM;
2539 }
2540
2541 /* Counting semaphore to track command slot usage */
2542 sema_init(&dd->port->cmd_slot, num_command_slots - 1);
2543
2544 /* Spinlock to prevent concurrent issue */
2545 spin_lock_init(&dd->port->cmd_issue_lock);
2546
2547 /* Set the port mmio base address. */
2548 dd->port->mmio = dd->mmio + PORT_OFFSET;
2549 dd->port->dd = dd;
2550
2551 /* Allocate memory for the command list. */
2552 dd->port->command_list =
2553 dmam_alloc_coherent(&dd->pdev->dev,
2554 HW_PORT_PRIV_DMA_SZ + (ATA_SECT_SIZE * 2),
2555 &dd->port->command_list_dma,
2556 GFP_KERNEL);
2557 if (!dd->port->command_list) {
2558 dev_err(&dd->pdev->dev,
2559 "Memory allocation: command list\n");
2560 rv = -ENOMEM;
2561 goto out1;
2562 }
2563
2564 /* Clear the memory we have allocated. */
2565 memset(dd->port->command_list,
2566 0,
2567 HW_PORT_PRIV_DMA_SZ + (ATA_SECT_SIZE * 2));
2568
2569 /* Setup the addresse of the RX FIS. */
2570 dd->port->rxfis = dd->port->command_list + HW_CMD_SLOT_SZ;
2571 dd->port->rxfis_dma = dd->port->command_list_dma + HW_CMD_SLOT_SZ;
2572
2573 /* Setup the address of the command tables. */
2574 dd->port->command_table = dd->port->rxfis + AHCI_RX_FIS_SZ;
2575 dd->port->command_tbl_dma = dd->port->rxfis_dma + AHCI_RX_FIS_SZ;
2576
2577 /* Setup the address of the identify data. */
2578 dd->port->identify = dd->port->command_table +
2579 HW_CMD_TBL_AR_SZ;
2580 dd->port->identify_dma = dd->port->command_tbl_dma +
2581 HW_CMD_TBL_AR_SZ;
2582
2583 /* Setup the address of the sector buffer. */
2584 dd->port->sector_buffer = (void *) dd->port->identify + ATA_SECT_SIZE;
2585 dd->port->sector_buffer_dma = dd->port->identify_dma + ATA_SECT_SIZE;
2586
2587 /* Point the command headers at the command tables. */
2588 for (i = 0; i < num_command_slots; i++) {
2589 dd->port->commands[i].command_header =
2590 dd->port->command_list +
2591 (sizeof(struct mtip_cmd_hdr) * i);
2592 dd->port->commands[i].command_header_dma =
2593 dd->port->command_list_dma +
2594 (sizeof(struct mtip_cmd_hdr) * i);
2595
2596 dd->port->commands[i].command =
2597 dd->port->command_table + (HW_CMD_TBL_SZ * i);
2598 dd->port->commands[i].command_dma =
2599 dd->port->command_tbl_dma + (HW_CMD_TBL_SZ * i);
2600
2601 if (readl(dd->mmio + HOST_CAP) & HOST_CAP_64)
2602 dd->port->commands[i].command_header->ctbau =
2603 __force_bit2int cpu_to_le32(
2604 (dd->port->commands[i].command_dma >> 16) >> 16);
2605 dd->port->commands[i].command_header->ctba =
2606 __force_bit2int cpu_to_le32(
2607 dd->port->commands[i].command_dma & 0xFFFFFFFF);
2608
2609 /*
2610 * If this is not done, a bug is reported by the stock
2611 * FC11 i386. Due to the fact that it has lots of kernel
2612 * debugging enabled.
2613 */
2614 sg_init_table(dd->port->commands[i].sg, MTIP_MAX_SG);
2615
2616 /* Mark all commands as currently inactive.*/
2617 atomic_set(&dd->port->commands[i].active, 0);
2618 }
2619
2620 /* Setup the pointers to the extended s_active and CI registers. */
2621 for (i = 0; i < dd->slot_groups; i++) {
2622 dd->port->s_active[i] =
2623 dd->port->mmio + i*0x80 + PORT_SCR_ACT;
2624 dd->port->cmd_issue[i] =
2625 dd->port->mmio + i*0x80 + PORT_COMMAND_ISSUE;
2626 dd->port->completed[i] =
2627 dd->port->mmio + i*0x80 + PORT_SDBV;
2628 }
2629
2630 /* Reset the HBA. */
2631 if (mtip_hba_reset(dd) < 0) {
2632 dev_err(&dd->pdev->dev,
2633 "Card did not reset within timeout\n");
2634 rv = -EIO;
2635 goto out2;
2636 }
2637
2638 mtip_init_port(dd->port);
2639 mtip_start_port(dd->port);
2640
2641 /* Setup the ISR and enable interrupts. */
2642 rv = devm_request_irq(&dd->pdev->dev,
2643 dd->pdev->irq,
2644 mtip_irq_handler,
2645 IRQF_SHARED,
2646 dev_driver_string(&dd->pdev->dev),
2647 dd);
2648
2649 if (rv) {
2650 dev_err(&dd->pdev->dev,
2651 "Unable to allocate IRQ %d\n", dd->pdev->irq);
2652 goto out2;
2653 }
2654
2655 /* Enable interrupts on the HBA. */
2656 writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN,
2657 dd->mmio + HOST_CTL);
2658
2659 init_timer(&dd->port->cmd_timer);
2660 init_waitqueue_head(&dd->port->svc_wait);
2661
2662 dd->port->cmd_timer.data = (unsigned long int) dd->port;
2663 dd->port->cmd_timer.function = mtip_timeout_function;
2664 mod_timer(&dd->port->cmd_timer,
2665 jiffies + msecs_to_jiffies(MTIP_TIMEOUT_CHECK_PERIOD));
2666
2667 if (mtip_get_identify(dd->port, NULL) < 0) {
2668 rv = -EFAULT;
2669 goto out3;
2670 }
2671
2672 if (*(dd->port->identify + MTIP_FTL_REBUILD_OFFSET) ==
2673 MTIP_FTL_REBUILD_MAGIC) {
2674 set_bit(MTIP_FLAG_REBUILD_BIT, &dd->port->flags);
2675 return MTIP_FTL_REBUILD_MAGIC;
2676 }
2677 mtip_dump_identify(dd->port);
2678 return rv;
2679
2680out3:
2681 del_timer_sync(&dd->port->cmd_timer);
2682
2683 /* Disable interrupts on the HBA. */
2684 writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN,
2685 dd->mmio + HOST_CTL);
2686
2687 /*Release the IRQ. */
2688 devm_free_irq(&dd->pdev->dev, dd->pdev->irq, dd);
2689
2690out2:
2691 mtip_deinit_port(dd->port);
2692
2693 /* Free the command/command header memory. */
2694 dmam_free_coherent(&dd->pdev->dev,
2695 HW_PORT_PRIV_DMA_SZ + (ATA_SECT_SIZE * 2),
2696 dd->port->command_list,
2697 dd->port->command_list_dma);
2698out1:
2699 /* Free the memory allocated for the for structure. */
2700 kfree(dd->port);
2701
2702 return rv;
2703}
2704
2705/*
2706 * Called to deinitialize an interface.
2707 *
2708 * @dd Pointer to the driver data structure.
2709 *
2710 * return value
2711 * 0
2712 */
2713static int mtip_hw_exit(struct driver_data *dd)
2714{
2715 /*
2716 * Send standby immediate (E0h) to the drive so that it
2717 * saves its state.
2718 */
2719 if (atomic_read(&dd->drv_cleanup_done) != true) {
2720
2721 mtip_standby_immediate(dd->port);
2722
2723 /* de-initialize the port. */
2724 mtip_deinit_port(dd->port);
2725
2726 /* Disable interrupts on the HBA. */
2727 writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN,
2728 dd->mmio + HOST_CTL);
2729 }
2730
2731 del_timer_sync(&dd->port->cmd_timer);
2732
2733 /* Release the IRQ. */
2734 devm_free_irq(&dd->pdev->dev, dd->pdev->irq, dd);
2735
2736 /* Stop the bottom half tasklet. */
2737 tasklet_kill(&dd->tasklet);
2738
2739 /* Free the command/command header memory. */
2740 dmam_free_coherent(&dd->pdev->dev,
2741 HW_PORT_PRIV_DMA_SZ + (ATA_SECT_SIZE * 2),
2742 dd->port->command_list,
2743 dd->port->command_list_dma);
2744 /* Free the memory allocated for the for structure. */
2745 kfree(dd->port);
2746
2747 return 0;
2748}
2749
2750/*
2751 * Issue a Standby Immediate command to the device.
2752 *
2753 * This function is called by the Block Layer just before the
2754 * system powers off during a shutdown.
2755 *
2756 * @dd Pointer to the driver data structure.
2757 *
2758 * return value
2759 * 0
2760 */
2761static int mtip_hw_shutdown(struct driver_data *dd)
2762{
2763 /*
2764 * Send standby immediate (E0h) to the drive so that it
2765 * saves its state.
2766 */
2767 mtip_standby_immediate(dd->port);
2768
2769 return 0;
2770}
2771
2772/*
2773 * Suspend function
2774 *
2775 * This function is called by the Block Layer just before the
2776 * system hibernates.
2777 *
2778 * @dd Pointer to the driver data structure.
2779 *
2780 * return value
2781 * 0 Suspend was successful
2782 * -EFAULT Suspend was not successful
2783 */
2784static int mtip_hw_suspend(struct driver_data *dd)
2785{
2786 /*
2787 * Send standby immediate (E0h) to the drive
2788 * so that it saves its state.
2789 */
2790 if (mtip_standby_immediate(dd->port) != 0) {
2791 dev_err(&dd->pdev->dev,
2792 "Failed standby-immediate command\n");
2793 return -EFAULT;
2794 }
2795
2796 /* Disable interrupts on the HBA.*/
2797 writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN,
2798 dd->mmio + HOST_CTL);
2799 mtip_deinit_port(dd->port);
2800
2801 return 0;
2802}
2803
2804/*
2805 * Resume function
2806 *
2807 * This function is called by the Block Layer as the
2808 * system resumes.
2809 *
2810 * @dd Pointer to the driver data structure.
2811 *
2812 * return value
2813 * 0 Resume was successful
2814 * -EFAULT Resume was not successful
2815 */
2816static int mtip_hw_resume(struct driver_data *dd)
2817{
2818 /* Perform any needed hardware setup steps */
2819 hba_setup(dd);
2820
2821 /* Reset the HBA */
2822 if (mtip_hba_reset(dd) != 0) {
2823 dev_err(&dd->pdev->dev,
2824 "Unable to reset the HBA\n");
2825 return -EFAULT;
2826 }
2827
2828 /*
2829 * Enable the port, DMA engine, and FIS reception specific
2830 * h/w in controller.
2831 */
2832 mtip_init_port(dd->port);
2833 mtip_start_port(dd->port);
2834
2835 /* Enable interrupts on the HBA.*/
2836 writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN,
2837 dd->mmio + HOST_CTL);
2838
2839 return 0;
2840}
2841
2842/*
2843 * Helper function for reusing disk name
2844 * upon hot insertion.
2845 */
2846static int rssd_disk_name_format(char *prefix,
2847 int index,
2848 char *buf,
2849 int buflen)
2850{
2851 const int base = 'z' - 'a' + 1;
2852 char *begin = buf + strlen(prefix);
2853 char *end = buf + buflen;
2854 char *p;
2855 int unit;
2856
2857 p = end - 1;
2858 *p = '\0';
2859 unit = base;
2860 do {
2861 if (p == begin)
2862 return -EINVAL;
2863 *--p = 'a' + (index % unit);
2864 index = (index / unit) - 1;
2865 } while (index >= 0);
2866
2867 memmove(begin, p, end - p);
2868 memcpy(buf, prefix, strlen(prefix));
2869
2870 return 0;
2871}
2872
2873/*
2874 * Block layer IOCTL handler.
2875 *
2876 * @dev Pointer to the block_device structure.
2877 * @mode ignored
2878 * @cmd IOCTL command passed from the user application.
2879 * @arg Argument passed from the user application.
2880 *
2881 * return value
2882 * 0 IOCTL completed successfully.
2883 * -ENOTTY IOCTL not supported or invalid driver data
2884 * structure pointer.
2885 */
2886static int mtip_block_ioctl(struct block_device *dev,
2887 fmode_t mode,
2888 unsigned cmd,
2889 unsigned long arg)
2890{
2891 struct driver_data *dd = dev->bd_disk->private_data;
2892
2893 if (!capable(CAP_SYS_ADMIN))
2894 return -EACCES;
2895
2896 if (!dd)
2897 return -ENOTTY;
2898
2899 switch (cmd) {
2900 case BLKFLSBUF:
2901 return -ENOTTY;
2902 default:
2903 return mtip_hw_ioctl(dd, cmd, arg);
2904 }
2905}
2906
2907#ifdef CONFIG_COMPAT
2908/*
2909 * Block layer compat IOCTL handler.
2910 *
2911 * @dev Pointer to the block_device structure.
2912 * @mode ignored
2913 * @cmd IOCTL command passed from the user application.
2914 * @arg Argument passed from the user application.
2915 *
2916 * return value
2917 * 0 IOCTL completed successfully.
2918 * -ENOTTY IOCTL not supported or invalid driver data
2919 * structure pointer.
2920 */
2921static int mtip_block_compat_ioctl(struct block_device *dev,
2922 fmode_t mode,
2923 unsigned cmd,
2924 unsigned long arg)
2925{
2926 struct driver_data *dd = dev->bd_disk->private_data;
2927
2928 if (!capable(CAP_SYS_ADMIN))
2929 return -EACCES;
2930
2931 if (!dd)
2932 return -ENOTTY;
2933
2934 switch (cmd) {
2935 case BLKFLSBUF:
2936 return -ENOTTY;
2937 case HDIO_DRIVE_TASKFILE: {
2938 struct mtip_compat_ide_task_request_s __user *compat_req_task;
2939 ide_task_request_t req_task;
2940 int compat_tasksize, outtotal, ret;
2941
2942 compat_tasksize =
2943 sizeof(struct mtip_compat_ide_task_request_s);
2944
2945 compat_req_task =
2946 (struct mtip_compat_ide_task_request_s __user *) arg;
2947
2948 if (copy_from_user(&req_task, (void __user *) arg,
2949 compat_tasksize - (2 * sizeof(compat_long_t))))
2950 return -EFAULT;
2951
2952 if (get_user(req_task.out_size, &compat_req_task->out_size))
2953 return -EFAULT;
2954
2955 if (get_user(req_task.in_size, &compat_req_task->in_size))
2956 return -EFAULT;
2957
2958 outtotal = sizeof(struct mtip_compat_ide_task_request_s);
2959
2960 ret = exec_drive_taskfile(dd, (void __user *) arg,
2961 &req_task, outtotal);
2962
2963 if (copy_to_user((void __user *) arg, &req_task,
2964 compat_tasksize -
2965 (2 * sizeof(compat_long_t))))
2966 return -EFAULT;
2967
2968 if (put_user(req_task.out_size, &compat_req_task->out_size))
2969 return -EFAULT;
2970
2971 if (put_user(req_task.in_size, &compat_req_task->in_size))
2972 return -EFAULT;
2973
2974 return ret;
2975 }
2976 default:
2977 return mtip_hw_ioctl(dd, cmd, arg);
2978 }
2979}
2980#endif
2981
2982/*
2983 * Obtain the geometry of the device.
2984 *
2985 * You may think that this function is obsolete, but some applications,
2986 * fdisk for example still used CHS values. This function describes the
2987 * device as having 224 heads and 56 sectors per cylinder. These values are
2988 * chosen so that each cylinder is aligned on a 4KB boundary. Since a
2989 * partition is described in terms of a start and end cylinder this means
2990 * that each partition is also 4KB aligned. Non-aligned partitions adversely
2991 * affects performance.
2992 *
2993 * @dev Pointer to the block_device strucutre.
2994 * @geo Pointer to a hd_geometry structure.
2995 *
2996 * return value
2997 * 0 Operation completed successfully.
2998 * -ENOTTY An error occurred while reading the drive capacity.
2999 */
3000static int mtip_block_getgeo(struct block_device *dev,
3001 struct hd_geometry *geo)
3002{
3003 struct driver_data *dd = dev->bd_disk->private_data;
3004 sector_t capacity;
3005
3006 if (!dd)
3007 return -ENOTTY;
3008
3009 if (!(mtip_hw_get_capacity(dd, &capacity))) {
3010 dev_warn(&dd->pdev->dev,
3011 "Could not get drive capacity.\n");
3012 return -ENOTTY;
3013 }
3014
3015 geo->heads = 224;
3016 geo->sectors = 56;
3017 sector_div(capacity, (geo->heads * geo->sectors));
3018 geo->cylinders = capacity;
3019 return 0;
3020}
3021
3022/*
3023 * Block device operation function.
3024 *
3025 * This structure contains pointers to the functions required by the block
3026 * layer.
3027 */
3028static const struct block_device_operations mtip_block_ops = {
3029 .ioctl = mtip_block_ioctl,
3030#ifdef CONFIG_COMPAT
3031 .compat_ioctl = mtip_block_compat_ioctl,
3032#endif
3033 .getgeo = mtip_block_getgeo,
3034 .owner = THIS_MODULE
3035};
3036
3037/*
3038 * Block layer make request function.
3039 *
3040 * This function is called by the kernel to process a BIO for
3041 * the P320 device.
3042 *
3043 * @queue Pointer to the request queue. Unused other than to obtain
3044 * the driver data structure.
3045 * @bio Pointer to the BIO.
3046 *
3047 */
3048static void mtip_make_request(struct request_queue *queue, struct bio *bio)
3049{
3050 struct driver_data *dd = queue->queuedata;
3051 struct scatterlist *sg;
3052 struct bio_vec *bvec;
3053 int nents = 0;
3054 int tag = 0;
3055
3056 if (unlikely(!bio_has_data(bio))) {
3057 blk_queue_flush(queue, 0);
3058 bio_endio(bio, 0);
3059 return;
3060 }
3061
3062 sg = mtip_hw_get_scatterlist(dd, &tag);
3063 if (likely(sg != NULL)) {
3064 blk_queue_bounce(queue, &bio);
3065
3066 if (unlikely((bio)->bi_vcnt > MTIP_MAX_SG)) {
3067 dev_warn(&dd->pdev->dev,
3068 "Maximum number of SGL entries exceeded");
3069 bio_io_error(bio);
3070 mtip_hw_release_scatterlist(dd, tag);
3071 return;
3072 }
3073
3074 /* Create the scatter list for this bio. */
3075 bio_for_each_segment(bvec, bio, nents) {
3076 sg_set_page(&sg[nents],
3077 bvec->bv_page,
3078 bvec->bv_len,
3079 bvec->bv_offset);
3080 }
3081
3082 /* Issue the read/write. */
3083 mtip_hw_submit_io(dd,
3084 bio->bi_sector,
3085 bio_sectors(bio),
3086 nents,
3087 tag,
3088 bio_endio,
3089 bio,
3090 bio->bi_rw & REQ_FUA,
3091 bio_data_dir(bio));
3092 } else
3093 bio_io_error(bio);
3094}
3095
3096/*
3097 * Block layer initialization function.
3098 *
3099 * This function is called once by the PCI layer for each P320
3100 * device that is connected to the system.
3101 *
3102 * @dd Pointer to the driver data structure.
3103 *
3104 * return value
3105 * 0 on success else an error code.
3106 */
3107static int mtip_block_initialize(struct driver_data *dd)
3108{
3109 int rv = 0, wait_for_rebuild = 0;
3110 sector_t capacity;
3111 unsigned int index = 0;
3112 struct kobject *kobj;
3113 unsigned char thd_name[16];
3114
3115 if (dd->disk)
3116 goto skip_create_disk; /* hw init done, before rebuild */
3117
3118 /* Initialize the protocol layer. */
3119 wait_for_rebuild = mtip_hw_init(dd);
3120 if (wait_for_rebuild < 0) {
3121 dev_err(&dd->pdev->dev,
3122 "Protocol layer initialization failed\n");
3123 rv = -EINVAL;
3124 goto protocol_init_error;
3125 }
3126
3127 dd->disk = alloc_disk(MTIP_MAX_MINORS);
3128 if (dd->disk == NULL) {
3129 dev_err(&dd->pdev->dev,
3130 "Unable to allocate gendisk structure\n");
3131 rv = -EINVAL;
3132 goto alloc_disk_error;
3133 }
3134
3135 /* Generate the disk name, implemented same as in sd.c */
3136 do {
3137 if (!ida_pre_get(&rssd_index_ida, GFP_KERNEL))
3138 goto ida_get_error;
3139
3140 spin_lock(&rssd_index_lock);
3141 rv = ida_get_new(&rssd_index_ida, &index);
3142 spin_unlock(&rssd_index_lock);
3143 } while (rv == -EAGAIN);
3144
3145 if (rv)
3146 goto ida_get_error;
3147
3148 rv = rssd_disk_name_format("rssd",
3149 index,
3150 dd->disk->disk_name,
3151 DISK_NAME_LEN);
3152 if (rv)
3153 goto disk_index_error;
3154
3155 dd->disk->driverfs_dev = &dd->pdev->dev;
3156 dd->disk->major = dd->major;
3157 dd->disk->first_minor = dd->instance * MTIP_MAX_MINORS;
3158 dd->disk->fops = &mtip_block_ops;
3159 dd->disk->private_data = dd;
3160 dd->index = index;
3161
3162 /*
3163 * if rebuild pending, start the service thread, and delay the block
3164 * queue creation and add_disk()
3165 */
3166 if (wait_for_rebuild == MTIP_FTL_REBUILD_MAGIC)
3167 goto start_service_thread;
3168
3169skip_create_disk:
3170 /* Allocate the request queue. */
3171 dd->queue = blk_alloc_queue(GFP_KERNEL);
3172 if (dd->queue == NULL) {
3173 dev_err(&dd->pdev->dev,
3174 "Unable to allocate request queue\n");
3175 rv = -ENOMEM;
3176 goto block_queue_alloc_init_error;
3177 }
3178
3179 /* Attach our request function to the request queue. */
3180 blk_queue_make_request(dd->queue, mtip_make_request);
3181
3182 dd->disk->queue = dd->queue;
3183 dd->queue->queuedata = dd;
3184
3185 /* Set device limits. */
3186 set_bit(QUEUE_FLAG_NONROT, &dd->queue->queue_flags);
3187 blk_queue_max_segments(dd->queue, MTIP_MAX_SG);
3188 blk_queue_physical_block_size(dd->queue, 4096);
3189 blk_queue_io_min(dd->queue, 4096);
3190 blk_queue_flush(dd->queue, 0);
3191
3192 /* Set the capacity of the device in 512 byte sectors. */
3193 if (!(mtip_hw_get_capacity(dd, &capacity))) {
3194 dev_warn(&dd->pdev->dev,
3195 "Could not read drive capacity\n");
3196 rv = -EIO;
3197 goto read_capacity_error;
3198 }
3199 set_capacity(dd->disk, capacity);
3200
3201 /* Enable the block device and add it to /dev */
3202 add_disk(dd->disk);
3203
3204 /*
3205 * Now that the disk is active, initialize any sysfs attributes
3206 * managed by the protocol layer.
3207 */
3208 kobj = kobject_get(&disk_to_dev(dd->disk)->kobj);
3209 if (kobj) {
3210 mtip_hw_sysfs_init(dd, kobj);
3211 kobject_put(kobj);
3212 }
3213
3214 if (dd->mtip_svc_handler)
3215 return rv; /* service thread created for handling rebuild */
3216
3217start_service_thread:
3218 sprintf(thd_name, "mtip_svc_thd_%02d", index);
3219
3220 dd->mtip_svc_handler = kthread_run(mtip_service_thread,
3221 dd, thd_name);
3222
3223 if (IS_ERR(dd->mtip_svc_handler)) {
3224 printk(KERN_ERR "mtip32xx: service thread failed to start\n");
3225 dd->mtip_svc_handler = NULL;
3226 rv = -EFAULT;
3227 goto kthread_run_error;
3228 }
3229
3230 return rv;
3231
3232kthread_run_error:
3233 /* Delete our gendisk. This also removes the device from /dev */
3234 del_gendisk(dd->disk);
3235
3236read_capacity_error:
3237 blk_cleanup_queue(dd->queue);
3238
3239block_queue_alloc_init_error:
3240disk_index_error:
3241 spin_lock(&rssd_index_lock);
3242 ida_remove(&rssd_index_ida, index);
3243 spin_unlock(&rssd_index_lock);
3244
3245ida_get_error:
3246 put_disk(dd->disk);
3247
3248alloc_disk_error:
3249 mtip_hw_exit(dd); /* De-initialize the protocol layer. */
3250
3251protocol_init_error:
3252 return rv;
3253}
3254
3255/*
3256 * Block layer deinitialization function.
3257 *
3258 * Called by the PCI layer as each P320 device is removed.
3259 *
3260 * @dd Pointer to the driver data structure.
3261 *
3262 * return value
3263 * 0
3264 */
3265static int mtip_block_remove(struct driver_data *dd)
3266{
3267 struct kobject *kobj;
3268
3269 if (dd->mtip_svc_handler) {
3270 set_bit(MTIP_FLAG_SVC_THD_SHOULD_STOP_BIT, &dd->port->flags);
3271 wake_up_interruptible(&dd->port->svc_wait);
3272 kthread_stop(dd->mtip_svc_handler);
3273 }
3274
3275 /* Clean up the sysfs attributes managed by the protocol layer. */
3276 kobj = kobject_get(&disk_to_dev(dd->disk)->kobj);
3277 if (kobj) {
3278 mtip_hw_sysfs_exit(dd, kobj);
3279 kobject_put(kobj);
3280 }
3281
3282 /*
3283 * Delete our gendisk structure. This also removes the device
3284 * from /dev
3285 */
3286 del_gendisk(dd->disk);
3287 blk_cleanup_queue(dd->queue);
3288 dd->disk = NULL;
3289 dd->queue = NULL;
3290
3291 /* De-initialize the protocol layer. */
3292 mtip_hw_exit(dd);
3293
3294 return 0;
3295}
3296
3297/*
3298 * Function called by the PCI layer when just before the
3299 * machine shuts down.
3300 *
3301 * If a protocol layer shutdown function is present it will be called
3302 * by this function.
3303 *
3304 * @dd Pointer to the driver data structure.
3305 *
3306 * return value
3307 * 0
3308 */
3309static int mtip_block_shutdown(struct driver_data *dd)
3310{
3311 dev_info(&dd->pdev->dev,
3312 "Shutting down %s ...\n", dd->disk->disk_name);
3313
3314 /* Delete our gendisk structure, and cleanup the blk queue. */
3315 del_gendisk(dd->disk);
3316 blk_cleanup_queue(dd->queue);
3317 dd->disk = NULL;
3318 dd->queue = NULL;
3319
3320 mtip_hw_shutdown(dd);
3321 return 0;
3322}
3323
3324static int mtip_block_suspend(struct driver_data *dd)
3325{
3326 dev_info(&dd->pdev->dev,
3327 "Suspending %s ...\n", dd->disk->disk_name);
3328 mtip_hw_suspend(dd);
3329 return 0;
3330}
3331
3332static int mtip_block_resume(struct driver_data *dd)
3333{
3334 dev_info(&dd->pdev->dev, "Resuming %s ...\n",
3335 dd->disk->disk_name);
3336 mtip_hw_resume(dd);
3337 return 0;
3338}
3339
3340/*
3341 * Called for each supported PCI device detected.
3342 *
3343 * This function allocates the private data structure, enables the
3344 * PCI device and then calls the block layer initialization function.
3345 *
3346 * return value
3347 * 0 on success else an error code.
3348 */
3349static int mtip_pci_probe(struct pci_dev *pdev,
3350 const struct pci_device_id *ent)
3351{
3352 int rv = 0;
3353 struct driver_data *dd = NULL;
3354
3355 /* Allocate memory for this devices private data. */
3356 dd = kzalloc(sizeof(struct driver_data), GFP_KERNEL);
3357 if (dd == NULL) {
3358 dev_err(&pdev->dev,
3359 "Unable to allocate memory for driver data\n");
3360 return -ENOMEM;
3361 }
3362
3363 /* Set the atomic variable as 1 in case of SRSI */
3364 atomic_set(&dd->drv_cleanup_done, true);
3365
3366 atomic_set(&dd->resumeflag, false);
3367
3368 /* Attach the private data to this PCI device. */
3369 pci_set_drvdata(pdev, dd);
3370
3371 rv = pcim_enable_device(pdev);
3372 if (rv < 0) {
3373 dev_err(&pdev->dev, "Unable to enable device\n");
3374 goto iomap_err;
3375 }
3376
3377 /* Map BAR5 to memory. */
3378 rv = pcim_iomap_regions(pdev, 1 << MTIP_ABAR, MTIP_DRV_NAME);
3379 if (rv < 0) {
3380 dev_err(&pdev->dev, "Unable to map regions\n");
3381 goto iomap_err;
3382 }
3383
3384 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
3385 rv = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3386
3387 if (rv) {
3388 rv = pci_set_consistent_dma_mask(pdev,
3389 DMA_BIT_MASK(32));
3390 if (rv) {
3391 dev_warn(&pdev->dev,
3392 "64-bit DMA enable failed\n");
3393 goto setmask_err;
3394 }
3395 }
3396 }
3397
3398 pci_set_master(pdev);
3399
3400 if (pci_enable_msi(pdev)) {
3401 dev_warn(&pdev->dev,
3402 "Unable to enable MSI interrupt.\n");
3403 goto block_initialize_err;
3404 }
3405
3406 /* Copy the info we may need later into the private data structure. */
3407 dd->major = mtip_major;
3408 dd->instance = instance;
3409 dd->pdev = pdev;
3410
3411 /* Initialize the block layer. */
3412 rv = mtip_block_initialize(dd);
3413 if (rv < 0) {
3414 dev_err(&pdev->dev,
3415 "Unable to initialize block layer\n");
3416 goto block_initialize_err;
3417 }
3418
3419 /*
3420 * Increment the instance count so that each device has a unique
3421 * instance number.
3422 */
3423 instance++;
3424
3425 goto done;
3426
3427block_initialize_err:
3428 pci_disable_msi(pdev);
3429
3430setmask_err:
3431 pcim_iounmap_regions(pdev, 1 << MTIP_ABAR);
3432
3433iomap_err:
3434 kfree(dd);
3435 pci_set_drvdata(pdev, NULL);
3436 return rv;
3437done:
3438 /* Set the atomic variable as 0 in case of SRSI */
3439 atomic_set(&dd->drv_cleanup_done, true);
3440
3441 return rv;
3442}
3443
3444/*
3445 * Called for each probed device when the device is removed or the
3446 * driver is unloaded.
3447 *
3448 * return value
3449 * None
3450 */
3451static void mtip_pci_remove(struct pci_dev *pdev)
3452{
3453 struct driver_data *dd = pci_get_drvdata(pdev);
3454 int counter = 0;
3455
3456 if (mtip_check_surprise_removal(pdev)) {
3457 while (atomic_read(&dd->drv_cleanup_done) == false) {
3458 counter++;
3459 msleep(20);
3460 if (counter == 10) {
3461 /* Cleanup the outstanding commands */
3462 mtip_command_cleanup(dd);
3463 break;
3464 }
3465 }
3466 }
3467 /* Set the atomic variable as 1 in case of SRSI */
3468 atomic_set(&dd->drv_cleanup_done, true);
3469
3470 /* Clean up the block layer. */
3471 mtip_block_remove(dd);
3472
3473 pci_disable_msi(pdev);
3474
3475 kfree(dd);
3476 pcim_iounmap_regions(pdev, 1 << MTIP_ABAR);
3477}
3478
3479/*
3480 * Called for each probed device when the device is suspended.
3481 *
3482 * return value
3483 * 0 Success
3484 * <0 Error
3485 */
3486static int mtip_pci_suspend(struct pci_dev *pdev, pm_message_t mesg)
3487{
3488 int rv = 0;
3489 struct driver_data *dd = pci_get_drvdata(pdev);
3490
3491 if (!dd) {
3492 dev_err(&pdev->dev,
3493 "Driver private datastructure is NULL\n");
3494 return -EFAULT;
3495 }
3496
3497 atomic_set(&dd->resumeflag, true);
3498
3499 /* Disable ports & interrupts then send standby immediate */
3500 rv = mtip_block_suspend(dd);
3501 if (rv < 0) {
3502 dev_err(&pdev->dev,
3503 "Failed to suspend controller\n");
3504 return rv;
3505 }
3506
3507 /*
3508 * Save the pci config space to pdev structure &
3509 * disable the device
3510 */
3511 pci_save_state(pdev);
3512 pci_disable_device(pdev);
3513
3514 /* Move to Low power state*/
3515 pci_set_power_state(pdev, PCI_D3hot);
3516
3517 return rv;
3518}
3519
3520/*
3521 * Called for each probed device when the device is resumed.
3522 *
3523 * return value
3524 * 0 Success
3525 * <0 Error
3526 */
3527static int mtip_pci_resume(struct pci_dev *pdev)
3528{
3529 int rv = 0;
3530 struct driver_data *dd;
3531
3532 dd = pci_get_drvdata(pdev);
3533 if (!dd) {
3534 dev_err(&pdev->dev,
3535 "Driver private datastructure is NULL\n");
3536 return -EFAULT;
3537 }
3538
3539 /* Move the device to active State */
3540 pci_set_power_state(pdev, PCI_D0);
3541
3542 /* Restore PCI configuration space */
3543 pci_restore_state(pdev);
3544
3545 /* Enable the PCI device*/
3546 rv = pcim_enable_device(pdev);
3547 if (rv < 0) {
3548 dev_err(&pdev->dev,
3549 "Failed to enable card during resume\n");
3550 goto err;
3551 }
3552 pci_set_master(pdev);
3553
3554 /*
3555 * Calls hbaReset, initPort, & startPort function
3556 * then enables interrupts
3557 */
3558 rv = mtip_block_resume(dd);
3559 if (rv < 0)
3560 dev_err(&pdev->dev, "Unable to resume\n");
3561
3562err:
3563 atomic_set(&dd->resumeflag, false);
3564
3565 return rv;
3566}
3567
3568/*
3569 * Shutdown routine
3570 *
3571 * return value
3572 * None
3573 */
3574static void mtip_pci_shutdown(struct pci_dev *pdev)
3575{
3576 struct driver_data *dd = pci_get_drvdata(pdev);
3577 if (dd)
3578 mtip_block_shutdown(dd);
3579}
3580
3581/* Table of device ids supported by this driver. */
3582static DEFINE_PCI_DEVICE_TABLE(mtip_pci_tbl) = {
3583 { PCI_DEVICE(PCI_VENDOR_ID_MICRON, P320_DEVICE_ID) },
3584 { 0 }
3585};
3586
3587/* Structure that describes the PCI driver functions. */
3588static struct pci_driver mtip_pci_driver = {
3589 .name = MTIP_DRV_NAME,
3590 .id_table = mtip_pci_tbl,
3591 .probe = mtip_pci_probe,
3592 .remove = mtip_pci_remove,
3593 .suspend = mtip_pci_suspend,
3594 .resume = mtip_pci_resume,
3595 .shutdown = mtip_pci_shutdown,
3596};
3597
3598MODULE_DEVICE_TABLE(pci, mtip_pci_tbl);
3599
3600/*
3601 * Module initialization function.
3602 *
3603 * Called once when the module is loaded. This function allocates a major
3604 * block device number to the Cyclone devices and registers the PCI layer
3605 * of the driver.
3606 *
3607 * Return value
3608 * 0 on success else error code.
3609 */
3610static int __init mtip_init(void)
3611{
3612 printk(KERN_INFO MTIP_DRV_NAME " Version " MTIP_DRV_VERSION "\n");
3613
3614 /* Allocate a major block device number to use with this driver. */
3615 mtip_major = register_blkdev(0, MTIP_DRV_NAME);
3616 if (mtip_major < 0) {
3617 printk(KERN_ERR "Unable to register block device (%d)\n",
3618 mtip_major);
3619 return -EBUSY;
3620 }
3621
3622 /* Register our PCI operations. */
3623 return pci_register_driver(&mtip_pci_driver);
3624}
3625
3626/*
3627 * Module de-initialization function.
3628 *
3629 * Called once when the module is unloaded. This function deallocates
3630 * the major block device number allocated by mtip_init() and
3631 * unregisters the PCI layer of the driver.
3632 *
3633 * Return value
3634 * none
3635 */
3636static void __exit mtip_exit(void)
3637{
3638 /* Release the allocated major block device number. */
3639 unregister_blkdev(mtip_major, MTIP_DRV_NAME);
3640
3641 /* Unregister the PCI driver. */
3642 pci_unregister_driver(&mtip_pci_driver);
3643}
3644
3645MODULE_AUTHOR("Micron Technology, Inc");
3646MODULE_DESCRIPTION("Micron RealSSD PCIe Block Driver");
3647MODULE_LICENSE("GPL");
3648MODULE_VERSION(MTIP_DRV_VERSION);
3649
3650module_init(mtip_init);
3651module_exit(mtip_exit);
diff --git a/drivers/block/mtip32xx/mtip32xx.h b/drivers/block/mtip32xx/mtip32xx.h
new file mode 100644
index 000000000000..723d7c4946dc
--- /dev/null
+++ b/drivers/block/mtip32xx/mtip32xx.h
@@ -0,0 +1,423 @@
1/*
2 * mtip32xx.h - Header file for the P320 SSD Block Driver
3 * Copyright (C) 2011 Micron Technology, Inc.
4 *
5 * Portions of this code were derived from works subjected to the
6 * following copyright:
7 * Copyright (C) 2009 Integrated Device Technology, Inc.
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
13 *
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
18 *
19 */
20
21#ifndef __MTIP32XX_H__
22#define __MTIP32XX_H__
23
24#include <linux/spinlock.h>
25#include <linux/rwsem.h>
26#include <linux/ata.h>
27#include <linux/interrupt.h>
28#include <linux/genhd.h>
29#include <linux/version.h>
30
31/* Offset of Subsystem Device ID in pci confoguration space */
32#define PCI_SUBSYSTEM_DEVICEID 0x2E
33
34/* offset of Device Control register in PCIe extended capabilites space */
35#define PCIE_CONFIG_EXT_DEVICE_CONTROL_OFFSET 0x48
36
37/* # of times to retry timed out IOs */
38#define MTIP_MAX_RETRIES 5
39
40/* Various timeout values in ms */
41#define MTIP_NCQ_COMMAND_TIMEOUT_MS 5000
42#define MTIP_IOCTL_COMMAND_TIMEOUT_MS 5000
43#define MTIP_INTERNAL_COMMAND_TIMEOUT_MS 5000
44
45/* check for timeouts every 500ms */
46#define MTIP_TIMEOUT_CHECK_PERIOD 500
47
48/* ftl rebuild */
49#define MTIP_FTL_REBUILD_OFFSET 142
50#define MTIP_FTL_REBUILD_MAGIC 0xED51
51#define MTIP_FTL_REBUILD_TIMEOUT_MS 2400000
52
53/* Macro to extract the tag bit number from a tag value. */
54#define MTIP_TAG_BIT(tag) (tag & 0x1F)
55
56/*
57 * Macro to extract the tag index from a tag value. The index
58 * is used to access the correct s_active/Command Issue register based
59 * on the tag value.
60 */
61#define MTIP_TAG_INDEX(tag) (tag >> 5)
62
63/*
64 * Maximum number of scatter gather entries
65 * a single command may have.
66 */
67#define MTIP_MAX_SG 128
68
69/*
70 * Maximum number of slot groups (Command Issue & s_active registers)
71 * NOTE: This is the driver maximum; check dd->slot_groups for actual value.
72 */
73#define MTIP_MAX_SLOT_GROUPS 8
74
75/* Internal command tag. */
76#define MTIP_TAG_INTERNAL 0
77
78/* Micron Vendor ID & P320x SSD Device ID */
79#define PCI_VENDOR_ID_MICRON 0x1344
80#define P320_DEVICE_ID 0x5150
81
82/* Driver name and version strings */
83#define MTIP_DRV_NAME "mtip32xx"
84#define MTIP_DRV_VERSION "1.2.6os3"
85
86/* Maximum number of minor device numbers per device. */
87#define MTIP_MAX_MINORS 16
88
89/* Maximum number of supported command slots. */
90#define MTIP_MAX_COMMAND_SLOTS (MTIP_MAX_SLOT_GROUPS * 32)
91
92/*
93 * Per-tag bitfield size in longs.
94 * Linux bit manipulation functions
95 * (i.e. test_and_set_bit, find_next_zero_bit)
96 * manipulate memory in longs, so we try to make the math work.
97 * take the slot groups and find the number of longs, rounding up.
98 * Careful! i386 and x86_64 use different size longs!
99 */
100#define U32_PER_LONG (sizeof(long) / sizeof(u32))
101#define SLOTBITS_IN_LONGS ((MTIP_MAX_SLOT_GROUPS + \
102 (U32_PER_LONG-1))/U32_PER_LONG)
103
104/* BAR number used to access the HBA registers. */
105#define MTIP_ABAR 5
106
107/* Forced Unit Access Bit */
108#define FUA_BIT 0x80
109
110#ifdef DEBUG
111 #define dbg_printk(format, arg...) \
112 printk(pr_fmt(format), ##arg);
113#else
114 #define dbg_printk(format, arg...)
115#endif
116
117#define __force_bit2int (unsigned int __force)
118
119/* below are bit numbers in 'flags' defined in mtip_port */
120#define MTIP_FLAG_IC_ACTIVE_BIT 0
121#define MTIP_FLAG_EH_ACTIVE_BIT 1
122#define MTIP_FLAG_SVC_THD_ACTIVE_BIT 2
123#define MTIP_FLAG_ISSUE_CMDS_BIT 4
124#define MTIP_FLAG_REBUILD_BIT 5
125#define MTIP_FLAG_SVC_THD_SHOULD_STOP_BIT 8
126
127/* Register Frame Information Structure (FIS), host to device. */
128struct host_to_dev_fis {
129 /*
130 * FIS type.
131 * - 27h Register FIS, host to device.
132 * - 34h Register FIS, device to host.
133 * - 39h DMA Activate FIS, device to host.
134 * - 41h DMA Setup FIS, bi-directional.
135 * - 46h Data FIS, bi-directional.
136 * - 58h BIST Activate FIS, bi-directional.
137 * - 5Fh PIO Setup FIS, device to host.
138 * - A1h Set Device Bits FIS, device to host.
139 */
140 unsigned char type;
141 unsigned char opts;
142 unsigned char command;
143 unsigned char features;
144
145 union {
146 unsigned char lba_low;
147 unsigned char sector;
148 };
149 union {
150 unsigned char lba_mid;
151 unsigned char cyl_low;
152 };
153 union {
154 unsigned char lba_hi;
155 unsigned char cyl_hi;
156 };
157 union {
158 unsigned char device;
159 unsigned char head;
160 };
161
162 union {
163 unsigned char lba_low_ex;
164 unsigned char sector_ex;
165 };
166 union {
167 unsigned char lba_mid_ex;
168 unsigned char cyl_low_ex;
169 };
170 union {
171 unsigned char lba_hi_ex;
172 unsigned char cyl_hi_ex;
173 };
174 unsigned char features_ex;
175
176 unsigned char sect_count;
177 unsigned char sect_cnt_ex;
178 unsigned char res2;
179 unsigned char control;
180
181 unsigned int res3;
182};
183
184/* Command header structure. */
185struct mtip_cmd_hdr {
186 /*
187 * Command options.
188 * - Bits 31:16 Number of PRD entries.
189 * - Bits 15:8 Unused in this implementation.
190 * - Bit 7 Prefetch bit, informs the drive to prefetch PRD entries.
191 * - Bit 6 Write bit, should be set when writing data to the device.
192 * - Bit 5 Unused in this implementation.
193 * - Bits 4:0 Length of the command FIS in DWords (DWord = 4 bytes).
194 */
195 unsigned int opts;
196 /* This field is unsed when using NCQ. */
197 union {
198 unsigned int byte_count;
199 unsigned int status;
200 };
201 /*
202 * Lower 32 bits of the command table address associated with this
203 * header. The command table addresses must be 128 byte aligned.
204 */
205 unsigned int ctba;
206 /*
207 * If 64 bit addressing is used this field is the upper 32 bits
208 * of the command table address associated with this command.
209 */
210 unsigned int ctbau;
211 /* Reserved and unused. */
212 unsigned int res[4];
213};
214
215/* Command scatter gather structure (PRD). */
216struct mtip_cmd_sg {
217 /*
218 * Low 32 bits of the data buffer address. For P320 this
219 * address must be 8 byte aligned signified by bits 2:0 being
220 * set to 0.
221 */
222 unsigned int dba;
223 /*
224 * When 64 bit addressing is used this field is the upper
225 * 32 bits of the data buffer address.
226 */
227 unsigned int dba_upper;
228 /* Unused. */
229 unsigned int reserved;
230 /*
231 * Bit 31: interrupt when this data block has been transferred.
232 * Bits 30..22: reserved
233 * Bits 21..0: byte count (minus 1). For P320 the byte count must be
234 * 8 byte aligned signified by bits 2:0 being set to 1.
235 */
236 unsigned int info;
237};
238struct mtip_port;
239
240/* Structure used to describe a command. */
241struct mtip_cmd {
242
243 struct mtip_cmd_hdr *command_header; /* ptr to command header entry */
244
245 dma_addr_t command_header_dma; /* corresponding physical address */
246
247 void *command; /* ptr to command table entry */
248
249 dma_addr_t command_dma; /* corresponding physical address */
250
251 void *comp_data; /* data passed to completion function comp_func() */
252 /*
253 * Completion function called by the ISR upon completion of
254 * a command.
255 */
256 void (*comp_func)(struct mtip_port *port,
257 int tag,
258 void *data,
259 int status);
260 /* Additional callback function that may be called by comp_func() */
261 void (*async_callback)(void *data, int status);
262
263 void *async_data; /* Addl. data passed to async_callback() */
264
265 int scatter_ents; /* Number of scatter list entries used */
266
267 struct scatterlist sg[MTIP_MAX_SG]; /* Scatter list entries */
268
269 int retries; /* The number of retries left for this command. */
270
271 int direction; /* Data transfer direction */
272
273 unsigned long comp_time; /* command completion time, in jiffies */
274
275 atomic_t active; /* declares if this command sent to the drive. */
276};
277
278/* Structure used to describe a port. */
279struct mtip_port {
280 /* Pointer back to the driver data for this port. */
281 struct driver_data *dd;
282 /*
283 * Used to determine if the data pointed to by the
284 * identify field is valid.
285 */
286 unsigned long identify_valid;
287 /* Base address of the memory mapped IO for the port. */
288 void __iomem *mmio;
289 /* Array of pointers to the memory mapped s_active registers. */
290 void __iomem *s_active[MTIP_MAX_SLOT_GROUPS];
291 /* Array of pointers to the memory mapped completed registers. */
292 void __iomem *completed[MTIP_MAX_SLOT_GROUPS];
293 /* Array of pointers to the memory mapped Command Issue registers. */
294 void __iomem *cmd_issue[MTIP_MAX_SLOT_GROUPS];
295 /*
296 * Pointer to the beginning of the command header memory as used
297 * by the driver.
298 */
299 void *command_list;
300 /*
301 * Pointer to the beginning of the command header memory as used
302 * by the DMA.
303 */
304 dma_addr_t command_list_dma;
305 /*
306 * Pointer to the beginning of the RX FIS memory as used
307 * by the driver.
308 */
309 void *rxfis;
310 /*
311 * Pointer to the beginning of the RX FIS memory as used
312 * by the DMA.
313 */
314 dma_addr_t rxfis_dma;
315 /*
316 * Pointer to the beginning of the command table memory as used
317 * by the driver.
318 */
319 void *command_table;
320 /*
321 * Pointer to the beginning of the command table memory as used
322 * by the DMA.
323 */
324 dma_addr_t command_tbl_dma;
325 /*
326 * Pointer to the beginning of the identify data memory as used
327 * by the driver.
328 */
329 u16 *identify;
330 /*
331 * Pointer to the beginning of the identify data memory as used
332 * by the DMA.
333 */
334 dma_addr_t identify_dma;
335 /*
336 * Pointer to the beginning of a sector buffer that is used
337 * by the driver when issuing internal commands.
338 */
339 u16 *sector_buffer;
340 /*
341 * Pointer to the beginning of a sector buffer that is used
342 * by the DMA when the driver issues internal commands.
343 */
344 dma_addr_t sector_buffer_dma;
345 /*
346 * Bit significant, used to determine if a command slot has
347 * been allocated. i.e. the slot is in use. Bits are cleared
348 * when the command slot and all associated data structures
349 * are no longer needed.
350 */
351 unsigned long allocated[SLOTBITS_IN_LONGS];
352 /*
353 * used to queue commands when an internal command is in progress
354 * or error handling is active
355 */
356 unsigned long cmds_to_issue[SLOTBITS_IN_LONGS];
357 /*
358 * Array of command slots. Structure includes pointers to the
359 * command header and command table, and completion function and data
360 * pointers.
361 */
362 struct mtip_cmd commands[MTIP_MAX_COMMAND_SLOTS];
363 /* Used by mtip_service_thread to wait for an event */
364 wait_queue_head_t svc_wait;
365 /*
366 * indicates the state of the port. Also, helps the service thread
367 * to determine its action on wake up.
368 */
369 unsigned long flags;
370 /*
371 * Timer used to complete commands that have been active for too long.
372 */
373 struct timer_list cmd_timer;
374 /*
375 * Semaphore used to block threads if there are no
376 * command slots available.
377 */
378 struct semaphore cmd_slot;
379 /* Spinlock for working around command-issue bug. */
380 spinlock_t cmd_issue_lock;
381};
382
383/*
384 * Driver private data structure.
385 *
386 * One structure is allocated per probed device.
387 */
388struct driver_data {
389 void __iomem *mmio; /* Base address of the HBA registers. */
390
391 int major; /* Major device number. */
392
393 int instance; /* Instance number. First device probed is 0, ... */
394
395 struct gendisk *disk; /* Pointer to our gendisk structure. */
396
397 struct pci_dev *pdev; /* Pointer to the PCI device structure. */
398
399 struct request_queue *queue; /* Our request queue. */
400
401 struct mtip_port *port; /* Pointer to the port data structure. */
402
403 /* Tasklet used to process the bottom half of the ISR. */
404 struct tasklet_struct tasklet;
405
406 unsigned product_type; /* magic value declaring the product type */
407
408 unsigned slot_groups; /* number of slot groups the product supports */
409
410 atomic_t drv_cleanup_done; /* Atomic variable for SRSI */
411
412 unsigned long index; /* Index to determine the disk name */
413
414 unsigned int ftlrebuildflag; /* FTL rebuild flag */
415
416 atomic_t resumeflag; /* Atomic variable to track suspend/resume */
417
418 atomic_t eh_active; /* Flag for error handling tracking */
419
420 struct task_struct *mtip_svc_handler; /* task_struct of svc thd */
421};
422
423#endif
diff --git a/drivers/block/xen-blkback/blkback.c b/drivers/block/xen-blkback/blkback.c
index 15ec4db194d1..0088bf60f368 100644
--- a/drivers/block/xen-blkback/blkback.c
+++ b/drivers/block/xen-blkback/blkback.c
@@ -39,9 +39,6 @@
39#include <linux/list.h> 39#include <linux/list.h>
40#include <linux/delay.h> 40#include <linux/delay.h>
41#include <linux/freezer.h> 41#include <linux/freezer.h>
42#include <linux/loop.h>
43#include <linux/falloc.h>
44#include <linux/fs.h>
45 42
46#include <xen/events.h> 43#include <xen/events.h>
47#include <xen/page.h> 44#include <xen/page.h>
@@ -362,7 +359,7 @@ static int xen_blkbk_map(struct blkif_request *req,
362{ 359{
363 struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST]; 360 struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST];
364 int i; 361 int i;
365 int nseg = req->nr_segments; 362 int nseg = req->u.rw.nr_segments;
366 int ret = 0; 363 int ret = 0;
367 364
368 /* 365 /*
@@ -416,30 +413,25 @@ static int xen_blkbk_map(struct blkif_request *req,
416 return ret; 413 return ret;
417} 414}
418 415
419static void xen_blk_discard(struct xen_blkif *blkif, struct blkif_request *req) 416static int dispatch_discard_io(struct xen_blkif *blkif,
417 struct blkif_request *req)
420{ 418{
421 int err = 0; 419 int err = 0;
422 int status = BLKIF_RSP_OKAY; 420 int status = BLKIF_RSP_OKAY;
423 struct block_device *bdev = blkif->vbd.bdev; 421 struct block_device *bdev = blkif->vbd.bdev;
424 422
425 if (blkif->blk_backend_type == BLKIF_BACKEND_PHY) 423 blkif->st_ds_req++;
426 /* just forward the discard request */ 424
425 xen_blkif_get(blkif);
426 if (blkif->blk_backend_type == BLKIF_BACKEND_PHY ||
427 blkif->blk_backend_type == BLKIF_BACKEND_FILE) {
428 unsigned long secure = (blkif->vbd.discard_secure &&
429 (req->u.discard.flag & BLKIF_DISCARD_SECURE)) ?
430 BLKDEV_DISCARD_SECURE : 0;
427 err = blkdev_issue_discard(bdev, 431 err = blkdev_issue_discard(bdev,
428 req->u.discard.sector_number, 432 req->u.discard.sector_number,
429 req->u.discard.nr_sectors, 433 req->u.discard.nr_sectors,
430 GFP_KERNEL, 0); 434 GFP_KERNEL, secure);
431 else if (blkif->blk_backend_type == BLKIF_BACKEND_FILE) {
432 /* punch a hole in the backing file */
433 struct loop_device *lo = bdev->bd_disk->private_data;
434 struct file *file = lo->lo_backing_file;
435
436 if (file->f_op->fallocate)
437 err = file->f_op->fallocate(file,
438 FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
439 req->u.discard.sector_number << 9,
440 req->u.discard.nr_sectors << 9);
441 else
442 err = -EOPNOTSUPP;
443 } else 435 } else
444 err = -EOPNOTSUPP; 436 err = -EOPNOTSUPP;
445 437
@@ -449,7 +441,9 @@ static void xen_blk_discard(struct xen_blkif *blkif, struct blkif_request *req)
449 } else if (err) 441 } else if (err)
450 status = BLKIF_RSP_ERROR; 442 status = BLKIF_RSP_ERROR;
451 443
452 make_response(blkif, req->id, req->operation, status); 444 make_response(blkif, req->u.discard.id, req->operation, status);
445 xen_blkif_put(blkif);
446 return err;
453} 447}
454 448
455static void xen_blk_drain_io(struct xen_blkif *blkif) 449static void xen_blk_drain_io(struct xen_blkif *blkif)
@@ -573,8 +567,11 @@ __do_block_io_op(struct xen_blkif *blkif)
573 567
574 /* Apply all sanity checks to /private copy/ of request. */ 568 /* Apply all sanity checks to /private copy/ of request. */
575 barrier(); 569 barrier();
576 570 if (unlikely(req.operation == BLKIF_OP_DISCARD)) {
577 if (dispatch_rw_block_io(blkif, &req, pending_req)) 571 free_req(pending_req);
572 if (dispatch_discard_io(blkif, &req))
573 break;
574 } else if (dispatch_rw_block_io(blkif, &req, pending_req))
578 break; 575 break;
579 576
580 /* Yield point for this unbounded loop. */ 577 /* Yield point for this unbounded loop. */
@@ -633,10 +630,6 @@ static int dispatch_rw_block_io(struct xen_blkif *blkif,
633 blkif->st_f_req++; 630 blkif->st_f_req++;
634 operation = WRITE_FLUSH; 631 operation = WRITE_FLUSH;
635 break; 632 break;
636 case BLKIF_OP_DISCARD:
637 blkif->st_ds_req++;
638 operation = REQ_DISCARD;
639 break;
640 default: 633 default:
641 operation = 0; /* make gcc happy */ 634 operation = 0; /* make gcc happy */
642 goto fail_response; 635 goto fail_response;
@@ -644,9 +637,9 @@ static int dispatch_rw_block_io(struct xen_blkif *blkif,
644 } 637 }
645 638
646 /* Check that the number of segments is sane. */ 639 /* Check that the number of segments is sane. */
647 nseg = req->nr_segments; 640 nseg = req->u.rw.nr_segments;
648 if (unlikely(nseg == 0 && operation != WRITE_FLUSH && 641
649 operation != REQ_DISCARD) || 642 if (unlikely(nseg == 0 && operation != WRITE_FLUSH) ||
650 unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) { 643 unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
651 pr_debug(DRV_PFX "Bad number of segments in request (%d)\n", 644 pr_debug(DRV_PFX "Bad number of segments in request (%d)\n",
652 nseg); 645 nseg);
@@ -654,12 +647,12 @@ static int dispatch_rw_block_io(struct xen_blkif *blkif,
654 goto fail_response; 647 goto fail_response;
655 } 648 }
656 649
657 preq.dev = req->handle; 650 preq.dev = req->u.rw.handle;
658 preq.sector_number = req->u.rw.sector_number; 651 preq.sector_number = req->u.rw.sector_number;
659 preq.nr_sects = 0; 652 preq.nr_sects = 0;
660 653
661 pending_req->blkif = blkif; 654 pending_req->blkif = blkif;
662 pending_req->id = req->id; 655 pending_req->id = req->u.rw.id;
663 pending_req->operation = req->operation; 656 pending_req->operation = req->operation;
664 pending_req->status = BLKIF_RSP_OKAY; 657 pending_req->status = BLKIF_RSP_OKAY;
665 pending_req->nr_pages = nseg; 658 pending_req->nr_pages = nseg;
@@ -707,7 +700,7 @@ static int dispatch_rw_block_io(struct xen_blkif *blkif,
707 * the hypercall to unmap the grants - that is all done in 700 * the hypercall to unmap the grants - that is all done in
708 * xen_blkbk_unmap. 701 * xen_blkbk_unmap.
709 */ 702 */
710 if (operation != REQ_DISCARD && xen_blkbk_map(req, pending_req, seg)) 703 if (xen_blkbk_map(req, pending_req, seg))
711 goto fail_flush; 704 goto fail_flush;
712 705
713 /* 706 /*
@@ -739,23 +732,16 @@ static int dispatch_rw_block_io(struct xen_blkif *blkif,
739 732
740 /* This will be hit if the operation was a flush or discard. */ 733 /* This will be hit if the operation was a flush or discard. */
741 if (!bio) { 734 if (!bio) {
742 BUG_ON(operation != WRITE_FLUSH && operation != REQ_DISCARD); 735 BUG_ON(operation != WRITE_FLUSH);
743 736
744 if (operation == WRITE_FLUSH) { 737 bio = bio_alloc(GFP_KERNEL, 0);
745 bio = bio_alloc(GFP_KERNEL, 0); 738 if (unlikely(bio == NULL))
746 if (unlikely(bio == NULL)) 739 goto fail_put_bio;
747 goto fail_put_bio;
748 740
749 biolist[nbio++] = bio; 741 biolist[nbio++] = bio;
750 bio->bi_bdev = preq.bdev; 742 bio->bi_bdev = preq.bdev;
751 bio->bi_private = pending_req; 743 bio->bi_private = pending_req;
752 bio->bi_end_io = end_block_io_op; 744 bio->bi_end_io = end_block_io_op;
753 } else if (operation == REQ_DISCARD) {
754 xen_blk_discard(blkif, req);
755 xen_blkif_put(blkif);
756 free_req(pending_req);
757 return 0;
758 }
759 } 745 }
760 746
761 /* 747 /*
@@ -784,7 +770,7 @@ static int dispatch_rw_block_io(struct xen_blkif *blkif,
784 xen_blkbk_unmap(pending_req); 770 xen_blkbk_unmap(pending_req);
785 fail_response: 771 fail_response:
786 /* Haven't submitted any bio's yet. */ 772 /* Haven't submitted any bio's yet. */
787 make_response(blkif, req->id, req->operation, BLKIF_RSP_ERROR); 773 make_response(blkif, req->u.rw.id, req->operation, BLKIF_RSP_ERROR);
788 free_req(pending_req); 774 free_req(pending_req);
789 msleep(1); /* back off a bit */ 775 msleep(1); /* back off a bit */
790 return -EIO; 776 return -EIO;
diff --git a/drivers/block/xen-blkback/common.h b/drivers/block/xen-blkback/common.h
index dfb1b3a43a5d..d0ee7edc9be8 100644
--- a/drivers/block/xen-blkback/common.h
+++ b/drivers/block/xen-blkback/common.h
@@ -60,58 +60,66 @@ struct blkif_common_response {
60 char dummy; 60 char dummy;
61}; 61};
62 62
63/* i386 protocol version */
64#pragma pack(push, 4)
65
66struct blkif_x86_32_request_rw { 63struct blkif_x86_32_request_rw {
64 uint8_t nr_segments; /* number of segments */
65 blkif_vdev_t handle; /* only for read/write requests */
66 uint64_t id; /* private guest value, echoed in resp */
67 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 67 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
68 struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; 68 struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
69}; 69} __attribute__((__packed__));
70 70
71struct blkif_x86_32_request_discard { 71struct blkif_x86_32_request_discard {
72 uint8_t flag; /* BLKIF_DISCARD_SECURE or zero */
73 blkif_vdev_t _pad1; /* was "handle" for read/write requests */
74 uint64_t id; /* private guest value, echoed in resp */
72 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 75 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
73 uint64_t nr_sectors; 76 uint64_t nr_sectors;
74}; 77} __attribute__((__packed__));
75 78
76struct blkif_x86_32_request { 79struct blkif_x86_32_request {
77 uint8_t operation; /* BLKIF_OP_??? */ 80 uint8_t operation; /* BLKIF_OP_??? */
78 uint8_t nr_segments; /* number of segments */
79 blkif_vdev_t handle; /* only for read/write requests */
80 uint64_t id; /* private guest value, echoed in resp */
81 union { 81 union {
82 struct blkif_x86_32_request_rw rw; 82 struct blkif_x86_32_request_rw rw;
83 struct blkif_x86_32_request_discard discard; 83 struct blkif_x86_32_request_discard discard;
84 } u; 84 } u;
85}; 85} __attribute__((__packed__));
86
87/* i386 protocol version */
88#pragma pack(push, 4)
86struct blkif_x86_32_response { 89struct blkif_x86_32_response {
87 uint64_t id; /* copied from request */ 90 uint64_t id; /* copied from request */
88 uint8_t operation; /* copied from request */ 91 uint8_t operation; /* copied from request */
89 int16_t status; /* BLKIF_RSP_??? */ 92 int16_t status; /* BLKIF_RSP_??? */
90}; 93};
91#pragma pack(pop) 94#pragma pack(pop)
92
93/* x86_64 protocol version */ 95/* x86_64 protocol version */
94 96
95struct blkif_x86_64_request_rw { 97struct blkif_x86_64_request_rw {
98 uint8_t nr_segments; /* number of segments */
99 blkif_vdev_t handle; /* only for read/write requests */
100 uint32_t _pad1; /* offsetof(blkif_reqest..,u.rw.id)==8 */
101 uint64_t id;
96 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 102 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
97 struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; 103 struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
98}; 104} __attribute__((__packed__));
99 105
100struct blkif_x86_64_request_discard { 106struct blkif_x86_64_request_discard {
107 uint8_t flag; /* BLKIF_DISCARD_SECURE or zero */
108 blkif_vdev_t _pad1; /* was "handle" for read/write requests */
109 uint32_t _pad2; /* offsetof(blkif_..,u.discard.id)==8 */
110 uint64_t id;
101 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 111 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
102 uint64_t nr_sectors; 112 uint64_t nr_sectors;
103}; 113} __attribute__((__packed__));
104 114
105struct blkif_x86_64_request { 115struct blkif_x86_64_request {
106 uint8_t operation; /* BLKIF_OP_??? */ 116 uint8_t operation; /* BLKIF_OP_??? */
107 uint8_t nr_segments; /* number of segments */
108 blkif_vdev_t handle; /* only for read/write requests */
109 uint64_t __attribute__((__aligned__(8))) id;
110 union { 117 union {
111 struct blkif_x86_64_request_rw rw; 118 struct blkif_x86_64_request_rw rw;
112 struct blkif_x86_64_request_discard discard; 119 struct blkif_x86_64_request_discard discard;
113 } u; 120 } u;
114}; 121} __attribute__((__packed__));
122
115struct blkif_x86_64_response { 123struct blkif_x86_64_response {
116 uint64_t __attribute__((__aligned__(8))) id; 124 uint64_t __attribute__((__aligned__(8))) id;
117 uint8_t operation; /* copied from request */ 125 uint8_t operation; /* copied from request */
@@ -156,6 +164,7 @@ struct xen_vbd {
156 /* Cached size parameter. */ 164 /* Cached size parameter. */
157 sector_t size; 165 sector_t size;
158 bool flush_support; 166 bool flush_support;
167 bool discard_secure;
159}; 168};
160 169
161struct backend_info; 170struct backend_info;
@@ -237,22 +246,23 @@ static inline void blkif_get_x86_32_req(struct blkif_request *dst,
237{ 246{
238 int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST; 247 int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST;
239 dst->operation = src->operation; 248 dst->operation = src->operation;
240 dst->nr_segments = src->nr_segments;
241 dst->handle = src->handle;
242 dst->id = src->id;
243 switch (src->operation) { 249 switch (src->operation) {
244 case BLKIF_OP_READ: 250 case BLKIF_OP_READ:
245 case BLKIF_OP_WRITE: 251 case BLKIF_OP_WRITE:
246 case BLKIF_OP_WRITE_BARRIER: 252 case BLKIF_OP_WRITE_BARRIER:
247 case BLKIF_OP_FLUSH_DISKCACHE: 253 case BLKIF_OP_FLUSH_DISKCACHE:
254 dst->u.rw.nr_segments = src->u.rw.nr_segments;
255 dst->u.rw.handle = src->u.rw.handle;
256 dst->u.rw.id = src->u.rw.id;
248 dst->u.rw.sector_number = src->u.rw.sector_number; 257 dst->u.rw.sector_number = src->u.rw.sector_number;
249 barrier(); 258 barrier();
250 if (n > dst->nr_segments) 259 if (n > dst->u.rw.nr_segments)
251 n = dst->nr_segments; 260 n = dst->u.rw.nr_segments;
252 for (i = 0; i < n; i++) 261 for (i = 0; i < n; i++)
253 dst->u.rw.seg[i] = src->u.rw.seg[i]; 262 dst->u.rw.seg[i] = src->u.rw.seg[i];
254 break; 263 break;
255 case BLKIF_OP_DISCARD: 264 case BLKIF_OP_DISCARD:
265 dst->u.discard.flag = src->u.discard.flag;
256 dst->u.discard.sector_number = src->u.discard.sector_number; 266 dst->u.discard.sector_number = src->u.discard.sector_number;
257 dst->u.discard.nr_sectors = src->u.discard.nr_sectors; 267 dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
258 break; 268 break;
@@ -266,22 +276,23 @@ static inline void blkif_get_x86_64_req(struct blkif_request *dst,
266{ 276{
267 int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST; 277 int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST;
268 dst->operation = src->operation; 278 dst->operation = src->operation;
269 dst->nr_segments = src->nr_segments;
270 dst->handle = src->handle;
271 dst->id = src->id;
272 switch (src->operation) { 279 switch (src->operation) {
273 case BLKIF_OP_READ: 280 case BLKIF_OP_READ:
274 case BLKIF_OP_WRITE: 281 case BLKIF_OP_WRITE:
275 case BLKIF_OP_WRITE_BARRIER: 282 case BLKIF_OP_WRITE_BARRIER:
276 case BLKIF_OP_FLUSH_DISKCACHE: 283 case BLKIF_OP_FLUSH_DISKCACHE:
284 dst->u.rw.nr_segments = src->u.rw.nr_segments;
285 dst->u.rw.handle = src->u.rw.handle;
286 dst->u.rw.id = src->u.rw.id;
277 dst->u.rw.sector_number = src->u.rw.sector_number; 287 dst->u.rw.sector_number = src->u.rw.sector_number;
278 barrier(); 288 barrier();
279 if (n > dst->nr_segments) 289 if (n > dst->u.rw.nr_segments)
280 n = dst->nr_segments; 290 n = dst->u.rw.nr_segments;
281 for (i = 0; i < n; i++) 291 for (i = 0; i < n; i++)
282 dst->u.rw.seg[i] = src->u.rw.seg[i]; 292 dst->u.rw.seg[i] = src->u.rw.seg[i];
283 break; 293 break;
284 case BLKIF_OP_DISCARD: 294 case BLKIF_OP_DISCARD:
295 dst->u.discard.flag = src->u.discard.flag;
285 dst->u.discard.sector_number = src->u.discard.sector_number; 296 dst->u.discard.sector_number = src->u.discard.sector_number;
286 dst->u.discard.nr_sectors = src->u.discard.nr_sectors; 297 dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
287 break; 298 break;
diff --git a/drivers/block/xen-blkback/xenbus.c b/drivers/block/xen-blkback/xenbus.c
index 37c794d31264..24a2fb57e5d0 100644
--- a/drivers/block/xen-blkback/xenbus.c
+++ b/drivers/block/xen-blkback/xenbus.c
@@ -338,6 +338,9 @@ static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
338 if (q && q->flush_flags) 338 if (q && q->flush_flags)
339 vbd->flush_support = true; 339 vbd->flush_support = true;
340 340
341 if (q && blk_queue_secdiscard(q))
342 vbd->discard_secure = true;
343
341 DPRINTK("Successful creation of handle=%04x (dom=%u)\n", 344 DPRINTK("Successful creation of handle=%04x (dom=%u)\n",
342 handle, blkif->domid); 345 handle, blkif->domid);
343 return 0; 346 return 0;
@@ -420,6 +423,15 @@ int xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
420 state = 1; 423 state = 1;
421 blkif->blk_backend_type = BLKIF_BACKEND_PHY; 424 blkif->blk_backend_type = BLKIF_BACKEND_PHY;
422 } 425 }
426 /* Optional. */
427 err = xenbus_printf(xbt, dev->nodename,
428 "discard-secure", "%d",
429 blkif->vbd.discard_secure);
430 if (err) {
431 xenbus_dev_fatal(dev, err,
432 "writting discard-secure");
433 goto kfree;
434 }
423 } 435 }
424 } else { 436 } else {
425 err = PTR_ERR(type); 437 err = PTR_ERR(type);
diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index 9fd3ee203b1e..2f22874c0a37 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -98,7 +98,8 @@ struct blkfront_info
98 unsigned long shadow_free; 98 unsigned long shadow_free;
99 unsigned int feature_flush; 99 unsigned int feature_flush;
100 unsigned int flush_op; 100 unsigned int flush_op;
101 unsigned int feature_discard; 101 unsigned int feature_discard:1;
102 unsigned int feature_secdiscard:1;
102 unsigned int discard_granularity; 103 unsigned int discard_granularity;
103 unsigned int discard_alignment; 104 unsigned int discard_alignment;
104 int is_ready; 105 int is_ready;
@@ -135,15 +136,15 @@ static int get_id_from_freelist(struct blkfront_info *info)
135{ 136{
136 unsigned long free = info->shadow_free; 137 unsigned long free = info->shadow_free;
137 BUG_ON(free >= BLK_RING_SIZE); 138 BUG_ON(free >= BLK_RING_SIZE);
138 info->shadow_free = info->shadow[free].req.id; 139 info->shadow_free = info->shadow[free].req.u.rw.id;
139 info->shadow[free].req.id = 0x0fffffee; /* debug */ 140 info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
140 return free; 141 return free;
141} 142}
142 143
143static void add_id_to_freelist(struct blkfront_info *info, 144static void add_id_to_freelist(struct blkfront_info *info,
144 unsigned long id) 145 unsigned long id)
145{ 146{
146 info->shadow[id].req.id = info->shadow_free; 147 info->shadow[id].req.u.rw.id = info->shadow_free;
147 info->shadow[id].request = NULL; 148 info->shadow[id].request = NULL;
148 info->shadow_free = id; 149 info->shadow_free = id;
149} 150}
@@ -156,7 +157,7 @@ static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
156 if (end > nr_minors) { 157 if (end > nr_minors) {
157 unsigned long *bitmap, *old; 158 unsigned long *bitmap, *old;
158 159
159 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap), 160 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
160 GFP_KERNEL); 161 GFP_KERNEL);
161 if (bitmap == NULL) 162 if (bitmap == NULL)
162 return -ENOMEM; 163 return -ENOMEM;
@@ -287,9 +288,9 @@ static int blkif_queue_request(struct request *req)
287 id = get_id_from_freelist(info); 288 id = get_id_from_freelist(info);
288 info->shadow[id].request = req; 289 info->shadow[id].request = req;
289 290
290 ring_req->id = id; 291 ring_req->u.rw.id = id;
291 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req); 292 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
292 ring_req->handle = info->handle; 293 ring_req->u.rw.handle = info->handle;
293 294
294 ring_req->operation = rq_data_dir(req) ? 295 ring_req->operation = rq_data_dir(req) ?
295 BLKIF_OP_WRITE : BLKIF_OP_READ; 296 BLKIF_OP_WRITE : BLKIF_OP_READ;
@@ -305,16 +306,21 @@ static int blkif_queue_request(struct request *req)
305 ring_req->operation = info->flush_op; 306 ring_req->operation = info->flush_op;
306 } 307 }
307 308
308 if (unlikely(req->cmd_flags & REQ_DISCARD)) { 309 if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
309 /* id, sector_number and handle are set above. */ 310 /* id, sector_number and handle are set above. */
310 ring_req->operation = BLKIF_OP_DISCARD; 311 ring_req->operation = BLKIF_OP_DISCARD;
311 ring_req->nr_segments = 0;
312 ring_req->u.discard.nr_sectors = blk_rq_sectors(req); 312 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
313 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
314 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
315 else
316 ring_req->u.discard.flag = 0;
313 } else { 317 } else {
314 ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg); 318 ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
315 BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST); 319 info->sg);
320 BUG_ON(ring_req->u.rw.nr_segments >
321 BLKIF_MAX_SEGMENTS_PER_REQUEST);
316 322
317 for_each_sg(info->sg, sg, ring_req->nr_segments, i) { 323 for_each_sg(info->sg, sg, ring_req->u.rw.nr_segments, i) {
318 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg))); 324 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
319 fsect = sg->offset >> 9; 325 fsect = sg->offset >> 9;
320 lsect = fsect + (sg->length >> 9) - 1; 326 lsect = fsect + (sg->length >> 9) - 1;
@@ -424,6 +430,8 @@ static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
424 blk_queue_max_discard_sectors(rq, get_capacity(gd)); 430 blk_queue_max_discard_sectors(rq, get_capacity(gd));
425 rq->limits.discard_granularity = info->discard_granularity; 431 rq->limits.discard_granularity = info->discard_granularity;
426 rq->limits.discard_alignment = info->discard_alignment; 432 rq->limits.discard_alignment = info->discard_alignment;
433 if (info->feature_secdiscard)
434 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
427 } 435 }
428 436
429 /* Hard sector size and max sectors impersonate the equiv. hardware. */ 437 /* Hard sector size and max sectors impersonate the equiv. hardware. */
@@ -705,7 +713,9 @@ static void blkif_free(struct blkfront_info *info, int suspend)
705static void blkif_completion(struct blk_shadow *s) 713static void blkif_completion(struct blk_shadow *s)
706{ 714{
707 int i; 715 int i;
708 for (i = 0; i < s->req.nr_segments; i++) 716 /* Do not let BLKIF_OP_DISCARD as nr_segment is in the same place
717 * flag. */
718 for (i = 0; i < s->req.u.rw.nr_segments; i++)
709 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL); 719 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
710} 720}
711 721
@@ -736,7 +746,8 @@ static irqreturn_t blkif_interrupt(int irq, void *dev_id)
736 id = bret->id; 746 id = bret->id;
737 req = info->shadow[id].request; 747 req = info->shadow[id].request;
738 748
739 blkif_completion(&info->shadow[id]); 749 if (bret->operation != BLKIF_OP_DISCARD)
750 blkif_completion(&info->shadow[id]);
740 751
741 add_id_to_freelist(info, id); 752 add_id_to_freelist(info, id);
742 753
@@ -749,7 +760,9 @@ static irqreturn_t blkif_interrupt(int irq, void *dev_id)
749 info->gd->disk_name); 760 info->gd->disk_name);
750 error = -EOPNOTSUPP; 761 error = -EOPNOTSUPP;
751 info->feature_discard = 0; 762 info->feature_discard = 0;
763 info->feature_secdiscard = 0;
752 queue_flag_clear(QUEUE_FLAG_DISCARD, rq); 764 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
765 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
753 } 766 }
754 __blk_end_request_all(req, error); 767 __blk_end_request_all(req, error);
755 break; 768 break;
@@ -763,7 +776,7 @@ static irqreturn_t blkif_interrupt(int irq, void *dev_id)
763 error = -EOPNOTSUPP; 776 error = -EOPNOTSUPP;
764 } 777 }
765 if (unlikely(bret->status == BLKIF_RSP_ERROR && 778 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
766 info->shadow[id].req.nr_segments == 0)) { 779 info->shadow[id].req.u.rw.nr_segments == 0)) {
767 printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n", 780 printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n",
768 info->flush_op == BLKIF_OP_WRITE_BARRIER ? 781 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
769 "barrier" : "flush disk cache", 782 "barrier" : "flush disk cache",
@@ -984,8 +997,8 @@ static int blkfront_probe(struct xenbus_device *dev,
984 INIT_WORK(&info->work, blkif_restart_queue); 997 INIT_WORK(&info->work, blkif_restart_queue);
985 998
986 for (i = 0; i < BLK_RING_SIZE; i++) 999 for (i = 0; i < BLK_RING_SIZE; i++)
987 info->shadow[i].req.id = i+1; 1000 info->shadow[i].req.u.rw.id = i+1;
988 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff; 1001 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
989 1002
990 /* Front end dir is a number, which is used as the id. */ 1003 /* Front end dir is a number, which is used as the id. */
991 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0); 1004 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
@@ -1019,9 +1032,9 @@ static int blkif_recover(struct blkfront_info *info)
1019 /* Stage 2: Set up free list. */ 1032 /* Stage 2: Set up free list. */
1020 memset(&info->shadow, 0, sizeof(info->shadow)); 1033 memset(&info->shadow, 0, sizeof(info->shadow));
1021 for (i = 0; i < BLK_RING_SIZE; i++) 1034 for (i = 0; i < BLK_RING_SIZE; i++)
1022 info->shadow[i].req.id = i+1; 1035 info->shadow[i].req.u.rw.id = i+1;
1023 info->shadow_free = info->ring.req_prod_pvt; 1036 info->shadow_free = info->ring.req_prod_pvt;
1024 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff; 1037 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1025 1038
1026 /* Stage 3: Find pending requests and requeue them. */ 1039 /* Stage 3: Find pending requests and requeue them. */
1027 for (i = 0; i < BLK_RING_SIZE; i++) { 1040 for (i = 0; i < BLK_RING_SIZE; i++) {
@@ -1034,17 +1047,19 @@ static int blkif_recover(struct blkfront_info *info)
1034 *req = copy[i].req; 1047 *req = copy[i].req;
1035 1048
1036 /* We get a new request id, and must reset the shadow state. */ 1049 /* We get a new request id, and must reset the shadow state. */
1037 req->id = get_id_from_freelist(info); 1050 req->u.rw.id = get_id_from_freelist(info);
1038 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i])); 1051 memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1039 1052
1053 if (req->operation != BLKIF_OP_DISCARD) {
1040 /* Rewrite any grant references invalidated by susp/resume. */ 1054 /* Rewrite any grant references invalidated by susp/resume. */
1041 for (j = 0; j < req->nr_segments; j++) 1055 for (j = 0; j < req->u.rw.nr_segments; j++)
1042 gnttab_grant_foreign_access_ref( 1056 gnttab_grant_foreign_access_ref(
1043 req->u.rw.seg[j].gref, 1057 req->u.rw.seg[j].gref,
1044 info->xbdev->otherend_id, 1058 info->xbdev->otherend_id,
1045 pfn_to_mfn(info->shadow[req->id].frame[j]), 1059 pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
1046 rq_data_dir(info->shadow[req->id].request)); 1060 rq_data_dir(info->shadow[req->u.rw.id].request));
1047 info->shadow[req->id].req = *req; 1061 }
1062 info->shadow[req->u.rw.id].req = *req;
1048 1063
1049 info->ring.req_prod_pvt++; 1064 info->ring.req_prod_pvt++;
1050 } 1065 }
@@ -1135,11 +1150,13 @@ static void blkfront_setup_discard(struct blkfront_info *info)
1135 char *type; 1150 char *type;
1136 unsigned int discard_granularity; 1151 unsigned int discard_granularity;
1137 unsigned int discard_alignment; 1152 unsigned int discard_alignment;
1153 unsigned int discard_secure;
1138 1154
1139 type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL); 1155 type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1140 if (IS_ERR(type)) 1156 if (IS_ERR(type))
1141 return; 1157 return;
1142 1158
1159 info->feature_secdiscard = 0;
1143 if (strncmp(type, "phy", 3) == 0) { 1160 if (strncmp(type, "phy", 3) == 0) {
1144 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 1161 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1145 "discard-granularity", "%u", &discard_granularity, 1162 "discard-granularity", "%u", &discard_granularity,
@@ -1150,6 +1167,12 @@ static void blkfront_setup_discard(struct blkfront_info *info)
1150 info->discard_granularity = discard_granularity; 1167 info->discard_granularity = discard_granularity;
1151 info->discard_alignment = discard_alignment; 1168 info->discard_alignment = discard_alignment;
1152 } 1169 }
1170 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1171 "discard-secure", "%d", &discard_secure,
1172 NULL);
1173 if (!err)
1174 info->feature_secdiscard = discard_secure;
1175
1153 } else if (strncmp(type, "file", 4) == 0) 1176 } else if (strncmp(type, "file", 4) == 0)
1154 info->feature_discard = 1; 1177 info->feature_discard = 1;
1155 1178
diff --git a/include/xen/interface/io/blkif.h b/include/xen/interface/io/blkif.h
index 9324488f23f0..ee338bfde18b 100644
--- a/include/xen/interface/io/blkif.h
+++ b/include/xen/interface/io/blkif.h
@@ -84,6 +84,21 @@ typedef uint64_t blkif_sector_t;
84 * e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc 84 * e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc
85 * http://www.seagate.com/staticfiles/support/disc/manuals/ 85 * http://www.seagate.com/staticfiles/support/disc/manuals/
86 * Interface%20manuals/100293068c.pdf 86 * Interface%20manuals/100293068c.pdf
87 * The backend can optionally provide three extra XenBus attributes to
88 * further optimize the discard functionality:
89 * 'discard-aligment' - Devices that support discard functionality may
90 * internally allocate space in units that are bigger than the exported
91 * logical block size. The discard-alignment parameter indicates how many bytes
92 * the beginning of the partition is offset from the internal allocation unit's
93 * natural alignment.
94 * 'discard-granularity' - Devices that support discard functionality may
95 * internally allocate space using units that are bigger than the logical block
96 * size. The discard-granularity parameter indicates the size of the internal
97 * allocation unit in bytes if reported by the device. Otherwise the
98 * discard-granularity will be set to match the device's physical block size.
99 * 'discard-secure' - All copies of the discarded sectors (potentially created
100 * by garbage collection) must also be erased. To use this feature, the flag
101 * BLKIF_DISCARD_SECURE must be set in the blkif_request_trim.
87 */ 102 */
88#define BLKIF_OP_DISCARD 5 103#define BLKIF_OP_DISCARD 5
89 104
@@ -95,6 +110,12 @@ typedef uint64_t blkif_sector_t;
95#define BLKIF_MAX_SEGMENTS_PER_REQUEST 11 110#define BLKIF_MAX_SEGMENTS_PER_REQUEST 11
96 111
97struct blkif_request_rw { 112struct blkif_request_rw {
113 uint8_t nr_segments; /* number of segments */
114 blkif_vdev_t handle; /* only for read/write requests */
115#ifdef CONFIG_X86_64
116 uint32_t _pad1; /* offsetof(blkif_request,u.rw.id) == 8 */
117#endif
118 uint64_t id; /* private guest value, echoed in resp */
98 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 119 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
99 struct blkif_request_segment { 120 struct blkif_request_segment {
100 grant_ref_t gref; /* reference to I/O buffer frame */ 121 grant_ref_t gref; /* reference to I/O buffer frame */
@@ -102,23 +123,28 @@ struct blkif_request_rw {
102 /* @last_sect: last sector in frame to transfer (inclusive). */ 123 /* @last_sect: last sector in frame to transfer (inclusive). */
103 uint8_t first_sect, last_sect; 124 uint8_t first_sect, last_sect;
104 } seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; 125 } seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
105}; 126} __attribute__((__packed__));
106 127
107struct blkif_request_discard { 128struct blkif_request_discard {
129 uint8_t flag; /* BLKIF_DISCARD_SECURE or zero. */
130#define BLKIF_DISCARD_SECURE (1<<0) /* ignored if discard-secure=0 */
131 blkif_vdev_t _pad1; /* only for read/write requests */
132#ifdef CONFIG_X86_64
133 uint32_t _pad2; /* offsetof(blkif_req..,u.discard.id)==8*/
134#endif
135 uint64_t id; /* private guest value, echoed in resp */
108 blkif_sector_t sector_number; 136 blkif_sector_t sector_number;
109 uint64_t nr_sectors; 137 uint64_t nr_sectors;
110}; 138 uint8_t _pad3;
139} __attribute__((__packed__));
111 140
112struct blkif_request { 141struct blkif_request {
113 uint8_t operation; /* BLKIF_OP_??? */ 142 uint8_t operation; /* BLKIF_OP_??? */
114 uint8_t nr_segments; /* number of segments */
115 blkif_vdev_t handle; /* only for read/write requests */
116 uint64_t id; /* private guest value, echoed in resp */
117 union { 143 union {
118 struct blkif_request_rw rw; 144 struct blkif_request_rw rw;
119 struct blkif_request_discard discard; 145 struct blkif_request_discard discard;
120 } u; 146 } u;
121}; 147} __attribute__((__packed__));
122 148
123struct blkif_response { 149struct blkif_response {
124 uint64_t id; /* copied from request */ 150 uint64_t id; /* copied from request */
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-28 13:40:28 -0500