Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 1 | /* |
| 2 | * libata-eh.c - libata error handling |
| 3 | * |
| 4 | * Maintained by: Jeff Garzik <jgarzik@pobox.com> |
| 5 | * Please ALWAYS copy linux-ide@vger.kernel.org |
| 6 | * on emails. |
| 7 | * |
| 8 | * Copyright 2006 Tejun Heo <htejun@gmail.com> |
| 9 | * |
| 10 | * |
| 11 | * This program is free software; you can redistribute it and/or |
| 12 | * modify it under the terms of the GNU General Public License as |
| 13 | * published by the Free Software Foundation; either version 2, or |
| 14 | * (at your option) any later version. |
| 15 | * |
| 16 | * This program is distributed in the hope that it will be useful, |
| 17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 19 | * General Public License for more details. |
| 20 | * |
| 21 | * You should have received a copy of the GNU General Public License |
| 22 | * along with this program; see the file COPYING. If not, write to |
| 23 | * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, |
| 24 | * USA. |
| 25 | * |
| 26 | * |
| 27 | * libata documentation is available via 'make {ps|pdf}docs', |
| 28 | * as Documentation/DocBook/libata.* |
| 29 | * |
| 30 | * Hardware documentation available from http://www.t13.org/ and |
| 31 | * http://www.sata-io.org/ |
| 32 | * |
| 33 | */ |
| 34 | |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 35 | #include <linux/kernel.h> |
| 36 | #include <scsi/scsi.h> |
| 37 | #include <scsi/scsi_host.h> |
| 38 | #include <scsi/scsi_eh.h> |
| 39 | #include <scsi/scsi_device.h> |
| 40 | #include <scsi/scsi_cmnd.h> |
Jeff Garzik | c6fd280 | 2006-08-10 07:31:37 -0400 | [diff] [blame] | 41 | #include "../scsi/scsi_transport_api.h" |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 42 | |
| 43 | #include <linux/libata.h> |
| 44 | |
| 45 | #include "libata.h" |
| 46 | |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 47 | static void __ata_port_freeze(struct ata_port *ap); |
Tejun Heo | 720ba12 | 2006-05-31 18:28:13 +0900 | [diff] [blame] | 48 | static void ata_eh_finish(struct ata_port *ap); |
Tejun Heo | 500530f | 2006-07-03 16:07:27 +0900 | [diff] [blame] | 49 | static void ata_eh_handle_port_suspend(struct ata_port *ap); |
| 50 | static void ata_eh_handle_port_resume(struct ata_port *ap); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 51 | |
Tejun Heo | 0c247c5 | 2006-05-15 20:58:19 +0900 | [diff] [blame] | 52 | static void ata_ering_record(struct ata_ering *ering, int is_io, |
| 53 | unsigned int err_mask) |
| 54 | { |
| 55 | struct ata_ering_entry *ent; |
| 56 | |
| 57 | WARN_ON(!err_mask); |
| 58 | |
| 59 | ering->cursor++; |
| 60 | ering->cursor %= ATA_ERING_SIZE; |
| 61 | |
| 62 | ent = &ering->ring[ering->cursor]; |
| 63 | ent->is_io = is_io; |
| 64 | ent->err_mask = err_mask; |
| 65 | ent->timestamp = get_jiffies_64(); |
| 66 | } |
| 67 | |
| 68 | static struct ata_ering_entry * ata_ering_top(struct ata_ering *ering) |
| 69 | { |
| 70 | struct ata_ering_entry *ent = &ering->ring[ering->cursor]; |
| 71 | if (!ent->err_mask) |
| 72 | return NULL; |
| 73 | return ent; |
| 74 | } |
| 75 | |
| 76 | static int ata_ering_map(struct ata_ering *ering, |
| 77 | int (*map_fn)(struct ata_ering_entry *, void *), |
| 78 | void *arg) |
| 79 | { |
| 80 | int idx, rc = 0; |
| 81 | struct ata_ering_entry *ent; |
| 82 | |
| 83 | idx = ering->cursor; |
| 84 | do { |
| 85 | ent = &ering->ring[idx]; |
| 86 | if (!ent->err_mask) |
| 87 | break; |
| 88 | rc = map_fn(ent, arg); |
| 89 | if (rc) |
| 90 | break; |
| 91 | idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE; |
| 92 | } while (idx != ering->cursor); |
| 93 | |
| 94 | return rc; |
| 95 | } |
| 96 | |
Tejun Heo | 64f65ca | 2006-06-24 20:30:18 +0900 | [diff] [blame] | 97 | static unsigned int ata_eh_dev_action(struct ata_device *dev) |
| 98 | { |
| 99 | struct ata_eh_context *ehc = &dev->ap->eh_context; |
| 100 | |
| 101 | return ehc->i.action | ehc->i.dev_action[dev->devno]; |
| 102 | } |
| 103 | |
Tejun Heo | af181c2 | 2006-06-24 20:30:18 +0900 | [diff] [blame] | 104 | static void ata_eh_clear_action(struct ata_device *dev, |
| 105 | struct ata_eh_info *ehi, unsigned int action) |
| 106 | { |
| 107 | int i; |
| 108 | |
| 109 | if (!dev) { |
| 110 | ehi->action &= ~action; |
| 111 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 112 | ehi->dev_action[i] &= ~action; |
| 113 | } else { |
| 114 | /* doesn't make sense for port-wide EH actions */ |
| 115 | WARN_ON(!(action & ATA_EH_PERDEV_MASK)); |
| 116 | |
| 117 | /* break ehi->action into ehi->dev_action */ |
| 118 | if (ehi->action & action) { |
| 119 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 120 | ehi->dev_action[i] |= ehi->action & action; |
| 121 | ehi->action &= ~action; |
| 122 | } |
| 123 | |
| 124 | /* turn off the specified per-dev action */ |
| 125 | ehi->dev_action[dev->devno] &= ~action; |
| 126 | } |
| 127 | } |
| 128 | |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 129 | /** |
| 130 | * ata_scsi_timed_out - SCSI layer time out callback |
| 131 | * @cmd: timed out SCSI command |
| 132 | * |
| 133 | * Handles SCSI layer timeout. We race with normal completion of |
| 134 | * the qc for @cmd. If the qc is already gone, we lose and let |
| 135 | * the scsi command finish (EH_HANDLED). Otherwise, the qc has |
| 136 | * timed out and EH should be invoked. Prevent ata_qc_complete() |
| 137 | * from finishing it by setting EH_SCHEDULED and return |
| 138 | * EH_NOT_HANDLED. |
| 139 | * |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 140 | * TODO: kill this function once old EH is gone. |
| 141 | * |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 142 | * LOCKING: |
| 143 | * Called from timer context |
| 144 | * |
| 145 | * RETURNS: |
| 146 | * EH_HANDLED or EH_NOT_HANDLED |
| 147 | */ |
| 148 | enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd) |
| 149 | { |
| 150 | struct Scsi_Host *host = cmd->device->host; |
Jeff Garzik | 35bb94b | 2006-04-11 13:12:34 -0400 | [diff] [blame] | 151 | struct ata_port *ap = ata_shost_to_port(host); |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 152 | unsigned long flags; |
| 153 | struct ata_queued_cmd *qc; |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 154 | enum scsi_eh_timer_return ret; |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 155 | |
| 156 | DPRINTK("ENTER\n"); |
| 157 | |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 158 | if (ap->ops->error_handler) { |
| 159 | ret = EH_NOT_HANDLED; |
| 160 | goto out; |
| 161 | } |
| 162 | |
| 163 | ret = EH_HANDLED; |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 164 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 165 | qc = ata_qc_from_tag(ap, ap->active_tag); |
| 166 | if (qc) { |
| 167 | WARN_ON(qc->scsicmd != cmd); |
| 168 | qc->flags |= ATA_QCFLAG_EH_SCHEDULED; |
| 169 | qc->err_mask |= AC_ERR_TIMEOUT; |
| 170 | ret = EH_NOT_HANDLED; |
| 171 | } |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 172 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 173 | |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 174 | out: |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 175 | DPRINTK("EXIT, ret=%d\n", ret); |
| 176 | return ret; |
| 177 | } |
| 178 | |
| 179 | /** |
| 180 | * ata_scsi_error - SCSI layer error handler callback |
| 181 | * @host: SCSI host on which error occurred |
| 182 | * |
| 183 | * Handles SCSI-layer-thrown error events. |
| 184 | * |
| 185 | * LOCKING: |
| 186 | * Inherited from SCSI layer (none, can sleep) |
| 187 | * |
| 188 | * RETURNS: |
| 189 | * Zero. |
| 190 | */ |
Jeff Garzik | 381544b | 2006-04-11 13:04:39 -0400 | [diff] [blame] | 191 | void ata_scsi_error(struct Scsi_Host *host) |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 192 | { |
Jeff Garzik | 35bb94b | 2006-04-11 13:12:34 -0400 | [diff] [blame] | 193 | struct ata_port *ap = ata_shost_to_port(host); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 194 | int i, repeat_cnt = ATA_EH_MAX_REPEAT; |
| 195 | unsigned long flags; |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 196 | |
| 197 | DPRINTK("ENTER\n"); |
| 198 | |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 199 | /* synchronize with port task */ |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 200 | ata_port_flush_task(ap); |
| 201 | |
Jeff Garzik | cca3974 | 2006-08-24 03:19:22 -0400 | [diff] [blame] | 202 | /* synchronize with host lock and sort out timeouts */ |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 203 | |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 204 | /* For new EH, all qcs are finished in one of three ways - |
| 205 | * normal completion, error completion, and SCSI timeout. |
| 206 | * Both cmpletions can race against SCSI timeout. When normal |
| 207 | * completion wins, the qc never reaches EH. When error |
| 208 | * completion wins, the qc has ATA_QCFLAG_FAILED set. |
| 209 | * |
| 210 | * When SCSI timeout wins, things are a bit more complex. |
| 211 | * Normal or error completion can occur after the timeout but |
| 212 | * before this point. In such cases, both types of |
| 213 | * completions are honored. A scmd is determined to have |
| 214 | * timed out iff its associated qc is active and not failed. |
| 215 | */ |
| 216 | if (ap->ops->error_handler) { |
| 217 | struct scsi_cmnd *scmd, *tmp; |
| 218 | int nr_timedout = 0; |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 219 | |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 220 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 221 | |
| 222 | list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) { |
| 223 | struct ata_queued_cmd *qc; |
| 224 | |
| 225 | for (i = 0; i < ATA_MAX_QUEUE; i++) { |
| 226 | qc = __ata_qc_from_tag(ap, i); |
| 227 | if (qc->flags & ATA_QCFLAG_ACTIVE && |
| 228 | qc->scsicmd == scmd) |
| 229 | break; |
| 230 | } |
| 231 | |
| 232 | if (i < ATA_MAX_QUEUE) { |
| 233 | /* the scmd has an associated qc */ |
| 234 | if (!(qc->flags & ATA_QCFLAG_FAILED)) { |
| 235 | /* which hasn't failed yet, timeout */ |
| 236 | qc->err_mask |= AC_ERR_TIMEOUT; |
| 237 | qc->flags |= ATA_QCFLAG_FAILED; |
| 238 | nr_timedout++; |
| 239 | } |
| 240 | } else { |
| 241 | /* Normal completion occurred after |
| 242 | * SCSI timeout but before this point. |
| 243 | * Successfully complete it. |
| 244 | */ |
| 245 | scmd->retries = scmd->allowed; |
| 246 | scsi_eh_finish_cmd(scmd, &ap->eh_done_q); |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | /* If we have timed out qcs. They belong to EH from |
| 251 | * this point but the state of the controller is |
| 252 | * unknown. Freeze the port to make sure the IRQ |
| 253 | * handler doesn't diddle with those qcs. This must |
| 254 | * be done atomically w.r.t. setting QCFLAG_FAILED. |
| 255 | */ |
| 256 | if (nr_timedout) |
| 257 | __ata_port_freeze(ap); |
| 258 | |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 259 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 260 | } else |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 261 | spin_unlock_wait(ap->lock); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 262 | |
| 263 | repeat: |
| 264 | /* invoke error handler */ |
| 265 | if (ap->ops->error_handler) { |
Tejun Heo | 500530f | 2006-07-03 16:07:27 +0900 | [diff] [blame] | 266 | /* process port resume request */ |
| 267 | ata_eh_handle_port_resume(ap); |
| 268 | |
Tejun Heo | f3e81b1 | 2006-05-15 20:58:21 +0900 | [diff] [blame] | 269 | /* fetch & clear EH info */ |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 270 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | f3e81b1 | 2006-05-15 20:58:21 +0900 | [diff] [blame] | 271 | |
| 272 | memset(&ap->eh_context, 0, sizeof(ap->eh_context)); |
| 273 | ap->eh_context.i = ap->eh_info; |
| 274 | memset(&ap->eh_info, 0, sizeof(ap->eh_info)); |
| 275 | |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 276 | ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS; |
| 277 | ap->pflags &= ~ATA_PFLAG_EH_PENDING; |
Tejun Heo | f3e81b1 | 2006-05-15 20:58:21 +0900 | [diff] [blame] | 278 | |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 279 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 280 | |
Tejun Heo | 500530f | 2006-07-03 16:07:27 +0900 | [diff] [blame] | 281 | /* invoke EH, skip if unloading or suspended */ |
| 282 | if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED))) |
Tejun Heo | 720ba12 | 2006-05-31 18:28:13 +0900 | [diff] [blame] | 283 | ap->ops->error_handler(ap); |
| 284 | else |
| 285 | ata_eh_finish(ap); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 286 | |
Tejun Heo | 500530f | 2006-07-03 16:07:27 +0900 | [diff] [blame] | 287 | /* process port suspend request */ |
| 288 | ata_eh_handle_port_suspend(ap); |
| 289 | |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 290 | /* Exception might have happend after ->error_handler |
| 291 | * recovered the port but before this point. Repeat |
| 292 | * EH in such case. |
| 293 | */ |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 294 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 295 | |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 296 | if (ap->pflags & ATA_PFLAG_EH_PENDING) { |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 297 | if (--repeat_cnt) { |
| 298 | ata_port_printk(ap, KERN_INFO, |
| 299 | "EH pending after completion, " |
| 300 | "repeating EH (cnt=%d)\n", repeat_cnt); |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 301 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 302 | goto repeat; |
| 303 | } |
| 304 | ata_port_printk(ap, KERN_ERR, "EH pending after %d " |
| 305 | "tries, giving up\n", ATA_EH_MAX_REPEAT); |
| 306 | } |
| 307 | |
Tejun Heo | f3e81b1 | 2006-05-15 20:58:21 +0900 | [diff] [blame] | 308 | /* this run is complete, make sure EH info is clear */ |
| 309 | memset(&ap->eh_info, 0, sizeof(ap->eh_info)); |
| 310 | |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 311 | /* Clear host_eh_scheduled while holding ap->lock such |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 312 | * that if exception occurs after this point but |
| 313 | * before EH completion, SCSI midlayer will |
| 314 | * re-initiate EH. |
| 315 | */ |
| 316 | host->host_eh_scheduled = 0; |
| 317 | |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 318 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 319 | } else { |
| 320 | WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL); |
| 321 | ap->ops->eng_timeout(ap); |
| 322 | } |
| 323 | |
| 324 | /* finish or retry handled scmd's and clean up */ |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 325 | WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q)); |
| 326 | |
| 327 | scsi_eh_flush_done_q(&ap->eh_done_q); |
| 328 | |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 329 | /* clean up */ |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 330 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 331 | |
Tejun Heo | 1cdaf53 | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 332 | if (ap->pflags & ATA_PFLAG_LOADING) |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 333 | ap->pflags &= ~ATA_PFLAG_LOADING; |
Tejun Heo | 1cdaf53 | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 334 | else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) |
David Howells | 52bad64 | 2006-11-22 14:54:01 +0000 | [diff] [blame] | 335 | queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0); |
Tejun Heo | 1cdaf53 | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 336 | |
| 337 | if (ap->pflags & ATA_PFLAG_RECOVERED) |
| 338 | ata_port_printk(ap, KERN_INFO, "EH complete\n"); |
Tejun Heo | 580b2102 | 2006-05-31 18:28:05 +0900 | [diff] [blame] | 339 | |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 340 | ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 341 | |
Tejun Heo | c6cf9e9 | 2006-05-31 18:27:27 +0900 | [diff] [blame] | 342 | /* tell wait_eh that we're done */ |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 343 | ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS; |
Tejun Heo | c6cf9e9 | 2006-05-31 18:27:27 +0900 | [diff] [blame] | 344 | wake_up_all(&ap->eh_wait_q); |
| 345 | |
Tejun Heo | e30349d | 2006-07-03 03:02:15 +0900 | [diff] [blame] | 346 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 347 | |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 348 | DPRINTK("EXIT\n"); |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 349 | } |
| 350 | |
| 351 | /** |
Tejun Heo | c6cf9e9 | 2006-05-31 18:27:27 +0900 | [diff] [blame] | 352 | * ata_port_wait_eh - Wait for the currently pending EH to complete |
| 353 | * @ap: Port to wait EH for |
| 354 | * |
| 355 | * Wait until the currently pending EH is complete. |
| 356 | * |
| 357 | * LOCKING: |
| 358 | * Kernel thread context (may sleep). |
| 359 | */ |
| 360 | void ata_port_wait_eh(struct ata_port *ap) |
| 361 | { |
| 362 | unsigned long flags; |
| 363 | DEFINE_WAIT(wait); |
| 364 | |
| 365 | retry: |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 366 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | c6cf9e9 | 2006-05-31 18:27:27 +0900 | [diff] [blame] | 367 | |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 368 | while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) { |
Tejun Heo | c6cf9e9 | 2006-05-31 18:27:27 +0900 | [diff] [blame] | 369 | prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE); |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 370 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | c6cf9e9 | 2006-05-31 18:27:27 +0900 | [diff] [blame] | 371 | schedule(); |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 372 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | c6cf9e9 | 2006-05-31 18:27:27 +0900 | [diff] [blame] | 373 | } |
Tejun Heo | 0a1b622 | 2006-06-11 11:01:38 +0900 | [diff] [blame] | 374 | finish_wait(&ap->eh_wait_q, &wait); |
Tejun Heo | c6cf9e9 | 2006-05-31 18:27:27 +0900 | [diff] [blame] | 375 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 376 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | c6cf9e9 | 2006-05-31 18:27:27 +0900 | [diff] [blame] | 377 | |
| 378 | /* make sure SCSI EH is complete */ |
Jeff Garzik | cca3974 | 2006-08-24 03:19:22 -0400 | [diff] [blame] | 379 | if (scsi_host_in_recovery(ap->scsi_host)) { |
Tejun Heo | c6cf9e9 | 2006-05-31 18:27:27 +0900 | [diff] [blame] | 380 | msleep(10); |
| 381 | goto retry; |
| 382 | } |
| 383 | } |
| 384 | |
| 385 | /** |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 386 | * ata_qc_timeout - Handle timeout of queued command |
| 387 | * @qc: Command that timed out |
| 388 | * |
| 389 | * Some part of the kernel (currently, only the SCSI layer) |
| 390 | * has noticed that the active command on port @ap has not |
| 391 | * completed after a specified length of time. Handle this |
| 392 | * condition by disabling DMA (if necessary) and completing |
| 393 | * transactions, with error if necessary. |
| 394 | * |
| 395 | * This also handles the case of the "lost interrupt", where |
| 396 | * for some reason (possibly hardware bug, possibly driver bug) |
| 397 | * an interrupt was not delivered to the driver, even though the |
| 398 | * transaction completed successfully. |
| 399 | * |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 400 | * TODO: kill this function once old EH is gone. |
| 401 | * |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 402 | * LOCKING: |
| 403 | * Inherited from SCSI layer (none, can sleep) |
| 404 | */ |
| 405 | static void ata_qc_timeout(struct ata_queued_cmd *qc) |
| 406 | { |
| 407 | struct ata_port *ap = qc->ap; |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 408 | u8 host_stat = 0, drv_stat; |
| 409 | unsigned long flags; |
| 410 | |
| 411 | DPRINTK("ENTER\n"); |
| 412 | |
| 413 | ap->hsm_task_state = HSM_ST_IDLE; |
| 414 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 415 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 416 | |
| 417 | switch (qc->tf.protocol) { |
| 418 | |
| 419 | case ATA_PROT_DMA: |
| 420 | case ATA_PROT_ATAPI_DMA: |
| 421 | host_stat = ap->ops->bmdma_status(ap); |
| 422 | |
| 423 | /* before we do anything else, clear DMA-Start bit */ |
| 424 | ap->ops->bmdma_stop(qc); |
| 425 | |
| 426 | /* fall through */ |
| 427 | |
| 428 | default: |
| 429 | ata_altstatus(ap); |
| 430 | drv_stat = ata_chk_status(ap); |
| 431 | |
| 432 | /* ack bmdma irq events */ |
| 433 | ap->ops->irq_clear(ap); |
| 434 | |
Tejun Heo | f15a1da | 2006-05-15 20:57:56 +0900 | [diff] [blame] | 435 | ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, " |
| 436 | "stat 0x%x host_stat 0x%x\n", |
| 437 | qc->tf.command, drv_stat, host_stat); |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 438 | |
| 439 | /* complete taskfile transaction */ |
Jeff Garzik | c13b56a | 2006-04-02 10:34:24 -0400 | [diff] [blame] | 440 | qc->err_mask |= AC_ERR_TIMEOUT; |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 441 | break; |
| 442 | } |
| 443 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 444 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 445 | |
| 446 | ata_eh_qc_complete(qc); |
| 447 | |
| 448 | DPRINTK("EXIT\n"); |
| 449 | } |
| 450 | |
| 451 | /** |
| 452 | * ata_eng_timeout - Handle timeout of queued command |
| 453 | * @ap: Port on which timed-out command is active |
| 454 | * |
| 455 | * Some part of the kernel (currently, only the SCSI layer) |
| 456 | * has noticed that the active command on port @ap has not |
| 457 | * completed after a specified length of time. Handle this |
| 458 | * condition by disabling DMA (if necessary) and completing |
| 459 | * transactions, with error if necessary. |
| 460 | * |
| 461 | * This also handles the case of the "lost interrupt", where |
| 462 | * for some reason (possibly hardware bug, possibly driver bug) |
| 463 | * an interrupt was not delivered to the driver, even though the |
| 464 | * transaction completed successfully. |
| 465 | * |
Tejun Heo | ad9e276 | 2006-05-15 20:58:12 +0900 | [diff] [blame] | 466 | * TODO: kill this function once old EH is gone. |
| 467 | * |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 468 | * LOCKING: |
| 469 | * Inherited from SCSI layer (none, can sleep) |
| 470 | */ |
| 471 | void ata_eng_timeout(struct ata_port *ap) |
| 472 | { |
| 473 | DPRINTK("ENTER\n"); |
| 474 | |
| 475 | ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag)); |
| 476 | |
| 477 | DPRINTK("EXIT\n"); |
| 478 | } |
| 479 | |
Tejun Heo | f686bcb | 2006-05-15 20:58:05 +0900 | [diff] [blame] | 480 | /** |
| 481 | * ata_qc_schedule_eh - schedule qc for error handling |
| 482 | * @qc: command to schedule error handling for |
| 483 | * |
| 484 | * Schedule error handling for @qc. EH will kick in as soon as |
| 485 | * other commands are drained. |
| 486 | * |
| 487 | * LOCKING: |
Jeff Garzik | cca3974 | 2006-08-24 03:19:22 -0400 | [diff] [blame] | 488 | * spin_lock_irqsave(host lock) |
Tejun Heo | f686bcb | 2006-05-15 20:58:05 +0900 | [diff] [blame] | 489 | */ |
| 490 | void ata_qc_schedule_eh(struct ata_queued_cmd *qc) |
| 491 | { |
| 492 | struct ata_port *ap = qc->ap; |
| 493 | |
| 494 | WARN_ON(!ap->ops->error_handler); |
| 495 | |
| 496 | qc->flags |= ATA_QCFLAG_FAILED; |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 497 | qc->ap->pflags |= ATA_PFLAG_EH_PENDING; |
Tejun Heo | f686bcb | 2006-05-15 20:58:05 +0900 | [diff] [blame] | 498 | |
| 499 | /* The following will fail if timeout has already expired. |
| 500 | * ata_scsi_error() takes care of such scmds on EH entry. |
| 501 | * Note that ATA_QCFLAG_FAILED is unconditionally set after |
| 502 | * this function completes. |
| 503 | */ |
| 504 | scsi_req_abort_cmd(qc->scsicmd); |
| 505 | } |
| 506 | |
Tejun Heo | 7b70fc0 | 2006-05-15 20:58:07 +0900 | [diff] [blame] | 507 | /** |
| 508 | * ata_port_schedule_eh - schedule error handling without a qc |
| 509 | * @ap: ATA port to schedule EH for |
| 510 | * |
| 511 | * Schedule error handling for @ap. EH will kick in as soon as |
| 512 | * all commands are drained. |
| 513 | * |
| 514 | * LOCKING: |
Jeff Garzik | cca3974 | 2006-08-24 03:19:22 -0400 | [diff] [blame] | 515 | * spin_lock_irqsave(host lock) |
Tejun Heo | 7b70fc0 | 2006-05-15 20:58:07 +0900 | [diff] [blame] | 516 | */ |
| 517 | void ata_port_schedule_eh(struct ata_port *ap) |
| 518 | { |
| 519 | WARN_ON(!ap->ops->error_handler); |
| 520 | |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 521 | ap->pflags |= ATA_PFLAG_EH_PENDING; |
Jeff Garzik | cca3974 | 2006-08-24 03:19:22 -0400 | [diff] [blame] | 522 | scsi_schedule_eh(ap->scsi_host); |
Tejun Heo | 7b70fc0 | 2006-05-15 20:58:07 +0900 | [diff] [blame] | 523 | |
| 524 | DPRINTK("port EH scheduled\n"); |
| 525 | } |
| 526 | |
| 527 | /** |
| 528 | * ata_port_abort - abort all qc's on the port |
| 529 | * @ap: ATA port to abort qc's for |
| 530 | * |
| 531 | * Abort all active qc's of @ap and schedule EH. |
| 532 | * |
| 533 | * LOCKING: |
Jeff Garzik | cca3974 | 2006-08-24 03:19:22 -0400 | [diff] [blame] | 534 | * spin_lock_irqsave(host lock) |
Tejun Heo | 7b70fc0 | 2006-05-15 20:58:07 +0900 | [diff] [blame] | 535 | * |
| 536 | * RETURNS: |
| 537 | * Number of aborted qc's. |
| 538 | */ |
| 539 | int ata_port_abort(struct ata_port *ap) |
| 540 | { |
| 541 | int tag, nr_aborted = 0; |
| 542 | |
| 543 | WARN_ON(!ap->ops->error_handler); |
| 544 | |
| 545 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 546 | struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag); |
| 547 | |
| 548 | if (qc) { |
| 549 | qc->flags |= ATA_QCFLAG_FAILED; |
| 550 | ata_qc_complete(qc); |
| 551 | nr_aborted++; |
| 552 | } |
| 553 | } |
| 554 | |
| 555 | if (!nr_aborted) |
| 556 | ata_port_schedule_eh(ap); |
| 557 | |
| 558 | return nr_aborted; |
| 559 | } |
| 560 | |
Tejun Heo | e318049 | 2006-05-15 20:58:09 +0900 | [diff] [blame] | 561 | /** |
| 562 | * __ata_port_freeze - freeze port |
| 563 | * @ap: ATA port to freeze |
| 564 | * |
| 565 | * This function is called when HSM violation or some other |
| 566 | * condition disrupts normal operation of the port. Frozen port |
| 567 | * is not allowed to perform any operation until the port is |
| 568 | * thawed, which usually follows a successful reset. |
| 569 | * |
| 570 | * ap->ops->freeze() callback can be used for freezing the port |
| 571 | * hardware-wise (e.g. mask interrupt and stop DMA engine). If a |
| 572 | * port cannot be frozen hardware-wise, the interrupt handler |
| 573 | * must ack and clear interrupts unconditionally while the port |
| 574 | * is frozen. |
| 575 | * |
| 576 | * LOCKING: |
Jeff Garzik | cca3974 | 2006-08-24 03:19:22 -0400 | [diff] [blame] | 577 | * spin_lock_irqsave(host lock) |
Tejun Heo | e318049 | 2006-05-15 20:58:09 +0900 | [diff] [blame] | 578 | */ |
| 579 | static void __ata_port_freeze(struct ata_port *ap) |
| 580 | { |
| 581 | WARN_ON(!ap->ops->error_handler); |
| 582 | |
| 583 | if (ap->ops->freeze) |
| 584 | ap->ops->freeze(ap); |
| 585 | |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 586 | ap->pflags |= ATA_PFLAG_FROZEN; |
Tejun Heo | e318049 | 2006-05-15 20:58:09 +0900 | [diff] [blame] | 587 | |
| 588 | DPRINTK("ata%u port frozen\n", ap->id); |
| 589 | } |
| 590 | |
| 591 | /** |
| 592 | * ata_port_freeze - abort & freeze port |
| 593 | * @ap: ATA port to freeze |
| 594 | * |
| 595 | * Abort and freeze @ap. |
| 596 | * |
| 597 | * LOCKING: |
Jeff Garzik | cca3974 | 2006-08-24 03:19:22 -0400 | [diff] [blame] | 598 | * spin_lock_irqsave(host lock) |
Tejun Heo | e318049 | 2006-05-15 20:58:09 +0900 | [diff] [blame] | 599 | * |
| 600 | * RETURNS: |
| 601 | * Number of aborted commands. |
| 602 | */ |
| 603 | int ata_port_freeze(struct ata_port *ap) |
| 604 | { |
| 605 | int nr_aborted; |
| 606 | |
| 607 | WARN_ON(!ap->ops->error_handler); |
| 608 | |
| 609 | nr_aborted = ata_port_abort(ap); |
| 610 | __ata_port_freeze(ap); |
| 611 | |
| 612 | return nr_aborted; |
| 613 | } |
| 614 | |
| 615 | /** |
| 616 | * ata_eh_freeze_port - EH helper to freeze port |
| 617 | * @ap: ATA port to freeze |
| 618 | * |
| 619 | * Freeze @ap. |
| 620 | * |
| 621 | * LOCKING: |
| 622 | * None. |
| 623 | */ |
| 624 | void ata_eh_freeze_port(struct ata_port *ap) |
| 625 | { |
| 626 | unsigned long flags; |
| 627 | |
| 628 | if (!ap->ops->error_handler) |
| 629 | return; |
| 630 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 631 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | e318049 | 2006-05-15 20:58:09 +0900 | [diff] [blame] | 632 | __ata_port_freeze(ap); |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 633 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | e318049 | 2006-05-15 20:58:09 +0900 | [diff] [blame] | 634 | } |
| 635 | |
| 636 | /** |
| 637 | * ata_port_thaw_port - EH helper to thaw port |
| 638 | * @ap: ATA port to thaw |
| 639 | * |
| 640 | * Thaw frozen port @ap. |
| 641 | * |
| 642 | * LOCKING: |
| 643 | * None. |
| 644 | */ |
| 645 | void ata_eh_thaw_port(struct ata_port *ap) |
| 646 | { |
| 647 | unsigned long flags; |
| 648 | |
| 649 | if (!ap->ops->error_handler) |
| 650 | return; |
| 651 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 652 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | e318049 | 2006-05-15 20:58:09 +0900 | [diff] [blame] | 653 | |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 654 | ap->pflags &= ~ATA_PFLAG_FROZEN; |
Tejun Heo | e318049 | 2006-05-15 20:58:09 +0900 | [diff] [blame] | 655 | |
| 656 | if (ap->ops->thaw) |
| 657 | ap->ops->thaw(ap); |
| 658 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 659 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | e318049 | 2006-05-15 20:58:09 +0900 | [diff] [blame] | 660 | |
| 661 | DPRINTK("ata%u port thawed\n", ap->id); |
| 662 | } |
| 663 | |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 664 | static void ata_eh_scsidone(struct scsi_cmnd *scmd) |
| 665 | { |
| 666 | /* nada */ |
| 667 | } |
| 668 | |
| 669 | static void __ata_eh_qc_complete(struct ata_queued_cmd *qc) |
| 670 | { |
| 671 | struct ata_port *ap = qc->ap; |
| 672 | struct scsi_cmnd *scmd = qc->scsicmd; |
| 673 | unsigned long flags; |
| 674 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 675 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 676 | qc->scsidone = ata_eh_scsidone; |
| 677 | __ata_qc_complete(qc); |
| 678 | WARN_ON(ata_tag_valid(qc->tag)); |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 679 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | ece1d63 | 2006-04-02 18:51:53 +0900 | [diff] [blame] | 680 | |
| 681 | scsi_eh_finish_cmd(scmd, &ap->eh_done_q); |
| 682 | } |
| 683 | |
| 684 | /** |
| 685 | * ata_eh_qc_complete - Complete an active ATA command from EH |
| 686 | * @qc: Command to complete |
| 687 | * |
| 688 | * Indicate to the mid and upper layers that an ATA command has |
| 689 | * completed. To be used from EH. |
| 690 | */ |
| 691 | void ata_eh_qc_complete(struct ata_queued_cmd *qc) |
| 692 | { |
| 693 | struct scsi_cmnd *scmd = qc->scsicmd; |
| 694 | scmd->retries = scmd->allowed; |
| 695 | __ata_eh_qc_complete(qc); |
| 696 | } |
| 697 | |
| 698 | /** |
| 699 | * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH |
| 700 | * @qc: Command to retry |
| 701 | * |
| 702 | * Indicate to the mid and upper layers that an ATA command |
| 703 | * should be retried. To be used from EH. |
| 704 | * |
| 705 | * SCSI midlayer limits the number of retries to scmd->allowed. |
| 706 | * scmd->retries is decremented for commands which get retried |
| 707 | * due to unrelated failures (qc->err_mask is zero). |
| 708 | */ |
| 709 | void ata_eh_qc_retry(struct ata_queued_cmd *qc) |
| 710 | { |
| 711 | struct scsi_cmnd *scmd = qc->scsicmd; |
| 712 | if (!qc->err_mask && scmd->retries) |
| 713 | scmd->retries--; |
| 714 | __ata_eh_qc_complete(qc); |
| 715 | } |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 716 | |
| 717 | /** |
Tejun Heo | 0ea035a | 2006-05-31 18:28:01 +0900 | [diff] [blame] | 718 | * ata_eh_detach_dev - detach ATA device |
| 719 | * @dev: ATA device to detach |
| 720 | * |
| 721 | * Detach @dev. |
| 722 | * |
| 723 | * LOCKING: |
| 724 | * None. |
| 725 | */ |
| 726 | static void ata_eh_detach_dev(struct ata_device *dev) |
| 727 | { |
| 728 | struct ata_port *ap = dev->ap; |
| 729 | unsigned long flags; |
| 730 | |
| 731 | ata_dev_disable(dev); |
| 732 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 733 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | 0ea035a | 2006-05-31 18:28:01 +0900 | [diff] [blame] | 734 | |
| 735 | dev->flags &= ~ATA_DFLAG_DETACH; |
| 736 | |
| 737 | if (ata_scsi_offline_dev(dev)) { |
| 738 | dev->flags |= ATA_DFLAG_DETACHED; |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 739 | ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG; |
Tejun Heo | 0ea035a | 2006-05-31 18:28:01 +0900 | [diff] [blame] | 740 | } |
| 741 | |
Tejun Heo | beb07c1 | 2006-06-24 20:30:19 +0900 | [diff] [blame] | 742 | /* clear per-dev EH actions */ |
| 743 | ata_eh_clear_action(dev, &ap->eh_info, ATA_EH_PERDEV_MASK); |
| 744 | ata_eh_clear_action(dev, &ap->eh_context.i, ATA_EH_PERDEV_MASK); |
| 745 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 746 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | 0ea035a | 2006-05-31 18:28:01 +0900 | [diff] [blame] | 747 | } |
| 748 | |
| 749 | /** |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 750 | * ata_eh_about_to_do - about to perform eh_action |
| 751 | * @ap: target ATA port |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 752 | * @dev: target ATA dev for per-dev action (can be NULL) |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 753 | * @action: action about to be performed |
| 754 | * |
| 755 | * Called just before performing EH actions to clear related bits |
| 756 | * in @ap->eh_info such that eh actions are not unnecessarily |
| 757 | * repeated. |
| 758 | * |
| 759 | * LOCKING: |
| 760 | * None. |
| 761 | */ |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 762 | static void ata_eh_about_to_do(struct ata_port *ap, struct ata_device *dev, |
| 763 | unsigned int action) |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 764 | { |
| 765 | unsigned long flags; |
Tejun Heo | 13abf50 | 2006-07-10 23:18:46 +0900 | [diff] [blame] | 766 | struct ata_eh_info *ehi = &ap->eh_info; |
| 767 | struct ata_eh_context *ehc = &ap->eh_context; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 768 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 769 | spin_lock_irqsave(ap->lock, flags); |
Tejun Heo | 1cdaf53 | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 770 | |
Tejun Heo | 13abf50 | 2006-07-10 23:18:46 +0900 | [diff] [blame] | 771 | /* Reset is represented by combination of actions and EHI |
| 772 | * flags. Suck in all related bits before clearing eh_info to |
| 773 | * avoid losing requested action. |
| 774 | */ |
| 775 | if (action & ATA_EH_RESET_MASK) { |
| 776 | ehc->i.action |= ehi->action & ATA_EH_RESET_MASK; |
| 777 | ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK; |
Tejun Heo | 1cdaf53 | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 778 | |
Tejun Heo | 13abf50 | 2006-07-10 23:18:46 +0900 | [diff] [blame] | 779 | /* make sure all reset actions are cleared & clear EHI flags */ |
| 780 | action |= ATA_EH_RESET_MASK; |
| 781 | ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK; |
| 782 | } |
| 783 | |
| 784 | ata_eh_clear_action(dev, ehi, action); |
| 785 | |
| 786 | if (!(ehc->i.flags & ATA_EHI_QUIET)) |
Tejun Heo | 1cdaf53 | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 787 | ap->pflags |= ATA_PFLAG_RECOVERED; |
| 788 | |
Jeff Garzik | ba6a130 | 2006-06-22 23:46:10 -0400 | [diff] [blame] | 789 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 790 | } |
| 791 | |
| 792 | /** |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 793 | * ata_eh_done - EH action complete |
| 794 | * @ap: target ATA port |
| 795 | * @dev: target ATA dev for per-dev action (can be NULL) |
| 796 | * @action: action just completed |
| 797 | * |
| 798 | * Called right after performing EH actions to clear related bits |
| 799 | * in @ap->eh_context. |
| 800 | * |
| 801 | * LOCKING: |
| 802 | * None. |
| 803 | */ |
| 804 | static void ata_eh_done(struct ata_port *ap, struct ata_device *dev, |
| 805 | unsigned int action) |
| 806 | { |
Tejun Heo | 13abf50 | 2006-07-10 23:18:46 +0900 | [diff] [blame] | 807 | /* if reset is complete, clear all reset actions & reset modifier */ |
| 808 | if (action & ATA_EH_RESET_MASK) { |
| 809 | action |= ATA_EH_RESET_MASK; |
| 810 | ap->eh_context.i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK; |
| 811 | } |
| 812 | |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 813 | ata_eh_clear_action(dev, &ap->eh_context.i, action); |
| 814 | } |
| 815 | |
| 816 | /** |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 817 | * ata_err_string - convert err_mask to descriptive string |
| 818 | * @err_mask: error mask to convert to string |
| 819 | * |
| 820 | * Convert @err_mask to descriptive string. Errors are |
| 821 | * prioritized according to severity and only the most severe |
| 822 | * error is reported. |
| 823 | * |
| 824 | * LOCKING: |
| 825 | * None. |
| 826 | * |
| 827 | * RETURNS: |
| 828 | * Descriptive string for @err_mask |
| 829 | */ |
| 830 | static const char * ata_err_string(unsigned int err_mask) |
| 831 | { |
| 832 | if (err_mask & AC_ERR_HOST_BUS) |
| 833 | return "host bus error"; |
| 834 | if (err_mask & AC_ERR_ATA_BUS) |
| 835 | return "ATA bus error"; |
| 836 | if (err_mask & AC_ERR_TIMEOUT) |
| 837 | return "timeout"; |
| 838 | if (err_mask & AC_ERR_HSM) |
| 839 | return "HSM violation"; |
| 840 | if (err_mask & AC_ERR_SYSTEM) |
| 841 | return "internal error"; |
| 842 | if (err_mask & AC_ERR_MEDIA) |
| 843 | return "media error"; |
| 844 | if (err_mask & AC_ERR_INVALID) |
| 845 | return "invalid argument"; |
| 846 | if (err_mask & AC_ERR_DEV) |
| 847 | return "device error"; |
| 848 | return "unknown error"; |
| 849 | } |
| 850 | |
| 851 | /** |
Tejun Heo | e8ee845 | 2006-05-15 21:03:46 +0900 | [diff] [blame] | 852 | * ata_read_log_page - read a specific log page |
| 853 | * @dev: target device |
| 854 | * @page: page to read |
| 855 | * @buf: buffer to store read page |
| 856 | * @sectors: number of sectors to read |
| 857 | * |
| 858 | * Read log page using READ_LOG_EXT command. |
| 859 | * |
| 860 | * LOCKING: |
| 861 | * Kernel thread context (may sleep). |
| 862 | * |
| 863 | * RETURNS: |
| 864 | * 0 on success, AC_ERR_* mask otherwise. |
| 865 | */ |
| 866 | static unsigned int ata_read_log_page(struct ata_device *dev, |
| 867 | u8 page, void *buf, unsigned int sectors) |
| 868 | { |
| 869 | struct ata_taskfile tf; |
| 870 | unsigned int err_mask; |
| 871 | |
| 872 | DPRINTK("read log page - page %d\n", page); |
| 873 | |
| 874 | ata_tf_init(dev, &tf); |
| 875 | tf.command = ATA_CMD_READ_LOG_EXT; |
| 876 | tf.lbal = page; |
| 877 | tf.nsect = sectors; |
| 878 | tf.hob_nsect = sectors >> 8; |
| 879 | tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE; |
| 880 | tf.protocol = ATA_PROT_PIO; |
| 881 | |
| 882 | err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE, |
| 883 | buf, sectors * ATA_SECT_SIZE); |
| 884 | |
| 885 | DPRINTK("EXIT, err_mask=%x\n", err_mask); |
| 886 | return err_mask; |
| 887 | } |
| 888 | |
| 889 | /** |
| 890 | * ata_eh_read_log_10h - Read log page 10h for NCQ error details |
| 891 | * @dev: Device to read log page 10h from |
| 892 | * @tag: Resulting tag of the failed command |
| 893 | * @tf: Resulting taskfile registers of the failed command |
| 894 | * |
| 895 | * Read log page 10h to obtain NCQ error details and clear error |
| 896 | * condition. |
| 897 | * |
| 898 | * LOCKING: |
| 899 | * Kernel thread context (may sleep). |
| 900 | * |
| 901 | * RETURNS: |
| 902 | * 0 on success, -errno otherwise. |
| 903 | */ |
| 904 | static int ata_eh_read_log_10h(struct ata_device *dev, |
| 905 | int *tag, struct ata_taskfile *tf) |
| 906 | { |
| 907 | u8 *buf = dev->ap->sector_buf; |
| 908 | unsigned int err_mask; |
| 909 | u8 csum; |
| 910 | int i; |
| 911 | |
| 912 | err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1); |
| 913 | if (err_mask) |
| 914 | return -EIO; |
| 915 | |
| 916 | csum = 0; |
| 917 | for (i = 0; i < ATA_SECT_SIZE; i++) |
| 918 | csum += buf[i]; |
| 919 | if (csum) |
| 920 | ata_dev_printk(dev, KERN_WARNING, |
| 921 | "invalid checksum 0x%x on log page 10h\n", csum); |
| 922 | |
| 923 | if (buf[0] & 0x80) |
| 924 | return -ENOENT; |
| 925 | |
| 926 | *tag = buf[0] & 0x1f; |
| 927 | |
| 928 | tf->command = buf[2]; |
| 929 | tf->feature = buf[3]; |
| 930 | tf->lbal = buf[4]; |
| 931 | tf->lbam = buf[5]; |
| 932 | tf->lbah = buf[6]; |
| 933 | tf->device = buf[7]; |
| 934 | tf->hob_lbal = buf[8]; |
| 935 | tf->hob_lbam = buf[9]; |
| 936 | tf->hob_lbah = buf[10]; |
| 937 | tf->nsect = buf[12]; |
| 938 | tf->hob_nsect = buf[13]; |
| 939 | |
| 940 | return 0; |
| 941 | } |
| 942 | |
| 943 | /** |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 944 | * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE |
| 945 | * @dev: device to perform REQUEST_SENSE to |
| 946 | * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long) |
| 947 | * |
| 948 | * Perform ATAPI REQUEST_SENSE after the device reported CHECK |
| 949 | * SENSE. This function is EH helper. |
| 950 | * |
| 951 | * LOCKING: |
| 952 | * Kernel thread context (may sleep). |
| 953 | * |
| 954 | * RETURNS: |
| 955 | * 0 on success, AC_ERR_* mask on failure |
| 956 | */ |
| 957 | static unsigned int atapi_eh_request_sense(struct ata_device *dev, |
| 958 | unsigned char *sense_buf) |
| 959 | { |
| 960 | struct ata_port *ap = dev->ap; |
| 961 | struct ata_taskfile tf; |
| 962 | u8 cdb[ATAPI_CDB_LEN]; |
| 963 | |
| 964 | DPRINTK("ATAPI request sense\n"); |
| 965 | |
| 966 | ata_tf_init(dev, &tf); |
| 967 | |
| 968 | /* FIXME: is this needed? */ |
| 969 | memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE); |
| 970 | |
| 971 | /* XXX: why tf_read here? */ |
| 972 | ap->ops->tf_read(ap, &tf); |
| 973 | |
| 974 | /* fill these in, for the case where they are -not- overwritten */ |
| 975 | sense_buf[0] = 0x70; |
| 976 | sense_buf[2] = tf.feature >> 4; |
| 977 | |
| 978 | memset(cdb, 0, ATAPI_CDB_LEN); |
| 979 | cdb[0] = REQUEST_SENSE; |
| 980 | cdb[4] = SCSI_SENSE_BUFFERSIZE; |
| 981 | |
| 982 | tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE; |
| 983 | tf.command = ATA_CMD_PACKET; |
| 984 | |
| 985 | /* is it pointless to prefer PIO for "safety reasons"? */ |
| 986 | if (ap->flags & ATA_FLAG_PIO_DMA) { |
| 987 | tf.protocol = ATA_PROT_ATAPI_DMA; |
| 988 | tf.feature |= ATAPI_PKT_DMA; |
| 989 | } else { |
| 990 | tf.protocol = ATA_PROT_ATAPI; |
| 991 | tf.lbam = (8 * 1024) & 0xff; |
| 992 | tf.lbah = (8 * 1024) >> 8; |
| 993 | } |
| 994 | |
| 995 | return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE, |
| 996 | sense_buf, SCSI_SENSE_BUFFERSIZE); |
| 997 | } |
| 998 | |
| 999 | /** |
| 1000 | * ata_eh_analyze_serror - analyze SError for a failed port |
| 1001 | * @ap: ATA port to analyze SError for |
| 1002 | * |
| 1003 | * Analyze SError if available and further determine cause of |
| 1004 | * failure. |
| 1005 | * |
| 1006 | * LOCKING: |
| 1007 | * None. |
| 1008 | */ |
| 1009 | static void ata_eh_analyze_serror(struct ata_port *ap) |
| 1010 | { |
| 1011 | struct ata_eh_context *ehc = &ap->eh_context; |
| 1012 | u32 serror = ehc->i.serror; |
| 1013 | unsigned int err_mask = 0, action = 0; |
| 1014 | |
| 1015 | if (serror & SERR_PERSISTENT) { |
| 1016 | err_mask |= AC_ERR_ATA_BUS; |
| 1017 | action |= ATA_EH_HARDRESET; |
| 1018 | } |
| 1019 | if (serror & |
| 1020 | (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) { |
| 1021 | err_mask |= AC_ERR_ATA_BUS; |
| 1022 | action |= ATA_EH_SOFTRESET; |
| 1023 | } |
| 1024 | if (serror & SERR_PROTOCOL) { |
| 1025 | err_mask |= AC_ERR_HSM; |
| 1026 | action |= ATA_EH_SOFTRESET; |
| 1027 | } |
| 1028 | if (serror & SERR_INTERNAL) { |
| 1029 | err_mask |= AC_ERR_SYSTEM; |
| 1030 | action |= ATA_EH_SOFTRESET; |
| 1031 | } |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 1032 | if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG)) |
| 1033 | ata_ehi_hotplugged(&ehc->i); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1034 | |
| 1035 | ehc->i.err_mask |= err_mask; |
| 1036 | ehc->i.action |= action; |
| 1037 | } |
| 1038 | |
| 1039 | /** |
Tejun Heo | e8ee845 | 2006-05-15 21:03:46 +0900 | [diff] [blame] | 1040 | * ata_eh_analyze_ncq_error - analyze NCQ error |
| 1041 | * @ap: ATA port to analyze NCQ error for |
| 1042 | * |
| 1043 | * Read log page 10h, determine the offending qc and acquire |
| 1044 | * error status TF. For NCQ device errors, all LLDDs have to do |
| 1045 | * is setting AC_ERR_DEV in ehi->err_mask. This function takes |
| 1046 | * care of the rest. |
| 1047 | * |
| 1048 | * LOCKING: |
| 1049 | * Kernel thread context (may sleep). |
| 1050 | */ |
| 1051 | static void ata_eh_analyze_ncq_error(struct ata_port *ap) |
| 1052 | { |
| 1053 | struct ata_eh_context *ehc = &ap->eh_context; |
| 1054 | struct ata_device *dev = ap->device; |
| 1055 | struct ata_queued_cmd *qc; |
| 1056 | struct ata_taskfile tf; |
| 1057 | int tag, rc; |
| 1058 | |
| 1059 | /* if frozen, we can't do much */ |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 1060 | if (ap->pflags & ATA_PFLAG_FROZEN) |
Tejun Heo | e8ee845 | 2006-05-15 21:03:46 +0900 | [diff] [blame] | 1061 | return; |
| 1062 | |
| 1063 | /* is it NCQ device error? */ |
| 1064 | if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV)) |
| 1065 | return; |
| 1066 | |
| 1067 | /* has LLDD analyzed already? */ |
| 1068 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 1069 | qc = __ata_qc_from_tag(ap, tag); |
| 1070 | |
| 1071 | if (!(qc->flags & ATA_QCFLAG_FAILED)) |
| 1072 | continue; |
| 1073 | |
| 1074 | if (qc->err_mask) |
| 1075 | return; |
| 1076 | } |
| 1077 | |
| 1078 | /* okay, this error is ours */ |
| 1079 | rc = ata_eh_read_log_10h(dev, &tag, &tf); |
| 1080 | if (rc) { |
| 1081 | ata_port_printk(ap, KERN_ERR, "failed to read log page 10h " |
| 1082 | "(errno=%d)\n", rc); |
| 1083 | return; |
| 1084 | } |
| 1085 | |
| 1086 | if (!(ap->sactive & (1 << tag))) { |
| 1087 | ata_port_printk(ap, KERN_ERR, "log page 10h reported " |
| 1088 | "inactive tag %d\n", tag); |
| 1089 | return; |
| 1090 | } |
| 1091 | |
| 1092 | /* we've got the perpetrator, condemn it */ |
| 1093 | qc = __ata_qc_from_tag(ap, tag); |
| 1094 | memcpy(&qc->result_tf, &tf, sizeof(tf)); |
| 1095 | qc->err_mask |= AC_ERR_DEV; |
| 1096 | ehc->i.err_mask &= ~AC_ERR_DEV; |
| 1097 | } |
| 1098 | |
| 1099 | /** |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1100 | * ata_eh_analyze_tf - analyze taskfile of a failed qc |
| 1101 | * @qc: qc to analyze |
| 1102 | * @tf: Taskfile registers to analyze |
| 1103 | * |
| 1104 | * Analyze taskfile of @qc and further determine cause of |
| 1105 | * failure. This function also requests ATAPI sense data if |
| 1106 | * avaliable. |
| 1107 | * |
| 1108 | * LOCKING: |
| 1109 | * Kernel thread context (may sleep). |
| 1110 | * |
| 1111 | * RETURNS: |
| 1112 | * Determined recovery action |
| 1113 | */ |
| 1114 | static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc, |
| 1115 | const struct ata_taskfile *tf) |
| 1116 | { |
| 1117 | unsigned int tmp, action = 0; |
| 1118 | u8 stat = tf->command, err = tf->feature; |
| 1119 | |
| 1120 | if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) { |
| 1121 | qc->err_mask |= AC_ERR_HSM; |
| 1122 | return ATA_EH_SOFTRESET; |
| 1123 | } |
| 1124 | |
| 1125 | if (!(qc->err_mask & AC_ERR_DEV)) |
| 1126 | return 0; |
| 1127 | |
| 1128 | switch (qc->dev->class) { |
| 1129 | case ATA_DEV_ATA: |
| 1130 | if (err & ATA_ICRC) |
| 1131 | qc->err_mask |= AC_ERR_ATA_BUS; |
| 1132 | if (err & ATA_UNC) |
| 1133 | qc->err_mask |= AC_ERR_MEDIA; |
| 1134 | if (err & ATA_IDNF) |
| 1135 | qc->err_mask |= AC_ERR_INVALID; |
| 1136 | break; |
| 1137 | |
| 1138 | case ATA_DEV_ATAPI: |
Tejun Heo | a569a30 | 2006-11-21 10:40:51 +0900 | [diff] [blame] | 1139 | if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) { |
| 1140 | tmp = atapi_eh_request_sense(qc->dev, |
| 1141 | qc->scsicmd->sense_buffer); |
| 1142 | if (!tmp) { |
| 1143 | /* ATA_QCFLAG_SENSE_VALID is used to |
| 1144 | * tell atapi_qc_complete() that sense |
| 1145 | * data is already valid. |
| 1146 | * |
| 1147 | * TODO: interpret sense data and set |
| 1148 | * appropriate err_mask. |
| 1149 | */ |
| 1150 | qc->flags |= ATA_QCFLAG_SENSE_VALID; |
| 1151 | } else |
| 1152 | qc->err_mask |= tmp; |
| 1153 | } |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1154 | } |
| 1155 | |
| 1156 | if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS)) |
| 1157 | action |= ATA_EH_SOFTRESET; |
| 1158 | |
| 1159 | return action; |
| 1160 | } |
| 1161 | |
| 1162 | static int ata_eh_categorize_ering_entry(struct ata_ering_entry *ent) |
| 1163 | { |
| 1164 | if (ent->err_mask & (AC_ERR_ATA_BUS | AC_ERR_TIMEOUT)) |
| 1165 | return 1; |
| 1166 | |
| 1167 | if (ent->is_io) { |
| 1168 | if (ent->err_mask & AC_ERR_HSM) |
| 1169 | return 1; |
| 1170 | if ((ent->err_mask & |
| 1171 | (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV) |
| 1172 | return 2; |
| 1173 | } |
| 1174 | |
| 1175 | return 0; |
| 1176 | } |
| 1177 | |
| 1178 | struct speed_down_needed_arg { |
| 1179 | u64 since; |
| 1180 | int nr_errors[3]; |
| 1181 | }; |
| 1182 | |
| 1183 | static int speed_down_needed_cb(struct ata_ering_entry *ent, void *void_arg) |
| 1184 | { |
| 1185 | struct speed_down_needed_arg *arg = void_arg; |
| 1186 | |
| 1187 | if (ent->timestamp < arg->since) |
| 1188 | return -1; |
| 1189 | |
| 1190 | arg->nr_errors[ata_eh_categorize_ering_entry(ent)]++; |
| 1191 | return 0; |
| 1192 | } |
| 1193 | |
| 1194 | /** |
| 1195 | * ata_eh_speed_down_needed - Determine wheter speed down is necessary |
| 1196 | * @dev: Device of interest |
| 1197 | * |
| 1198 | * This function examines error ring of @dev and determines |
| 1199 | * whether speed down is necessary. Speed down is necessary if |
| 1200 | * there have been more than 3 of Cat-1 errors or 10 of Cat-2 |
| 1201 | * errors during last 15 minutes. |
| 1202 | * |
| 1203 | * Cat-1 errors are ATA_BUS, TIMEOUT for any command and HSM |
| 1204 | * violation for known supported commands. |
| 1205 | * |
| 1206 | * Cat-2 errors are unclassified DEV error for known supported |
| 1207 | * command. |
| 1208 | * |
| 1209 | * LOCKING: |
| 1210 | * Inherited from caller. |
| 1211 | * |
| 1212 | * RETURNS: |
| 1213 | * 1 if speed down is necessary, 0 otherwise |
| 1214 | */ |
| 1215 | static int ata_eh_speed_down_needed(struct ata_device *dev) |
| 1216 | { |
| 1217 | const u64 interval = 15LLU * 60 * HZ; |
| 1218 | static const int err_limits[3] = { -1, 3, 10 }; |
| 1219 | struct speed_down_needed_arg arg; |
| 1220 | struct ata_ering_entry *ent; |
| 1221 | int err_cat; |
| 1222 | u64 j64; |
| 1223 | |
| 1224 | ent = ata_ering_top(&dev->ering); |
| 1225 | if (!ent) |
| 1226 | return 0; |
| 1227 | |
| 1228 | err_cat = ata_eh_categorize_ering_entry(ent); |
| 1229 | if (err_cat == 0) |
| 1230 | return 0; |
| 1231 | |
| 1232 | memset(&arg, 0, sizeof(arg)); |
| 1233 | |
| 1234 | j64 = get_jiffies_64(); |
| 1235 | if (j64 >= interval) |
| 1236 | arg.since = j64 - interval; |
| 1237 | else |
| 1238 | arg.since = 0; |
| 1239 | |
| 1240 | ata_ering_map(&dev->ering, speed_down_needed_cb, &arg); |
| 1241 | |
| 1242 | return arg.nr_errors[err_cat] > err_limits[err_cat]; |
| 1243 | } |
| 1244 | |
| 1245 | /** |
| 1246 | * ata_eh_speed_down - record error and speed down if necessary |
| 1247 | * @dev: Failed device |
| 1248 | * @is_io: Did the device fail during normal IO? |
| 1249 | * @err_mask: err_mask of the error |
| 1250 | * |
| 1251 | * Record error and examine error history to determine whether |
| 1252 | * adjusting transmission speed is necessary. It also sets |
| 1253 | * transmission limits appropriately if such adjustment is |
| 1254 | * necessary. |
| 1255 | * |
| 1256 | * LOCKING: |
| 1257 | * Kernel thread context (may sleep). |
| 1258 | * |
| 1259 | * RETURNS: |
| 1260 | * 0 on success, -errno otherwise |
| 1261 | */ |
| 1262 | static int ata_eh_speed_down(struct ata_device *dev, int is_io, |
| 1263 | unsigned int err_mask) |
| 1264 | { |
| 1265 | if (!err_mask) |
| 1266 | return 0; |
| 1267 | |
| 1268 | /* record error and determine whether speed down is necessary */ |
| 1269 | ata_ering_record(&dev->ering, is_io, err_mask); |
| 1270 | |
| 1271 | if (!ata_eh_speed_down_needed(dev)) |
| 1272 | return 0; |
| 1273 | |
| 1274 | /* speed down SATA link speed if possible */ |
| 1275 | if (sata_down_spd_limit(dev->ap) == 0) |
| 1276 | return ATA_EH_HARDRESET; |
| 1277 | |
| 1278 | /* lower transfer mode */ |
| 1279 | if (ata_down_xfermask_limit(dev, 0) == 0) |
| 1280 | return ATA_EH_SOFTRESET; |
| 1281 | |
| 1282 | ata_dev_printk(dev, KERN_ERR, |
| 1283 | "speed down requested but no transfer mode left\n"); |
| 1284 | return 0; |
| 1285 | } |
| 1286 | |
| 1287 | /** |
| 1288 | * ata_eh_autopsy - analyze error and determine recovery action |
| 1289 | * @ap: ATA port to perform autopsy on |
| 1290 | * |
| 1291 | * Analyze why @ap failed and determine which recovery action is |
| 1292 | * needed. This function also sets more detailed AC_ERR_* values |
| 1293 | * and fills sense data for ATAPI CHECK SENSE. |
| 1294 | * |
| 1295 | * LOCKING: |
| 1296 | * Kernel thread context (may sleep). |
| 1297 | */ |
| 1298 | static void ata_eh_autopsy(struct ata_port *ap) |
| 1299 | { |
| 1300 | struct ata_eh_context *ehc = &ap->eh_context; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1301 | unsigned int all_err_mask = 0; |
| 1302 | int tag, is_io = 0; |
| 1303 | u32 serror; |
| 1304 | int rc; |
| 1305 | |
| 1306 | DPRINTK("ENTER\n"); |
| 1307 | |
Tejun Heo | 1cdaf53 | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 1308 | if (ehc->i.flags & ATA_EHI_NO_AUTOPSY) |
| 1309 | return; |
| 1310 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1311 | /* obtain and analyze SError */ |
| 1312 | rc = sata_scr_read(ap, SCR_ERROR, &serror); |
| 1313 | if (rc == 0) { |
| 1314 | ehc->i.serror |= serror; |
| 1315 | ata_eh_analyze_serror(ap); |
| 1316 | } else if (rc != -EOPNOTSUPP) |
Tejun Heo | 4528e4d | 2006-07-08 20:17:26 +0900 | [diff] [blame] | 1317 | ehc->i.action |= ATA_EH_HARDRESET; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1318 | |
Tejun Heo | e8ee845 | 2006-05-15 21:03:46 +0900 | [diff] [blame] | 1319 | /* analyze NCQ failure */ |
| 1320 | ata_eh_analyze_ncq_error(ap); |
| 1321 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1322 | /* any real error trumps AC_ERR_OTHER */ |
| 1323 | if (ehc->i.err_mask & ~AC_ERR_OTHER) |
| 1324 | ehc->i.err_mask &= ~AC_ERR_OTHER; |
| 1325 | |
| 1326 | all_err_mask |= ehc->i.err_mask; |
| 1327 | |
| 1328 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 1329 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); |
| 1330 | |
| 1331 | if (!(qc->flags & ATA_QCFLAG_FAILED)) |
| 1332 | continue; |
| 1333 | |
| 1334 | /* inherit upper level err_mask */ |
| 1335 | qc->err_mask |= ehc->i.err_mask; |
| 1336 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1337 | /* analyze TF */ |
Tejun Heo | 4528e4d | 2006-07-08 20:17:26 +0900 | [diff] [blame] | 1338 | ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1339 | |
| 1340 | /* DEV errors are probably spurious in case of ATA_BUS error */ |
| 1341 | if (qc->err_mask & AC_ERR_ATA_BUS) |
| 1342 | qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA | |
| 1343 | AC_ERR_INVALID); |
| 1344 | |
| 1345 | /* any real error trumps unknown error */ |
| 1346 | if (qc->err_mask & ~AC_ERR_OTHER) |
| 1347 | qc->err_mask &= ~AC_ERR_OTHER; |
| 1348 | |
| 1349 | /* SENSE_VALID trumps dev/unknown error and revalidation */ |
| 1350 | if (qc->flags & ATA_QCFLAG_SENSE_VALID) { |
| 1351 | qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER); |
Tejun Heo | 4528e4d | 2006-07-08 20:17:26 +0900 | [diff] [blame] | 1352 | ehc->i.action &= ~ATA_EH_REVALIDATE; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1353 | } |
| 1354 | |
| 1355 | /* accumulate error info */ |
Tejun Heo | 4528e4d | 2006-07-08 20:17:26 +0900 | [diff] [blame] | 1356 | ehc->i.dev = qc->dev; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1357 | all_err_mask |= qc->err_mask; |
| 1358 | if (qc->flags & ATA_QCFLAG_IO) |
| 1359 | is_io = 1; |
| 1360 | } |
| 1361 | |
Tejun Heo | a20f33f | 2006-05-16 12:58:24 +0900 | [diff] [blame] | 1362 | /* enforce default EH actions */ |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 1363 | if (ap->pflags & ATA_PFLAG_FROZEN || |
Tejun Heo | a20f33f | 2006-05-16 12:58:24 +0900 | [diff] [blame] | 1364 | all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT)) |
Tejun Heo | 4528e4d | 2006-07-08 20:17:26 +0900 | [diff] [blame] | 1365 | ehc->i.action |= ATA_EH_SOFTRESET; |
Tejun Heo | a20f33f | 2006-05-16 12:58:24 +0900 | [diff] [blame] | 1366 | else if (all_err_mask) |
Tejun Heo | 4528e4d | 2006-07-08 20:17:26 +0900 | [diff] [blame] | 1367 | ehc->i.action |= ATA_EH_REVALIDATE; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1368 | |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 1369 | /* if we have offending qcs and the associated failed device */ |
Tejun Heo | 4528e4d | 2006-07-08 20:17:26 +0900 | [diff] [blame] | 1370 | if (ehc->i.dev) { |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 1371 | /* speed down */ |
Tejun Heo | 4528e4d | 2006-07-08 20:17:26 +0900 | [diff] [blame] | 1372 | ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io, |
| 1373 | all_err_mask); |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 1374 | |
| 1375 | /* perform per-dev EH action only on the offending device */ |
Tejun Heo | 4528e4d | 2006-07-08 20:17:26 +0900 | [diff] [blame] | 1376 | ehc->i.dev_action[ehc->i.dev->devno] |= |
| 1377 | ehc->i.action & ATA_EH_PERDEV_MASK; |
| 1378 | ehc->i.action &= ~ATA_EH_PERDEV_MASK; |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 1379 | } |
| 1380 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1381 | DPRINTK("EXIT\n"); |
| 1382 | } |
| 1383 | |
| 1384 | /** |
| 1385 | * ata_eh_report - report error handling to user |
| 1386 | * @ap: ATA port EH is going on |
| 1387 | * |
| 1388 | * Report EH to user. |
| 1389 | * |
| 1390 | * LOCKING: |
| 1391 | * None. |
| 1392 | */ |
| 1393 | static void ata_eh_report(struct ata_port *ap) |
| 1394 | { |
| 1395 | struct ata_eh_context *ehc = &ap->eh_context; |
| 1396 | const char *frozen, *desc; |
| 1397 | int tag, nr_failed = 0; |
| 1398 | |
| 1399 | desc = NULL; |
| 1400 | if (ehc->i.desc[0] != '\0') |
| 1401 | desc = ehc->i.desc; |
| 1402 | |
| 1403 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 1404 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); |
| 1405 | |
| 1406 | if (!(qc->flags & ATA_QCFLAG_FAILED)) |
| 1407 | continue; |
| 1408 | if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask) |
| 1409 | continue; |
| 1410 | |
| 1411 | nr_failed++; |
| 1412 | } |
| 1413 | |
| 1414 | if (!nr_failed && !ehc->i.err_mask) |
| 1415 | return; |
| 1416 | |
| 1417 | frozen = ""; |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 1418 | if (ap->pflags & ATA_PFLAG_FROZEN) |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1419 | frozen = " frozen"; |
| 1420 | |
| 1421 | if (ehc->i.dev) { |
Tejun Heo | e8ee845 | 2006-05-15 21:03:46 +0900 | [diff] [blame] | 1422 | ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x " |
| 1423 | "SAct 0x%x SErr 0x%x action 0x%x%s\n", |
| 1424 | ehc->i.err_mask, ap->sactive, ehc->i.serror, |
| 1425 | ehc->i.action, frozen); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1426 | if (desc) |
| 1427 | ata_dev_printk(ehc->i.dev, KERN_ERR, "(%s)\n", desc); |
| 1428 | } else { |
Tejun Heo | e8ee845 | 2006-05-15 21:03:46 +0900 | [diff] [blame] | 1429 | ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x " |
| 1430 | "SAct 0x%x SErr 0x%x action 0x%x%s\n", |
| 1431 | ehc->i.err_mask, ap->sactive, ehc->i.serror, |
| 1432 | ehc->i.action, frozen); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1433 | if (desc) |
| 1434 | ata_port_printk(ap, KERN_ERR, "(%s)\n", desc); |
| 1435 | } |
| 1436 | |
| 1437 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
Tejun Heo | 8a93758 | 2006-11-14 22:36:12 +0900 | [diff] [blame] | 1438 | static const char *dma_str[] = { |
| 1439 | [DMA_BIDIRECTIONAL] = "bidi", |
| 1440 | [DMA_TO_DEVICE] = "out", |
| 1441 | [DMA_FROM_DEVICE] = "in", |
| 1442 | [DMA_NONE] = "", |
| 1443 | }; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1444 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); |
Tejun Heo | 8a93758 | 2006-11-14 22:36:12 +0900 | [diff] [blame] | 1445 | struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf; |
| 1446 | unsigned int nbytes; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1447 | |
| 1448 | if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask) |
| 1449 | continue; |
| 1450 | |
Tejun Heo | 8a93758 | 2006-11-14 22:36:12 +0900 | [diff] [blame] | 1451 | nbytes = qc->nbytes; |
| 1452 | if (!nbytes) |
| 1453 | nbytes = qc->nsect << 9; |
| 1454 | |
| 1455 | ata_dev_printk(qc->dev, KERN_ERR, |
| 1456 | "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x " |
Tejun Heo | 664e850 | 2006-11-20 16:05:34 +0900 | [diff] [blame] | 1457 | "tag %d cdb 0x%x data %u %s\n " |
Tejun Heo | 8a93758 | 2006-11-14 22:36:12 +0900 | [diff] [blame] | 1458 | "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x " |
| 1459 | "Emask 0x%x (%s)\n", |
| 1460 | cmd->command, cmd->feature, cmd->nsect, |
| 1461 | cmd->lbal, cmd->lbam, cmd->lbah, |
| 1462 | cmd->hob_feature, cmd->hob_nsect, |
| 1463 | cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah, |
Tejun Heo | 664e850 | 2006-11-20 16:05:34 +0900 | [diff] [blame] | 1464 | cmd->device, qc->tag, qc->cdb[0], nbytes, |
| 1465 | dma_str[qc->dma_dir], |
Tejun Heo | 8a93758 | 2006-11-14 22:36:12 +0900 | [diff] [blame] | 1466 | res->command, res->feature, res->nsect, |
| 1467 | res->lbal, res->lbam, res->lbah, |
| 1468 | res->hob_feature, res->hob_nsect, |
| 1469 | res->hob_lbal, res->hob_lbam, res->hob_lbah, |
| 1470 | res->device, qc->err_mask, ata_err_string(qc->err_mask)); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1471 | } |
| 1472 | } |
| 1473 | |
Tejun Heo | d87fa38 | 2006-05-31 18:28:24 +0900 | [diff] [blame] | 1474 | static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset, |
| 1475 | unsigned int *classes) |
| 1476 | { |
| 1477 | int i, rc; |
| 1478 | |
| 1479 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 1480 | classes[i] = ATA_DEV_UNKNOWN; |
| 1481 | |
| 1482 | rc = reset(ap, classes); |
| 1483 | if (rc) |
| 1484 | return rc; |
| 1485 | |
| 1486 | /* If any class isn't ATA_DEV_UNKNOWN, consider classification |
| 1487 | * is complete and convert all ATA_DEV_UNKNOWN to |
| 1488 | * ATA_DEV_NONE. |
| 1489 | */ |
| 1490 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 1491 | if (classes[i] != ATA_DEV_UNKNOWN) |
| 1492 | break; |
| 1493 | |
| 1494 | if (i < ATA_MAX_DEVICES) |
| 1495 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 1496 | if (classes[i] == ATA_DEV_UNKNOWN) |
| 1497 | classes[i] = ATA_DEV_NONE; |
| 1498 | |
| 1499 | return 0; |
| 1500 | } |
| 1501 | |
Tejun Heo | 664faf0 | 2006-05-31 18:27:50 +0900 | [diff] [blame] | 1502 | static int ata_eh_followup_srst_needed(int rc, int classify, |
| 1503 | const unsigned int *classes) |
| 1504 | { |
| 1505 | if (rc == -EAGAIN) |
| 1506 | return 1; |
| 1507 | if (rc != 0) |
| 1508 | return 0; |
| 1509 | if (classify && classes[0] == ATA_DEV_UNKNOWN) |
| 1510 | return 1; |
| 1511 | return 0; |
| 1512 | } |
| 1513 | |
| 1514 | static int ata_eh_reset(struct ata_port *ap, int classify, |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 1515 | ata_prereset_fn_t prereset, ata_reset_fn_t softreset, |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1516 | ata_reset_fn_t hardreset, ata_postreset_fn_t postreset) |
| 1517 | { |
| 1518 | struct ata_eh_context *ehc = &ap->eh_context; |
Tejun Heo | 664faf0 | 2006-05-31 18:27:50 +0900 | [diff] [blame] | 1519 | unsigned int *classes = ehc->classes; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1520 | int tries = ATA_EH_RESET_TRIES; |
Tejun Heo | 1cdaf53 | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 1521 | int verbose = !(ehc->i.flags & ATA_EHI_QUIET); |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 1522 | unsigned int action; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1523 | ata_reset_fn_t reset; |
Tejun Heo | 664faf0 | 2006-05-31 18:27:50 +0900 | [diff] [blame] | 1524 | int i, did_followup_srst, rc; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1525 | |
Tejun Heo | 13abf50 | 2006-07-10 23:18:46 +0900 | [diff] [blame] | 1526 | /* about to reset */ |
| 1527 | ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK); |
| 1528 | |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 1529 | /* Determine which reset to use and record in ehc->i.action. |
| 1530 | * prereset() may examine and modify it. |
| 1531 | */ |
| 1532 | action = ehc->i.action; |
| 1533 | ehc->i.action &= ~ATA_EH_RESET_MASK; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1534 | if (softreset && (!hardreset || (!sata_set_spd_needed(ap) && |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 1535 | !(action & ATA_EH_HARDRESET)))) |
| 1536 | ehc->i.action |= ATA_EH_SOFTRESET; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1537 | else |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 1538 | ehc->i.action |= ATA_EH_HARDRESET; |
| 1539 | |
| 1540 | if (prereset) { |
| 1541 | rc = prereset(ap); |
| 1542 | if (rc) { |
Alan Cox | c961922 | 2006-09-26 17:53:38 +0100 | [diff] [blame] | 1543 | if (rc == -ENOENT) { |
| 1544 | ata_port_printk(ap, KERN_DEBUG, "port disabled. ignoring.\n"); |
| 1545 | ap->eh_context.i.action &= ~ATA_EH_RESET_MASK; |
| 1546 | } else |
| 1547 | ata_port_printk(ap, KERN_ERR, |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 1548 | "prereset failed (errno=%d)\n", rc); |
| 1549 | return rc; |
| 1550 | } |
| 1551 | } |
| 1552 | |
| 1553 | /* prereset() might have modified ehc->i.action */ |
| 1554 | if (ehc->i.action & ATA_EH_HARDRESET) |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1555 | reset = hardreset; |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 1556 | else if (ehc->i.action & ATA_EH_SOFTRESET) |
| 1557 | reset = softreset; |
| 1558 | else { |
| 1559 | /* prereset told us not to reset, bang classes and return */ |
| 1560 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 1561 | classes[i] = ATA_DEV_NONE; |
| 1562 | return 0; |
| 1563 | } |
| 1564 | |
| 1565 | /* did prereset() screw up? if so, fix up to avoid oopsing */ |
| 1566 | if (!reset) { |
| 1567 | ata_port_printk(ap, KERN_ERR, "BUG: prereset() requested " |
| 1568 | "invalid reset type\n"); |
| 1569 | if (softreset) |
| 1570 | reset = softreset; |
| 1571 | else |
| 1572 | reset = hardreset; |
| 1573 | } |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1574 | |
| 1575 | retry: |
Tejun Heo | 3e706399 | 2006-05-31 18:28:11 +0900 | [diff] [blame] | 1576 | /* shut up during boot probing */ |
| 1577 | if (verbose) |
| 1578 | ata_port_printk(ap, KERN_INFO, "%s resetting port\n", |
| 1579 | reset == softreset ? "soft" : "hard"); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1580 | |
Tejun Heo | 13abf50 | 2006-07-10 23:18:46 +0900 | [diff] [blame] | 1581 | /* mark that this EH session started with reset */ |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1582 | ehc->i.flags |= ATA_EHI_DID_RESET; |
| 1583 | |
| 1584 | rc = ata_do_reset(ap, reset, classes); |
| 1585 | |
Tejun Heo | 664faf0 | 2006-05-31 18:27:50 +0900 | [diff] [blame] | 1586 | did_followup_srst = 0; |
| 1587 | if (reset == hardreset && |
| 1588 | ata_eh_followup_srst_needed(rc, classify, classes)) { |
| 1589 | /* okay, let's do follow-up softreset */ |
| 1590 | did_followup_srst = 1; |
| 1591 | reset = softreset; |
| 1592 | |
| 1593 | if (!reset) { |
| 1594 | ata_port_printk(ap, KERN_ERR, |
| 1595 | "follow-up softreset required " |
| 1596 | "but no softreset avaliable\n"); |
| 1597 | return -EINVAL; |
| 1598 | } |
| 1599 | |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 1600 | ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK); |
Tejun Heo | 664faf0 | 2006-05-31 18:27:50 +0900 | [diff] [blame] | 1601 | rc = ata_do_reset(ap, reset, classes); |
| 1602 | |
| 1603 | if (rc == 0 && classify && |
| 1604 | classes[0] == ATA_DEV_UNKNOWN) { |
| 1605 | ata_port_printk(ap, KERN_ERR, |
| 1606 | "classification failed\n"); |
| 1607 | return -EINVAL; |
| 1608 | } |
| 1609 | } |
| 1610 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1611 | if (rc && --tries) { |
Tejun Heo | 664faf0 | 2006-05-31 18:27:50 +0900 | [diff] [blame] | 1612 | const char *type; |
| 1613 | |
| 1614 | if (reset == softreset) { |
| 1615 | if (did_followup_srst) |
| 1616 | type = "follow-up soft"; |
| 1617 | else |
| 1618 | type = "soft"; |
| 1619 | } else |
| 1620 | type = "hard"; |
| 1621 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1622 | ata_port_printk(ap, KERN_WARNING, |
Tejun Heo | 664faf0 | 2006-05-31 18:27:50 +0900 | [diff] [blame] | 1623 | "%sreset failed, retrying in 5 secs\n", type); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1624 | ssleep(5); |
| 1625 | |
| 1626 | if (reset == hardreset) |
| 1627 | sata_down_spd_limit(ap); |
| 1628 | if (hardreset) |
| 1629 | reset = hardreset; |
| 1630 | goto retry; |
| 1631 | } |
| 1632 | |
| 1633 | if (rc == 0) { |
Tejun Heo | 20952b6 | 2006-05-31 18:27:23 +0900 | [diff] [blame] | 1634 | /* After the reset, the device state is PIO 0 and the |
| 1635 | * controller state is undefined. Record the mode. |
| 1636 | */ |
| 1637 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 1638 | ap->device[i].pio_mode = XFER_PIO_0; |
| 1639 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1640 | if (postreset) |
| 1641 | postreset(ap, classes); |
| 1642 | |
| 1643 | /* reset successful, schedule revalidation */ |
Tejun Heo | 13abf50 | 2006-07-10 23:18:46 +0900 | [diff] [blame] | 1644 | ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1645 | ehc->i.action |= ATA_EH_REVALIDATE; |
| 1646 | } |
| 1647 | |
| 1648 | return rc; |
| 1649 | } |
| 1650 | |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 1651 | static int ata_eh_revalidate_and_attach(struct ata_port *ap, |
| 1652 | struct ata_device **r_failed_dev) |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1653 | { |
| 1654 | struct ata_eh_context *ehc = &ap->eh_context; |
| 1655 | struct ata_device *dev; |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 1656 | unsigned long flags; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1657 | int i, rc = 0; |
| 1658 | |
| 1659 | DPRINTK("ENTER\n"); |
| 1660 | |
| 1661 | for (i = 0; i < ATA_MAX_DEVICES; i++) { |
Tejun Heo | bff0464 | 2006-11-10 18:08:10 +0900 | [diff] [blame] | 1662 | unsigned int action, readid_flags = 0; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1663 | |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 1664 | dev = &ap->device[i]; |
Tejun Heo | 64f65ca | 2006-06-24 20:30:18 +0900 | [diff] [blame] | 1665 | action = ata_eh_dev_action(dev); |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 1666 | |
Tejun Heo | bff0464 | 2006-11-10 18:08:10 +0900 | [diff] [blame] | 1667 | if (ehc->i.flags & ATA_EHI_DID_RESET) |
| 1668 | readid_flags |= ATA_READID_POSTRESET; |
| 1669 | |
Tejun Heo | 02670bf | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 1670 | if (action & ATA_EH_REVALIDATE && ata_dev_ready(dev)) { |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1671 | if (ata_port_offline(ap)) { |
| 1672 | rc = -EIO; |
| 1673 | break; |
| 1674 | } |
| 1675 | |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 1676 | ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE); |
Tejun Heo | bff0464 | 2006-11-10 18:08:10 +0900 | [diff] [blame] | 1677 | rc = ata_dev_revalidate(dev, readid_flags); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1678 | if (rc) |
| 1679 | break; |
| 1680 | |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 1681 | ata_eh_done(ap, dev, ATA_EH_REVALIDATE); |
| 1682 | |
Tejun Heo | baa1e78 | 2006-11-01 18:39:27 +0900 | [diff] [blame] | 1683 | /* Configuration may have changed, reconfigure |
| 1684 | * transfer mode. |
| 1685 | */ |
| 1686 | ehc->i.flags |= ATA_EHI_SETMODE; |
| 1687 | |
zhao, forrest | 3057ac3 | 2006-06-12 12:01:34 +0800 | [diff] [blame] | 1688 | /* schedule the scsi_rescan_device() here */ |
| 1689 | queue_work(ata_aux_wq, &(ap->scsi_rescan_task)); |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 1690 | } else if (dev->class == ATA_DEV_UNKNOWN && |
| 1691 | ehc->tries[dev->devno] && |
| 1692 | ata_class_enabled(ehc->classes[dev->devno])) { |
| 1693 | dev->class = ehc->classes[dev->devno]; |
| 1694 | |
Tejun Heo | bff0464 | 2006-11-10 18:08:10 +0900 | [diff] [blame] | 1695 | rc = ata_dev_read_id(dev, &dev->class, readid_flags, |
| 1696 | dev->id); |
Tejun Heo | efdaedc | 2006-11-01 18:38:52 +0900 | [diff] [blame] | 1697 | if (rc == 0) { |
| 1698 | ehc->i.flags |= ATA_EHI_PRINTINFO; |
| 1699 | rc = ata_dev_configure(dev); |
| 1700 | ehc->i.flags &= ~ATA_EHI_PRINTINFO; |
Tejun Heo | 55a8e2c | 2006-11-10 18:08:10 +0900 | [diff] [blame] | 1701 | } else if (rc == -ENOENT) { |
| 1702 | /* IDENTIFY was issued to non-existent |
| 1703 | * device. No need to reset. Just |
| 1704 | * thaw and kill the device. |
| 1705 | */ |
| 1706 | ata_eh_thaw_port(ap); |
| 1707 | dev->class = ATA_DEV_UNKNOWN; |
| 1708 | rc = 0; |
Tejun Heo | efdaedc | 2006-11-01 18:38:52 +0900 | [diff] [blame] | 1709 | } |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 1710 | |
| 1711 | if (rc) { |
| 1712 | dev->class = ATA_DEV_UNKNOWN; |
| 1713 | break; |
| 1714 | } |
| 1715 | |
Tejun Heo | 55a8e2c | 2006-11-10 18:08:10 +0900 | [diff] [blame] | 1716 | if (ata_dev_enabled(dev)) { |
| 1717 | spin_lock_irqsave(ap->lock, flags); |
| 1718 | ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG; |
| 1719 | spin_unlock_irqrestore(ap->lock, flags); |
Tejun Heo | baa1e78 | 2006-11-01 18:39:27 +0900 | [diff] [blame] | 1720 | |
Tejun Heo | 55a8e2c | 2006-11-10 18:08:10 +0900 | [diff] [blame] | 1721 | /* new device discovered, configure xfermode */ |
| 1722 | ehc->i.flags |= ATA_EHI_SETMODE; |
| 1723 | } |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1724 | } |
| 1725 | } |
| 1726 | |
Tejun Heo | 47005f2 | 2006-06-19 18:27:23 +0900 | [diff] [blame] | 1727 | if (rc) |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1728 | *r_failed_dev = dev; |
| 1729 | |
| 1730 | DPRINTK("EXIT\n"); |
| 1731 | return rc; |
| 1732 | } |
| 1733 | |
Tejun Heo | 02670bf | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 1734 | /** |
| 1735 | * ata_eh_suspend - handle suspend EH action |
| 1736 | * @ap: target host port |
| 1737 | * @r_failed_dev: result parameter to indicate failing device |
| 1738 | * |
| 1739 | * Handle suspend EH action. Disk devices are spinned down and |
| 1740 | * other types of devices are just marked suspended. Once |
| 1741 | * suspended, no EH action to the device is allowed until it is |
| 1742 | * resumed. |
| 1743 | * |
| 1744 | * LOCKING: |
| 1745 | * Kernel thread context (may sleep). |
| 1746 | * |
| 1747 | * RETURNS: |
| 1748 | * 0 on success, -errno otherwise |
| 1749 | */ |
| 1750 | static int ata_eh_suspend(struct ata_port *ap, struct ata_device **r_failed_dev) |
| 1751 | { |
| 1752 | struct ata_device *dev; |
| 1753 | int i, rc = 0; |
| 1754 | |
| 1755 | DPRINTK("ENTER\n"); |
| 1756 | |
| 1757 | for (i = 0; i < ATA_MAX_DEVICES; i++) { |
| 1758 | unsigned long flags; |
| 1759 | unsigned int action, err_mask; |
| 1760 | |
| 1761 | dev = &ap->device[i]; |
| 1762 | action = ata_eh_dev_action(dev); |
| 1763 | |
| 1764 | if (!ata_dev_enabled(dev) || !(action & ATA_EH_SUSPEND)) |
| 1765 | continue; |
| 1766 | |
| 1767 | WARN_ON(dev->flags & ATA_DFLAG_SUSPENDED); |
| 1768 | |
| 1769 | ata_eh_about_to_do(ap, dev, ATA_EH_SUSPEND); |
| 1770 | |
| 1771 | if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) { |
| 1772 | /* flush cache */ |
| 1773 | rc = ata_flush_cache(dev); |
| 1774 | if (rc) |
| 1775 | break; |
| 1776 | |
| 1777 | /* spin down */ |
| 1778 | err_mask = ata_do_simple_cmd(dev, ATA_CMD_STANDBYNOW1); |
| 1779 | if (err_mask) { |
| 1780 | ata_dev_printk(dev, KERN_ERR, "failed to " |
| 1781 | "spin down (err_mask=0x%x)\n", |
| 1782 | err_mask); |
| 1783 | rc = -EIO; |
| 1784 | break; |
| 1785 | } |
| 1786 | } |
| 1787 | |
| 1788 | spin_lock_irqsave(ap->lock, flags); |
| 1789 | dev->flags |= ATA_DFLAG_SUSPENDED; |
| 1790 | spin_unlock_irqrestore(ap->lock, flags); |
| 1791 | |
| 1792 | ata_eh_done(ap, dev, ATA_EH_SUSPEND); |
| 1793 | } |
| 1794 | |
| 1795 | if (rc) |
| 1796 | *r_failed_dev = dev; |
| 1797 | |
| 1798 | DPRINTK("EXIT\n"); |
Tejun Heo | 03ee5b1 | 2007-01-26 20:10:25 +0900 | [diff] [blame] | 1799 | return rc; |
Tejun Heo | 02670bf | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 1800 | } |
| 1801 | |
| 1802 | /** |
| 1803 | * ata_eh_prep_resume - prep for resume EH action |
| 1804 | * @ap: target host port |
| 1805 | * |
| 1806 | * Clear SUSPENDED in preparation for scheduled resume actions. |
| 1807 | * This allows other parts of EH to access the devices being |
| 1808 | * resumed. |
| 1809 | * |
| 1810 | * LOCKING: |
| 1811 | * Kernel thread context (may sleep). |
| 1812 | */ |
| 1813 | static void ata_eh_prep_resume(struct ata_port *ap) |
| 1814 | { |
| 1815 | struct ata_device *dev; |
| 1816 | unsigned long flags; |
| 1817 | int i; |
| 1818 | |
| 1819 | DPRINTK("ENTER\n"); |
| 1820 | |
| 1821 | for (i = 0; i < ATA_MAX_DEVICES; i++) { |
| 1822 | unsigned int action; |
| 1823 | |
| 1824 | dev = &ap->device[i]; |
| 1825 | action = ata_eh_dev_action(dev); |
| 1826 | |
| 1827 | if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME)) |
| 1828 | continue; |
| 1829 | |
| 1830 | spin_lock_irqsave(ap->lock, flags); |
| 1831 | dev->flags &= ~ATA_DFLAG_SUSPENDED; |
| 1832 | spin_unlock_irqrestore(ap->lock, flags); |
| 1833 | } |
| 1834 | |
| 1835 | DPRINTK("EXIT\n"); |
| 1836 | } |
| 1837 | |
| 1838 | /** |
| 1839 | * ata_eh_resume - handle resume EH action |
| 1840 | * @ap: target host port |
| 1841 | * @r_failed_dev: result parameter to indicate failing device |
| 1842 | * |
| 1843 | * Handle resume EH action. Target devices are already reset and |
| 1844 | * revalidated. Spinning up is the only operation left. |
| 1845 | * |
| 1846 | * LOCKING: |
| 1847 | * Kernel thread context (may sleep). |
| 1848 | * |
| 1849 | * RETURNS: |
| 1850 | * 0 on success, -errno otherwise |
| 1851 | */ |
| 1852 | static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev) |
| 1853 | { |
| 1854 | struct ata_device *dev; |
| 1855 | int i, rc = 0; |
| 1856 | |
| 1857 | DPRINTK("ENTER\n"); |
| 1858 | |
| 1859 | for (i = 0; i < ATA_MAX_DEVICES; i++) { |
| 1860 | unsigned int action, err_mask; |
| 1861 | |
| 1862 | dev = &ap->device[i]; |
| 1863 | action = ata_eh_dev_action(dev); |
| 1864 | |
| 1865 | if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME)) |
| 1866 | continue; |
| 1867 | |
| 1868 | ata_eh_about_to_do(ap, dev, ATA_EH_RESUME); |
| 1869 | |
| 1870 | if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) { |
| 1871 | err_mask = ata_do_simple_cmd(dev, |
| 1872 | ATA_CMD_IDLEIMMEDIATE); |
| 1873 | if (err_mask) { |
| 1874 | ata_dev_printk(dev, KERN_ERR, "failed to " |
| 1875 | "spin up (err_mask=0x%x)\n", |
| 1876 | err_mask); |
| 1877 | rc = -EIO; |
| 1878 | break; |
| 1879 | } |
| 1880 | } |
| 1881 | |
| 1882 | ata_eh_done(ap, dev, ATA_EH_RESUME); |
| 1883 | } |
| 1884 | |
| 1885 | if (rc) |
| 1886 | *r_failed_dev = dev; |
| 1887 | |
| 1888 | DPRINTK("EXIT\n"); |
| 1889 | return 0; |
| 1890 | } |
| 1891 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1892 | static int ata_port_nr_enabled(struct ata_port *ap) |
| 1893 | { |
| 1894 | int i, cnt = 0; |
| 1895 | |
| 1896 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 1897 | if (ata_dev_enabled(&ap->device[i])) |
| 1898 | cnt++; |
| 1899 | return cnt; |
| 1900 | } |
| 1901 | |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 1902 | static int ata_port_nr_vacant(struct ata_port *ap) |
| 1903 | { |
| 1904 | int i, cnt = 0; |
| 1905 | |
| 1906 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 1907 | if (ap->device[i].class == ATA_DEV_UNKNOWN) |
| 1908 | cnt++; |
| 1909 | return cnt; |
| 1910 | } |
| 1911 | |
| 1912 | static int ata_eh_skip_recovery(struct ata_port *ap) |
| 1913 | { |
| 1914 | struct ata_eh_context *ehc = &ap->eh_context; |
| 1915 | int i; |
| 1916 | |
Tejun Heo | 02670bf | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 1917 | /* skip if all possible devices are suspended */ |
| 1918 | for (i = 0; i < ata_port_max_devices(ap); i++) { |
| 1919 | struct ata_device *dev = &ap->device[i]; |
| 1920 | |
Tejun Heo | 7c8c2cf | 2006-07-10 23:18:23 +0900 | [diff] [blame] | 1921 | if (!(dev->flags & ATA_DFLAG_SUSPENDED)) |
Tejun Heo | 02670bf | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 1922 | break; |
| 1923 | } |
| 1924 | |
| 1925 | if (i == ata_port_max_devices(ap)) |
| 1926 | return 1; |
| 1927 | |
Tejun Heo | 7c8c2cf | 2006-07-10 23:18:23 +0900 | [diff] [blame] | 1928 | /* thaw frozen port, resume link and recover failed devices */ |
| 1929 | if ((ap->pflags & ATA_PFLAG_FROZEN) || |
| 1930 | (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap)) |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 1931 | return 0; |
| 1932 | |
| 1933 | /* skip if class codes for all vacant slots are ATA_DEV_NONE */ |
| 1934 | for (i = 0; i < ATA_MAX_DEVICES; i++) { |
| 1935 | struct ata_device *dev = &ap->device[i]; |
| 1936 | |
| 1937 | if (dev->class == ATA_DEV_UNKNOWN && |
| 1938 | ehc->classes[dev->devno] != ATA_DEV_NONE) |
| 1939 | return 0; |
| 1940 | } |
| 1941 | |
| 1942 | return 1; |
| 1943 | } |
| 1944 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1945 | /** |
| 1946 | * ata_eh_recover - recover host port after error |
| 1947 | * @ap: host port to recover |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 1948 | * @prereset: prereset method (can be NULL) |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1949 | * @softreset: softreset method (can be NULL) |
| 1950 | * @hardreset: hardreset method (can be NULL) |
| 1951 | * @postreset: postreset method (can be NULL) |
| 1952 | * |
| 1953 | * This is the alpha and omega, eum and yang, heart and soul of |
| 1954 | * libata exception handling. On entry, actions required to |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 1955 | * recover the port and hotplug requests are recorded in |
| 1956 | * eh_context. This function executes all the operations with |
| 1957 | * appropriate retrials and fallbacks to resurrect failed |
| 1958 | * devices, detach goners and greet newcomers. |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1959 | * |
| 1960 | * LOCKING: |
| 1961 | * Kernel thread context (may sleep). |
| 1962 | * |
| 1963 | * RETURNS: |
| 1964 | * 0 on success, -errno on failure. |
| 1965 | */ |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 1966 | static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset, |
| 1967 | ata_reset_fn_t softreset, ata_reset_fn_t hardreset, |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1968 | ata_postreset_fn_t postreset) |
| 1969 | { |
| 1970 | struct ata_eh_context *ehc = &ap->eh_context; |
| 1971 | struct ata_device *dev; |
| 1972 | int down_xfermask, i, rc; |
| 1973 | |
| 1974 | DPRINTK("ENTER\n"); |
| 1975 | |
| 1976 | /* prep for recovery */ |
| 1977 | for (i = 0; i < ATA_MAX_DEVICES; i++) { |
| 1978 | dev = &ap->device[i]; |
| 1979 | |
| 1980 | ehc->tries[dev->devno] = ATA_EH_DEV_TRIES; |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 1981 | |
Tejun Heo | 79a55b7 | 2007-01-18 17:22:18 +0900 | [diff] [blame] | 1982 | /* collect port action mask recorded in dev actions */ |
| 1983 | ehc->i.action |= ehc->i.dev_action[i] & ~ATA_EH_PERDEV_MASK; |
| 1984 | ehc->i.dev_action[i] &= ATA_EH_PERDEV_MASK; |
| 1985 | |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 1986 | /* process hotplug request */ |
| 1987 | if (dev->flags & ATA_DFLAG_DETACH) |
| 1988 | ata_eh_detach_dev(dev); |
| 1989 | |
| 1990 | if (!ata_dev_enabled(dev) && |
| 1991 | ((ehc->i.probe_mask & (1 << dev->devno)) && |
| 1992 | !(ehc->did_probe_mask & (1 << dev->devno)))) { |
| 1993 | ata_eh_detach_dev(dev); |
| 1994 | ata_dev_init(dev); |
| 1995 | ehc->did_probe_mask |= (1 << dev->devno); |
| 1996 | ehc->i.action |= ATA_EH_SOFTRESET; |
| 1997 | } |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 1998 | } |
| 1999 | |
| 2000 | retry: |
| 2001 | down_xfermask = 0; |
| 2002 | rc = 0; |
| 2003 | |
Tejun Heo | aeb2ecd | 2006-06-12 14:11:43 +0900 | [diff] [blame] | 2004 | /* if UNLOADING, finish immediately */ |
Tejun Heo | b51e9e5 | 2006-06-29 01:29:30 +0900 | [diff] [blame] | 2005 | if (ap->pflags & ATA_PFLAG_UNLOADING) |
Tejun Heo | aeb2ecd | 2006-06-12 14:11:43 +0900 | [diff] [blame] | 2006 | goto out; |
| 2007 | |
Tejun Heo | 02670bf | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 2008 | /* prep for resume */ |
| 2009 | ata_eh_prep_resume(ap); |
| 2010 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2011 | /* skip EH if possible. */ |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 2012 | if (ata_eh_skip_recovery(ap)) |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2013 | ehc->i.action = 0; |
| 2014 | |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 2015 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 2016 | ehc->classes[i] = ATA_DEV_UNKNOWN; |
| 2017 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2018 | /* reset */ |
| 2019 | if (ehc->i.action & ATA_EH_RESET_MASK) { |
| 2020 | ata_eh_freeze_port(ap); |
| 2021 | |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 2022 | rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset, |
| 2023 | softreset, hardreset, postreset); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2024 | if (rc) { |
| 2025 | ata_port_printk(ap, KERN_ERR, |
| 2026 | "reset failed, giving up\n"); |
| 2027 | goto out; |
| 2028 | } |
| 2029 | |
| 2030 | ata_eh_thaw_port(ap); |
| 2031 | } |
| 2032 | |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 2033 | /* revalidate existing devices and attach new ones */ |
| 2034 | rc = ata_eh_revalidate_and_attach(ap, &dev); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2035 | if (rc) |
| 2036 | goto dev_fail; |
| 2037 | |
Tejun Heo | 02670bf | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 2038 | /* resume devices */ |
| 2039 | rc = ata_eh_resume(ap, &dev); |
| 2040 | if (rc) |
| 2041 | goto dev_fail; |
| 2042 | |
Tejun Heo | baa1e78 | 2006-11-01 18:39:27 +0900 | [diff] [blame] | 2043 | /* configure transfer mode if necessary */ |
| 2044 | if (ehc->i.flags & ATA_EHI_SETMODE) { |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2045 | rc = ata_set_mode(ap, &dev); |
| 2046 | if (rc) { |
| 2047 | down_xfermask = 1; |
| 2048 | goto dev_fail; |
| 2049 | } |
Tejun Heo | baa1e78 | 2006-11-01 18:39:27 +0900 | [diff] [blame] | 2050 | ehc->i.flags &= ~ATA_EHI_SETMODE; |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2051 | } |
| 2052 | |
Tejun Heo | 02670bf | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 2053 | /* suspend devices */ |
| 2054 | rc = ata_eh_suspend(ap, &dev); |
| 2055 | if (rc) |
| 2056 | goto dev_fail; |
| 2057 | |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2058 | goto out; |
| 2059 | |
| 2060 | dev_fail: |
| 2061 | switch (rc) { |
| 2062 | case -ENODEV: |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 2063 | /* device missing, schedule probing */ |
| 2064 | ehc->i.probe_mask |= (1 << dev->devno); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2065 | case -EINVAL: |
| 2066 | ehc->tries[dev->devno] = 0; |
| 2067 | break; |
| 2068 | case -EIO: |
| 2069 | sata_down_spd_limit(ap); |
| 2070 | default: |
| 2071 | ehc->tries[dev->devno]--; |
| 2072 | if (down_xfermask && |
| 2073 | ata_down_xfermask_limit(dev, ehc->tries[dev->devno] == 1)) |
| 2074 | ehc->tries[dev->devno] = 0; |
| 2075 | } |
| 2076 | |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 2077 | if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) { |
| 2078 | /* disable device if it has used up all its chances */ |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2079 | ata_dev_disable(dev); |
| 2080 | |
Tejun Heo | 084fe63 | 2006-05-31 18:28:03 +0900 | [diff] [blame] | 2081 | /* detach if offline */ |
| 2082 | if (ata_port_offline(ap)) |
| 2083 | ata_eh_detach_dev(dev); |
| 2084 | |
| 2085 | /* probe if requested */ |
| 2086 | if ((ehc->i.probe_mask & (1 << dev->devno)) && |
| 2087 | !(ehc->did_probe_mask & (1 << dev->devno))) { |
| 2088 | ata_eh_detach_dev(dev); |
| 2089 | ata_dev_init(dev); |
| 2090 | |
| 2091 | ehc->tries[dev->devno] = ATA_EH_DEV_TRIES; |
| 2092 | ehc->did_probe_mask |= (1 << dev->devno); |
| 2093 | ehc->i.action |= ATA_EH_SOFTRESET; |
| 2094 | } |
| 2095 | } else { |
| 2096 | /* soft didn't work? be haaaaard */ |
| 2097 | if (ehc->i.flags & ATA_EHI_DID_RESET) |
| 2098 | ehc->i.action |= ATA_EH_HARDRESET; |
| 2099 | else |
| 2100 | ehc->i.action |= ATA_EH_SOFTRESET; |
| 2101 | } |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2102 | |
| 2103 | if (ata_port_nr_enabled(ap)) { |
| 2104 | ata_port_printk(ap, KERN_WARNING, "failed to recover some " |
| 2105 | "devices, retrying in 5 secs\n"); |
| 2106 | ssleep(5); |
| 2107 | } else { |
| 2108 | /* no device left, repeat fast */ |
| 2109 | msleep(500); |
| 2110 | } |
| 2111 | |
| 2112 | goto retry; |
| 2113 | |
| 2114 | out: |
| 2115 | if (rc) { |
| 2116 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
| 2117 | ata_dev_disable(&ap->device[i]); |
| 2118 | } |
| 2119 | |
| 2120 | DPRINTK("EXIT, rc=%d\n", rc); |
| 2121 | return rc; |
| 2122 | } |
| 2123 | |
| 2124 | /** |
| 2125 | * ata_eh_finish - finish up EH |
| 2126 | * @ap: host port to finish EH for |
| 2127 | * |
| 2128 | * Recovery is complete. Clean up EH states and retry or finish |
| 2129 | * failed qcs. |
| 2130 | * |
| 2131 | * LOCKING: |
| 2132 | * None. |
| 2133 | */ |
| 2134 | static void ata_eh_finish(struct ata_port *ap) |
| 2135 | { |
| 2136 | int tag; |
| 2137 | |
| 2138 | /* retry or finish qcs */ |
| 2139 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 2140 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); |
| 2141 | |
| 2142 | if (!(qc->flags & ATA_QCFLAG_FAILED)) |
| 2143 | continue; |
| 2144 | |
| 2145 | if (qc->err_mask) { |
| 2146 | /* FIXME: Once EH migration is complete, |
| 2147 | * generate sense data in this function, |
| 2148 | * considering both err_mask and tf. |
| 2149 | */ |
| 2150 | if (qc->err_mask & AC_ERR_INVALID) |
| 2151 | ata_eh_qc_complete(qc); |
| 2152 | else |
| 2153 | ata_eh_qc_retry(qc); |
| 2154 | } else { |
| 2155 | if (qc->flags & ATA_QCFLAG_SENSE_VALID) { |
| 2156 | ata_eh_qc_complete(qc); |
| 2157 | } else { |
| 2158 | /* feed zero TF to sense generation */ |
| 2159 | memset(&qc->result_tf, 0, sizeof(qc->result_tf)); |
| 2160 | ata_eh_qc_retry(qc); |
| 2161 | } |
| 2162 | } |
| 2163 | } |
| 2164 | } |
| 2165 | |
| 2166 | /** |
| 2167 | * ata_do_eh - do standard error handling |
| 2168 | * @ap: host port to handle error for |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 2169 | * @prereset: prereset method (can be NULL) |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2170 | * @softreset: softreset method (can be NULL) |
| 2171 | * @hardreset: hardreset method (can be NULL) |
| 2172 | * @postreset: postreset method (can be NULL) |
| 2173 | * |
| 2174 | * Perform standard error handling sequence. |
| 2175 | * |
| 2176 | * LOCKING: |
| 2177 | * Kernel thread context (may sleep). |
| 2178 | */ |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 2179 | void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset, |
| 2180 | ata_reset_fn_t softreset, ata_reset_fn_t hardreset, |
| 2181 | ata_postreset_fn_t postreset) |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2182 | { |
Tejun Heo | 1cdaf53 | 2006-07-03 16:07:26 +0900 | [diff] [blame] | 2183 | ata_eh_autopsy(ap); |
| 2184 | ata_eh_report(ap); |
Tejun Heo | f5914a4 | 2006-05-31 18:27:48 +0900 | [diff] [blame] | 2185 | ata_eh_recover(ap, prereset, softreset, hardreset, postreset); |
Tejun Heo | 022bdb0 | 2006-05-15 20:58:22 +0900 | [diff] [blame] | 2186 | ata_eh_finish(ap); |
| 2187 | } |
Tejun Heo | 500530f | 2006-07-03 16:07:27 +0900 | [diff] [blame] | 2188 | |
| 2189 | /** |
| 2190 | * ata_eh_handle_port_suspend - perform port suspend operation |
| 2191 | * @ap: port to suspend |
| 2192 | * |
| 2193 | * Suspend @ap. |
| 2194 | * |
| 2195 | * LOCKING: |
| 2196 | * Kernel thread context (may sleep). |
| 2197 | */ |
| 2198 | static void ata_eh_handle_port_suspend(struct ata_port *ap) |
| 2199 | { |
| 2200 | unsigned long flags; |
| 2201 | int rc = 0; |
| 2202 | |
| 2203 | /* are we suspending? */ |
| 2204 | spin_lock_irqsave(ap->lock, flags); |
| 2205 | if (!(ap->pflags & ATA_PFLAG_PM_PENDING) || |
| 2206 | ap->pm_mesg.event == PM_EVENT_ON) { |
| 2207 | spin_unlock_irqrestore(ap->lock, flags); |
| 2208 | return; |
| 2209 | } |
| 2210 | spin_unlock_irqrestore(ap->lock, flags); |
| 2211 | |
| 2212 | WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED); |
| 2213 | |
| 2214 | /* suspend */ |
| 2215 | ata_eh_freeze_port(ap); |
| 2216 | |
| 2217 | if (ap->ops->port_suspend) |
| 2218 | rc = ap->ops->port_suspend(ap, ap->pm_mesg); |
| 2219 | |
| 2220 | /* report result */ |
| 2221 | spin_lock_irqsave(ap->lock, flags); |
| 2222 | |
| 2223 | ap->pflags &= ~ATA_PFLAG_PM_PENDING; |
| 2224 | if (rc == 0) |
| 2225 | ap->pflags |= ATA_PFLAG_SUSPENDED; |
| 2226 | else |
| 2227 | ata_port_schedule_eh(ap); |
| 2228 | |
| 2229 | if (ap->pm_result) { |
| 2230 | *ap->pm_result = rc; |
| 2231 | ap->pm_result = NULL; |
| 2232 | } |
| 2233 | |
| 2234 | spin_unlock_irqrestore(ap->lock, flags); |
| 2235 | |
| 2236 | return; |
| 2237 | } |
| 2238 | |
| 2239 | /** |
| 2240 | * ata_eh_handle_port_resume - perform port resume operation |
| 2241 | * @ap: port to resume |
| 2242 | * |
| 2243 | * Resume @ap. |
| 2244 | * |
| 2245 | * This function also waits upto one second until all devices |
| 2246 | * hanging off this port requests resume EH action. This is to |
| 2247 | * prevent invoking EH and thus reset multiple times on resume. |
| 2248 | * |
| 2249 | * On DPM resume, where some of devices might not be resumed |
| 2250 | * together, this may delay port resume upto one second, but such |
| 2251 | * DPM resumes are rare and 1 sec delay isn't too bad. |
| 2252 | * |
| 2253 | * LOCKING: |
| 2254 | * Kernel thread context (may sleep). |
| 2255 | */ |
| 2256 | static void ata_eh_handle_port_resume(struct ata_port *ap) |
| 2257 | { |
| 2258 | unsigned long timeout; |
| 2259 | unsigned long flags; |
| 2260 | int i, rc = 0; |
| 2261 | |
| 2262 | /* are we resuming? */ |
| 2263 | spin_lock_irqsave(ap->lock, flags); |
| 2264 | if (!(ap->pflags & ATA_PFLAG_PM_PENDING) || |
| 2265 | ap->pm_mesg.event != PM_EVENT_ON) { |
| 2266 | spin_unlock_irqrestore(ap->lock, flags); |
| 2267 | return; |
| 2268 | } |
| 2269 | spin_unlock_irqrestore(ap->lock, flags); |
| 2270 | |
| 2271 | /* spurious? */ |
| 2272 | if (!(ap->pflags & ATA_PFLAG_SUSPENDED)) |
| 2273 | goto done; |
| 2274 | |
| 2275 | if (ap->ops->port_resume) |
| 2276 | rc = ap->ops->port_resume(ap); |
| 2277 | |
| 2278 | /* give devices time to request EH */ |
| 2279 | timeout = jiffies + HZ; /* 1s max */ |
| 2280 | while (1) { |
| 2281 | for (i = 0; i < ATA_MAX_DEVICES; i++) { |
| 2282 | struct ata_device *dev = &ap->device[i]; |
| 2283 | unsigned int action = ata_eh_dev_action(dev); |
| 2284 | |
| 2285 | if ((dev->flags & ATA_DFLAG_SUSPENDED) && |
| 2286 | !(action & ATA_EH_RESUME)) |
| 2287 | break; |
| 2288 | } |
| 2289 | |
| 2290 | if (i == ATA_MAX_DEVICES || time_after(jiffies, timeout)) |
| 2291 | break; |
| 2292 | msleep(10); |
| 2293 | } |
| 2294 | |
| 2295 | done: |
| 2296 | spin_lock_irqsave(ap->lock, flags); |
| 2297 | ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED); |
| 2298 | if (ap->pm_result) { |
| 2299 | *ap->pm_result = rc; |
| 2300 | ap->pm_result = NULL; |
| 2301 | } |
| 2302 | spin_unlock_irqrestore(ap->lock, flags); |
| 2303 | } |