1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Driver for the Micron P320 SSD 4 * Copyright (C) 2011 Micron Technology, Inc. 5 * 6 * Portions of this code were derived from works subjected to the 7 * following copyright: 8 * Copyright (C) 2009 Integrated Device Technology, Inc. 9 */ 10 11 #include <linux/pci.h> 12 #include <linux/interrupt.h> 13 #include <linux/ata.h> 14 #include <linux/delay.h> 15 #include <linux/hdreg.h> 16 #include <linux/uaccess.h> 17 #include <linux/random.h> 18 #include <linux/smp.h> 19 #include <linux/compat.h> 20 #include <linux/fs.h> 21 #include <linux/module.h> 22 #include <linux/genhd.h> 23 #include <linux/blkdev.h> 24 #include <linux/blk-mq.h> 25 #include <linux/bio.h> 26 #include <linux/dma-mapping.h> 27 #include <linux/idr.h> 28 #include <linux/kthread.h> 29 #include <../drivers/ata/ahci.h> 30 #include <linux/export.h> 31 #include <linux/debugfs.h> 32 #include <linux/prefetch.h> 33 #include <linux/numa.h> 34 #include "mtip32xx.h" 35 36 #define HW_CMD_SLOT_SZ (MTIP_MAX_COMMAND_SLOTS * 32) 37 38 /* DMA region containing RX Fis, Identify, RLE10, and SMART buffers */ 39 #define AHCI_RX_FIS_SZ 0x100 40 #define AHCI_RX_FIS_OFFSET 0x0 41 #define AHCI_IDFY_SZ ATA_SECT_SIZE 42 #define AHCI_IDFY_OFFSET 0x400 43 #define AHCI_SECTBUF_SZ ATA_SECT_SIZE 44 #define AHCI_SECTBUF_OFFSET 0x800 45 #define AHCI_SMARTBUF_SZ ATA_SECT_SIZE 46 #define AHCI_SMARTBUF_OFFSET 0xC00 47 /* 0x100 + 0x200 + 0x200 + 0x200 is smaller than 4k but we pad it out */ 48 #define BLOCK_DMA_ALLOC_SZ 4096 49 50 /* DMA region containing command table (should be 8192 bytes) */ 51 #define AHCI_CMD_SLOT_SZ sizeof(struct mtip_cmd_hdr) 52 #define AHCI_CMD_TBL_SZ (MTIP_MAX_COMMAND_SLOTS * AHCI_CMD_SLOT_SZ) 53 #define AHCI_CMD_TBL_OFFSET 0x0 54 55 /* DMA region per command (contains header and SGL) */ 56 #define AHCI_CMD_TBL_HDR_SZ 0x80 57 #define AHCI_CMD_TBL_HDR_OFFSET 0x0 58 #define AHCI_CMD_TBL_SGL_SZ (MTIP_MAX_SG * sizeof(struct mtip_cmd_sg)) 59 #define AHCI_CMD_TBL_SGL_OFFSET AHCI_CMD_TBL_HDR_SZ 60 #define CMD_DMA_ALLOC_SZ (AHCI_CMD_TBL_SGL_SZ + AHCI_CMD_TBL_HDR_SZ) 61 62 63 #define HOST_CAP_NZDMA (1 << 19) 64 #define HOST_HSORG 0xFC 65 #define HSORG_DISABLE_SLOTGRP_INTR (1<<24) 66 #define HSORG_DISABLE_SLOTGRP_PXIS (1<<16) 67 #define HSORG_HWREV 0xFF00 68 #define HSORG_STYLE 0x8 69 #define HSORG_SLOTGROUPS 0x7 70 71 #define PORT_COMMAND_ISSUE 0x38 72 #define PORT_SDBV 0x7C 73 74 #define PORT_OFFSET 0x100 75 #define PORT_MEM_SIZE 0x80 76 77 #define PORT_IRQ_ERR \ 78 (PORT_IRQ_HBUS_ERR | PORT_IRQ_IF_ERR | PORT_IRQ_CONNECT | \ 79 PORT_IRQ_PHYRDY | PORT_IRQ_UNK_FIS | PORT_IRQ_BAD_PMP | \ 80 PORT_IRQ_TF_ERR | PORT_IRQ_HBUS_DATA_ERR | PORT_IRQ_IF_NONFATAL | \ 81 PORT_IRQ_OVERFLOW) 82 #define PORT_IRQ_LEGACY \ 83 (PORT_IRQ_PIOS_FIS | PORT_IRQ_D2H_REG_FIS) 84 #define PORT_IRQ_HANDLED \ 85 (PORT_IRQ_SDB_FIS | PORT_IRQ_LEGACY | \ 86 PORT_IRQ_TF_ERR | PORT_IRQ_IF_ERR | \ 87 PORT_IRQ_CONNECT | PORT_IRQ_PHYRDY) 88 #define DEF_PORT_IRQ \ 89 (PORT_IRQ_ERR | PORT_IRQ_LEGACY | PORT_IRQ_SDB_FIS) 90 91 /* product numbers */ 92 #define MTIP_PRODUCT_UNKNOWN 0x00 93 #define MTIP_PRODUCT_ASICFPGA 0x11 94 95 /* Device instance number, incremented each time a device is probed. */ 96 static int instance; 97 98 static struct list_head online_list; 99 static struct list_head removing_list; 100 static spinlock_t dev_lock; 101 102 /* 103 * Global variable used to hold the major block device number 104 * allocated in mtip_init(). 105 */ 106 static int mtip_major; 107 static struct dentry *dfs_parent; 108 static struct dentry *dfs_device_status; 109 110 static u32 cpu_use[NR_CPUS]; 111 112 static DEFINE_IDA(rssd_index_ida); 113 114 static int mtip_block_initialize(struct driver_data *dd); 115 116 #ifdef CONFIG_COMPAT 117 struct mtip_compat_ide_task_request_s { 118 __u8 io_ports[8]; 119 __u8 hob_ports[8]; 120 ide_reg_valid_t out_flags; 121 ide_reg_valid_t in_flags; 122 int data_phase; 123 int req_cmd; 124 compat_ulong_t out_size; 125 compat_ulong_t in_size; 126 }; 127 #endif 128 129 /* 130 * This function check_for_surprise_removal is called 131 * while card is removed from the system and it will 132 * read the vendor id from the configuration space 133 * 134 * @pdev Pointer to the pci_dev structure. 135 * 136 * return value 137 * true if device removed, else false 138 */ 139 static bool mtip_check_surprise_removal(struct pci_dev *pdev) 140 { 141 u16 vendor_id = 0; 142 struct driver_data *dd = pci_get_drvdata(pdev); 143 144 if (dd->sr) 145 return true; 146 147 /* Read the vendorID from the configuration space */ 148 pci_read_config_word(pdev, 0x00, &vendor_id); 149 if (vendor_id == 0xFFFF) { 150 dd->sr = true; 151 if (dd->queue) 152 blk_queue_flag_set(QUEUE_FLAG_DEAD, dd->queue); 153 else 154 dev_warn(&dd->pdev->dev, 155 "%s: dd->queue is NULL\n", __func__); 156 return true; /* device removed */ 157 } 158 159 return false; /* device present */ 160 } 161 162 static struct mtip_cmd *mtip_cmd_from_tag(struct driver_data *dd, 163 unsigned int tag) 164 { 165 struct blk_mq_hw_ctx *hctx = dd->queue->queue_hw_ctx[0]; 166 167 return blk_mq_rq_to_pdu(blk_mq_tag_to_rq(hctx->tags, tag)); 168 } 169 170 /* 171 * Reset the HBA (without sleeping) 172 * 173 * @dd Pointer to the driver data structure. 174 * 175 * return value 176 * 0 The reset was successful. 177 * -1 The HBA Reset bit did not clear. 178 */ 179 static int mtip_hba_reset(struct driver_data *dd) 180 { 181 unsigned long timeout; 182 183 /* Set the reset bit */ 184 writel(HOST_RESET, dd->mmio + HOST_CTL); 185 186 /* Flush */ 187 readl(dd->mmio + HOST_CTL); 188 189 /* 190 * Spin for up to 10 seconds waiting for reset acknowledgement. Spec 191 * is 1 sec but in LUN failure conditions, up to 10 secs are required 192 */ 193 timeout = jiffies + msecs_to_jiffies(10000); 194 do { 195 mdelay(10); 196 if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)) 197 return -1; 198 199 } while ((readl(dd->mmio + HOST_CTL) & HOST_RESET) 200 && time_before(jiffies, timeout)); 201 202 if (readl(dd->mmio + HOST_CTL) & HOST_RESET) 203 return -1; 204 205 return 0; 206 } 207 208 /* 209 * Issue a command to the hardware. 210 * 211 * Set the appropriate bit in the s_active and Command Issue hardware 212 * registers, causing hardware command processing to begin. 213 * 214 * @port Pointer to the port structure. 215 * @tag The tag of the command to be issued. 216 * 217 * return value 218 * None 219 */ 220 static inline void mtip_issue_ncq_command(struct mtip_port *port, int tag) 221 { 222 int group = tag >> 5; 223 224 /* guard SACT and CI registers */ 225 spin_lock(&port->cmd_issue_lock[group]); 226 writel((1 << MTIP_TAG_BIT(tag)), 227 port->s_active[MTIP_TAG_INDEX(tag)]); 228 writel((1 << MTIP_TAG_BIT(tag)), 229 port->cmd_issue[MTIP_TAG_INDEX(tag)]); 230 spin_unlock(&port->cmd_issue_lock[group]); 231 } 232 233 /* 234 * Enable/disable the reception of FIS 235 * 236 * @port Pointer to the port data structure 237 * @enable 1 to enable, 0 to disable 238 * 239 * return value 240 * Previous state: 1 enabled, 0 disabled 241 */ 242 static int mtip_enable_fis(struct mtip_port *port, int enable) 243 { 244 u32 tmp; 245 246 /* enable FIS reception */ 247 tmp = readl(port->mmio + PORT_CMD); 248 if (enable) 249 writel(tmp | PORT_CMD_FIS_RX, port->mmio + PORT_CMD); 250 else 251 writel(tmp & ~PORT_CMD_FIS_RX, port->mmio + PORT_CMD); 252 253 /* Flush */ 254 readl(port->mmio + PORT_CMD); 255 256 return (((tmp & PORT_CMD_FIS_RX) == PORT_CMD_FIS_RX)); 257 } 258 259 /* 260 * Enable/disable the DMA engine 261 * 262 * @port Pointer to the port data structure 263 * @enable 1 to enable, 0 to disable 264 * 265 * return value 266 * Previous state: 1 enabled, 0 disabled. 267 */ 268 static int mtip_enable_engine(struct mtip_port *port, int enable) 269 { 270 u32 tmp; 271 272 /* enable FIS reception */ 273 tmp = readl(port->mmio + PORT_CMD); 274 if (enable) 275 writel(tmp | PORT_CMD_START, port->mmio + PORT_CMD); 276 else 277 writel(tmp & ~PORT_CMD_START, port->mmio + PORT_CMD); 278 279 readl(port->mmio + PORT_CMD); 280 return (((tmp & PORT_CMD_START) == PORT_CMD_START)); 281 } 282 283 /* 284 * Enables the port DMA engine and FIS reception. 285 * 286 * return value 287 * None 288 */ 289 static inline void mtip_start_port(struct mtip_port *port) 290 { 291 /* Enable FIS reception */ 292 mtip_enable_fis(port, 1); 293 294 /* Enable the DMA engine */ 295 mtip_enable_engine(port, 1); 296 } 297 298 /* 299 * Deinitialize a port by disabling port interrupts, the DMA engine, 300 * and FIS reception. 301 * 302 * @port Pointer to the port structure 303 * 304 * return value 305 * None 306 */ 307 static inline void mtip_deinit_port(struct mtip_port *port) 308 { 309 /* Disable interrupts on this port */ 310 writel(0, port->mmio + PORT_IRQ_MASK); 311 312 /* Disable the DMA engine */ 313 mtip_enable_engine(port, 0); 314 315 /* Disable FIS reception */ 316 mtip_enable_fis(port, 0); 317 } 318 319 /* 320 * Initialize a port. 321 * 322 * This function deinitializes the port by calling mtip_deinit_port() and 323 * then initializes it by setting the command header and RX FIS addresses, 324 * clearing the SError register and any pending port interrupts before 325 * re-enabling the default set of port interrupts. 326 * 327 * @port Pointer to the port structure. 328 * 329 * return value 330 * None 331 */ 332 static void mtip_init_port(struct mtip_port *port) 333 { 334 int i; 335 mtip_deinit_port(port); 336 337 /* Program the command list base and FIS base addresses */ 338 if (readl(port->dd->mmio + HOST_CAP) & HOST_CAP_64) { 339 writel((port->command_list_dma >> 16) >> 16, 340 port->mmio + PORT_LST_ADDR_HI); 341 writel((port->rxfis_dma >> 16) >> 16, 342 port->mmio + PORT_FIS_ADDR_HI); 343 set_bit(MTIP_PF_HOST_CAP_64, &port->flags); 344 } 345 346 writel(port->command_list_dma & 0xFFFFFFFF, 347 port->mmio + PORT_LST_ADDR); 348 writel(port->rxfis_dma & 0xFFFFFFFF, port->mmio + PORT_FIS_ADDR); 349 350 /* Clear SError */ 351 writel(readl(port->mmio + PORT_SCR_ERR), port->mmio + PORT_SCR_ERR); 352 353 /* reset the completed registers.*/ 354 for (i = 0; i < port->dd->slot_groups; i++) 355 writel(0xFFFFFFFF, port->completed[i]); 356 357 /* Clear any pending interrupts for this port */ 358 writel(readl(port->mmio + PORT_IRQ_STAT), port->mmio + PORT_IRQ_STAT); 359 360 /* Clear any pending interrupts on the HBA. */ 361 writel(readl(port->dd->mmio + HOST_IRQ_STAT), 362 port->dd->mmio + HOST_IRQ_STAT); 363 364 /* Enable port interrupts */ 365 writel(DEF_PORT_IRQ, port->mmio + PORT_IRQ_MASK); 366 } 367 368 /* 369 * Restart a port 370 * 371 * @port Pointer to the port data structure. 372 * 373 * return value 374 * None 375 */ 376 static void mtip_restart_port(struct mtip_port *port) 377 { 378 unsigned long timeout; 379 380 /* Disable the DMA engine */ 381 mtip_enable_engine(port, 0); 382 383 /* Chip quirk: wait up to 500ms for PxCMD.CR == 0 */ 384 timeout = jiffies + msecs_to_jiffies(500); 385 while ((readl(port->mmio + PORT_CMD) & PORT_CMD_LIST_ON) 386 && time_before(jiffies, timeout)) 387 ; 388 389 if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag)) 390 return; 391 392 /* 393 * Chip quirk: escalate to hba reset if 394 * PxCMD.CR not clear after 500 ms 395 */ 396 if (readl(port->mmio + PORT_CMD) & PORT_CMD_LIST_ON) { 397 dev_warn(&port->dd->pdev->dev, 398 "PxCMD.CR not clear, escalating reset\n"); 399 400 if (mtip_hba_reset(port->dd)) 401 dev_err(&port->dd->pdev->dev, 402 "HBA reset escalation failed.\n"); 403 404 /* 30 ms delay before com reset to quiesce chip */ 405 mdelay(30); 406 } 407 408 dev_warn(&port->dd->pdev->dev, "Issuing COM reset\n"); 409 410 /* Set PxSCTL.DET */ 411 writel(readl(port->mmio + PORT_SCR_CTL) | 412 1, port->mmio + PORT_SCR_CTL); 413 readl(port->mmio + PORT_SCR_CTL); 414 415 /* Wait 1 ms to quiesce chip function */ 416 timeout = jiffies + msecs_to_jiffies(1); 417 while (time_before(jiffies, timeout)) 418 ; 419 420 if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag)) 421 return; 422 423 /* Clear PxSCTL.DET */ 424 writel(readl(port->mmio + PORT_SCR_CTL) & ~1, 425 port->mmio + PORT_SCR_CTL); 426 readl(port->mmio + PORT_SCR_CTL); 427 428 /* Wait 500 ms for bit 0 of PORT_SCR_STS to be set */ 429 timeout = jiffies + msecs_to_jiffies(500); 430 while (((readl(port->mmio + PORT_SCR_STAT) & 0x01) == 0) 431 && time_before(jiffies, timeout)) 432 ; 433 434 if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag)) 435 return; 436 437 if ((readl(port->mmio + PORT_SCR_STAT) & 0x01) == 0) 438 dev_warn(&port->dd->pdev->dev, 439 "COM reset failed\n"); 440 441 mtip_init_port(port); 442 mtip_start_port(port); 443 444 } 445 446 static int mtip_device_reset(struct driver_data *dd) 447 { 448 int rv = 0; 449 450 if (mtip_check_surprise_removal(dd->pdev)) 451 return 0; 452 453 if (mtip_hba_reset(dd) < 0) 454 rv = -EFAULT; 455 456 mdelay(1); 457 mtip_init_port(dd->port); 458 mtip_start_port(dd->port); 459 460 /* Enable interrupts on the HBA. */ 461 writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN, 462 dd->mmio + HOST_CTL); 463 return rv; 464 } 465 466 /* 467 * Helper function for tag logging 468 */ 469 static void print_tags(struct driver_data *dd, 470 char *msg, 471 unsigned long *tagbits, 472 int cnt) 473 { 474 unsigned char tagmap[128]; 475 int group, tagmap_len = 0; 476 477 memset(tagmap, 0, sizeof(tagmap)); 478 for (group = SLOTBITS_IN_LONGS; group > 0; group--) 479 tagmap_len += sprintf(tagmap + tagmap_len, "%016lX ", 480 tagbits[group-1]); 481 dev_warn(&dd->pdev->dev, 482 "%d command(s) %s: tagmap [%s]", cnt, msg, tagmap); 483 } 484 485 static int mtip_read_log_page(struct mtip_port *port, u8 page, u16 *buffer, 486 dma_addr_t buffer_dma, unsigned int sectors); 487 static int mtip_get_smart_attr(struct mtip_port *port, unsigned int id, 488 struct smart_attr *attrib); 489 490 static void mtip_complete_command(struct mtip_cmd *cmd, blk_status_t status) 491 { 492 struct request *req = blk_mq_rq_from_pdu(cmd); 493 494 cmd->status = status; 495 blk_mq_complete_request(req); 496 } 497 498 /* 499 * Handle an error. 500 * 501 * @dd Pointer to the DRIVER_DATA structure. 502 * 503 * return value 504 * None 505 */ 506 static void mtip_handle_tfe(struct driver_data *dd) 507 { 508 int group, tag, bit, reissue, rv; 509 struct mtip_port *port; 510 struct mtip_cmd *cmd; 511 u32 completed; 512 struct host_to_dev_fis *fis; 513 unsigned long tagaccum[SLOTBITS_IN_LONGS]; 514 unsigned int cmd_cnt = 0; 515 unsigned char *buf; 516 char *fail_reason = NULL; 517 int fail_all_ncq_write = 0, fail_all_ncq_cmds = 0; 518 519 dev_warn(&dd->pdev->dev, "Taskfile error\n"); 520 521 port = dd->port; 522 523 if (test_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags)) { 524 cmd = mtip_cmd_from_tag(dd, MTIP_TAG_INTERNAL); 525 dbg_printk(MTIP_DRV_NAME " TFE for the internal command\n"); 526 mtip_complete_command(cmd, BLK_STS_IOERR); 527 return; 528 } 529 530 /* clear the tag accumulator */ 531 memset(tagaccum, 0, SLOTBITS_IN_LONGS * sizeof(long)); 532 533 /* Loop through all the groups */ 534 for (group = 0; group < dd->slot_groups; group++) { 535 completed = readl(port->completed[group]); 536 537 dev_warn(&dd->pdev->dev, "g=%u, comp=%x\n", group, completed); 538 539 /* clear completed status register in the hardware.*/ 540 writel(completed, port->completed[group]); 541 542 /* Process successfully completed commands */ 543 for (bit = 0; bit < 32 && completed; bit++) { 544 if (!(completed & (1<<bit))) 545 continue; 546 tag = (group << 5) + bit; 547 548 /* Skip the internal command slot */ 549 if (tag == MTIP_TAG_INTERNAL) 550 continue; 551 552 cmd = mtip_cmd_from_tag(dd, tag); 553 mtip_complete_command(cmd, 0); 554 set_bit(tag, tagaccum); 555 cmd_cnt++; 556 } 557 } 558 559 print_tags(dd, "completed (TFE)", tagaccum, cmd_cnt); 560 561 /* Restart the port */ 562 mdelay(20); 563 mtip_restart_port(port); 564 565 /* Trying to determine the cause of the error */ 566 rv = mtip_read_log_page(dd->port, ATA_LOG_SATA_NCQ, 567 dd->port->log_buf, 568 dd->port->log_buf_dma, 1); 569 if (rv) { 570 dev_warn(&dd->pdev->dev, 571 "Error in READ LOG EXT (10h) command\n"); 572 /* non-critical error, don't fail the load */ 573 } else { 574 buf = (unsigned char *)dd->port->log_buf; 575 if (buf[259] & 0x1) { 576 dev_info(&dd->pdev->dev, 577 "Write protect bit is set.\n"); 578 set_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag); 579 fail_all_ncq_write = 1; 580 fail_reason = "write protect"; 581 } 582 if (buf[288] == 0xF7) { 583 dev_info(&dd->pdev->dev, 584 "Exceeded Tmax, drive in thermal shutdown.\n"); 585 set_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag); 586 fail_all_ncq_cmds = 1; 587 fail_reason = "thermal shutdown"; 588 } 589 if (buf[288] == 0xBF) { 590 set_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag); 591 dev_info(&dd->pdev->dev, 592 "Drive indicates rebuild has failed. Secure erase required.\n"); 593 fail_all_ncq_cmds = 1; 594 fail_reason = "rebuild failed"; 595 } 596 } 597 598 /* clear the tag accumulator */ 599 memset(tagaccum, 0, SLOTBITS_IN_LONGS * sizeof(long)); 600 601 /* Loop through all the groups */ 602 for (group = 0; group < dd->slot_groups; group++) { 603 for (bit = 0; bit < 32; bit++) { 604 reissue = 1; 605 tag = (group << 5) + bit; 606 cmd = mtip_cmd_from_tag(dd, tag); 607 608 fis = (struct host_to_dev_fis *)cmd->command; 609 610 /* Should re-issue? */ 611 if (tag == MTIP_TAG_INTERNAL || 612 fis->command == ATA_CMD_SET_FEATURES) 613 reissue = 0; 614 else { 615 if (fail_all_ncq_cmds || 616 (fail_all_ncq_write && 617 fis->command == ATA_CMD_FPDMA_WRITE)) { 618 dev_warn(&dd->pdev->dev, 619 " Fail: %s w/tag %d [%s].\n", 620 fis->command == ATA_CMD_FPDMA_WRITE ? 621 "write" : "read", 622 tag, 623 fail_reason != NULL ? 624 fail_reason : "unknown"); 625 mtip_complete_command(cmd, BLK_STS_MEDIUM); 626 continue; 627 } 628 } 629 630 /* 631 * First check if this command has 632 * exceeded its retries. 633 */ 634 if (reissue && (cmd->retries-- > 0)) { 635 636 set_bit(tag, tagaccum); 637 638 /* Re-issue the command. */ 639 mtip_issue_ncq_command(port, tag); 640 641 continue; 642 } 643 644 /* Retire a command that will not be reissued */ 645 dev_warn(&port->dd->pdev->dev, 646 "retiring tag %d\n", tag); 647 648 mtip_complete_command(cmd, BLK_STS_IOERR); 649 } 650 } 651 print_tags(dd, "reissued (TFE)", tagaccum, cmd_cnt); 652 } 653 654 /* 655 * Handle a set device bits interrupt 656 */ 657 static inline void mtip_workq_sdbfx(struct mtip_port *port, int group, 658 u32 completed) 659 { 660 struct driver_data *dd = port->dd; 661 int tag, bit; 662 struct mtip_cmd *command; 663 664 if (!completed) { 665 WARN_ON_ONCE(!completed); 666 return; 667 } 668 /* clear completed status register in the hardware.*/ 669 writel(completed, port->completed[group]); 670 671 /* Process completed commands. */ 672 for (bit = 0; (bit < 32) && completed; bit++) { 673 if (completed & 0x01) { 674 tag = (group << 5) | bit; 675 676 /* skip internal command slot. */ 677 if (unlikely(tag == MTIP_TAG_INTERNAL)) 678 continue; 679 680 command = mtip_cmd_from_tag(dd, tag); 681 mtip_complete_command(command, 0); 682 } 683 completed >>= 1; 684 } 685 686 /* If last, re-enable interrupts */ 687 if (atomic_dec_return(&dd->irq_workers_active) == 0) 688 writel(0xffffffff, dd->mmio + HOST_IRQ_STAT); 689 } 690 691 /* 692 * Process legacy pio and d2h interrupts 693 */ 694 static inline void mtip_process_legacy(struct driver_data *dd, u32 port_stat) 695 { 696 struct mtip_port *port = dd->port; 697 struct mtip_cmd *cmd = mtip_cmd_from_tag(dd, MTIP_TAG_INTERNAL); 698 699 if (test_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags) && cmd) { 700 int group = MTIP_TAG_INDEX(MTIP_TAG_INTERNAL); 701 int status = readl(port->cmd_issue[group]); 702 703 if (!(status & (1 << MTIP_TAG_BIT(MTIP_TAG_INTERNAL)))) 704 mtip_complete_command(cmd, 0); 705 } 706 } 707 708 /* 709 * Demux and handle errors 710 */ 711 static inline void mtip_process_errors(struct driver_data *dd, u32 port_stat) 712 { 713 if (unlikely(port_stat & PORT_IRQ_CONNECT)) { 714 dev_warn(&dd->pdev->dev, 715 "Clearing PxSERR.DIAG.x\n"); 716 writel((1 << 26), dd->port->mmio + PORT_SCR_ERR); 717 } 718 719 if (unlikely(port_stat & PORT_IRQ_PHYRDY)) { 720 dev_warn(&dd->pdev->dev, 721 "Clearing PxSERR.DIAG.n\n"); 722 writel((1 << 16), dd->port->mmio + PORT_SCR_ERR); 723 } 724 725 if (unlikely(port_stat & ~PORT_IRQ_HANDLED)) { 726 dev_warn(&dd->pdev->dev, 727 "Port stat errors %x unhandled\n", 728 (port_stat & ~PORT_IRQ_HANDLED)); 729 if (mtip_check_surprise_removal(dd->pdev)) 730 return; 731 } 732 if (likely(port_stat & (PORT_IRQ_TF_ERR | PORT_IRQ_IF_ERR))) { 733 set_bit(MTIP_PF_EH_ACTIVE_BIT, &dd->port->flags); 734 wake_up_interruptible(&dd->port->svc_wait); 735 } 736 } 737 738 static inline irqreturn_t mtip_handle_irq(struct driver_data *data) 739 { 740 struct driver_data *dd = (struct driver_data *) data; 741 struct mtip_port *port = dd->port; 742 u32 hba_stat, port_stat; 743 int rv = IRQ_NONE; 744 int do_irq_enable = 1, i, workers; 745 struct mtip_work *twork; 746 747 hba_stat = readl(dd->mmio + HOST_IRQ_STAT); 748 if (hba_stat) { 749 rv = IRQ_HANDLED; 750 751 /* Acknowledge the interrupt status on the port.*/ 752 port_stat = readl(port->mmio + PORT_IRQ_STAT); 753 if (unlikely(port_stat == 0xFFFFFFFF)) { 754 mtip_check_surprise_removal(dd->pdev); 755 return IRQ_HANDLED; 756 } 757 writel(port_stat, port->mmio + PORT_IRQ_STAT); 758 759 /* Demux port status */ 760 if (likely(port_stat & PORT_IRQ_SDB_FIS)) { 761 do_irq_enable = 0; 762 WARN_ON_ONCE(atomic_read(&dd->irq_workers_active) != 0); 763 764 /* Start at 1: group zero is always local? */ 765 for (i = 0, workers = 0; i < MTIP_MAX_SLOT_GROUPS; 766 i++) { 767 twork = &dd->work[i]; 768 twork->completed = readl(port->completed[i]); 769 if (twork->completed) 770 workers++; 771 } 772 773 atomic_set(&dd->irq_workers_active, workers); 774 if (workers) { 775 for (i = 1; i < MTIP_MAX_SLOT_GROUPS; i++) { 776 twork = &dd->work[i]; 777 if (twork->completed) 778 queue_work_on( 779 twork->cpu_binding, 780 dd->isr_workq, 781 &twork->work); 782 } 783 784 if (likely(dd->work[0].completed)) 785 mtip_workq_sdbfx(port, 0, 786 dd->work[0].completed); 787 788 } else { 789 /* 790 * Chip quirk: SDB interrupt but nothing 791 * to complete 792 */ 793 do_irq_enable = 1; 794 } 795 } 796 797 if (unlikely(port_stat & PORT_IRQ_ERR)) { 798 if (unlikely(mtip_check_surprise_removal(dd->pdev))) { 799 /* don't proceed further */ 800 return IRQ_HANDLED; 801 } 802 if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, 803 &dd->dd_flag)) 804 return rv; 805 806 mtip_process_errors(dd, port_stat & PORT_IRQ_ERR); 807 } 808 809 if (unlikely(port_stat & PORT_IRQ_LEGACY)) 810 mtip_process_legacy(dd, port_stat & PORT_IRQ_LEGACY); 811 } 812 813 /* acknowledge interrupt */ 814 if (unlikely(do_irq_enable)) 815 writel(hba_stat, dd->mmio + HOST_IRQ_STAT); 816 817 return rv; 818 } 819 820 /* 821 * HBA interrupt subroutine. 822 * 823 * @irq IRQ number. 824 * @instance Pointer to the driver data structure. 825 * 826 * return value 827 * IRQ_HANDLED A HBA interrupt was pending and handled. 828 * IRQ_NONE This interrupt was not for the HBA. 829 */ 830 static irqreturn_t mtip_irq_handler(int irq, void *instance) 831 { 832 struct driver_data *dd = instance; 833 834 return mtip_handle_irq(dd); 835 } 836 837 static void mtip_issue_non_ncq_command(struct mtip_port *port, int tag) 838 { 839 writel(1 << MTIP_TAG_BIT(tag), port->cmd_issue[MTIP_TAG_INDEX(tag)]); 840 } 841 842 static bool mtip_pause_ncq(struct mtip_port *port, 843 struct host_to_dev_fis *fis) 844 { 845 unsigned long task_file_data; 846 847 task_file_data = readl(port->mmio+PORT_TFDATA); 848 if ((task_file_data & 1)) 849 return false; 850 851 if (fis->command == ATA_CMD_SEC_ERASE_PREP) { 852 port->ic_pause_timer = jiffies; 853 return true; 854 } else if ((fis->command == ATA_CMD_DOWNLOAD_MICRO) && 855 (fis->features == 0x03)) { 856 set_bit(MTIP_PF_DM_ACTIVE_BIT, &port->flags); 857 port->ic_pause_timer = jiffies; 858 return true; 859 } else if ((fis->command == ATA_CMD_SEC_ERASE_UNIT) || 860 ((fis->command == 0xFC) && 861 (fis->features == 0x27 || fis->features == 0x72 || 862 fis->features == 0x62 || fis->features == 0x26))) { 863 clear_bit(MTIP_DDF_SEC_LOCK_BIT, &port->dd->dd_flag); 864 clear_bit(MTIP_DDF_REBUILD_FAILED_BIT, &port->dd->dd_flag); 865 /* Com reset after secure erase or lowlevel format */ 866 mtip_restart_port(port); 867 clear_bit(MTIP_PF_SE_ACTIVE_BIT, &port->flags); 868 return false; 869 } 870 871 return false; 872 } 873 874 static bool mtip_commands_active(struct mtip_port *port) 875 { 876 unsigned int active; 877 unsigned int n; 878 879 /* 880 * Ignore s_active bit 0 of array element 0. 881 * This bit will always be set 882 */ 883 active = readl(port->s_active[0]) & 0xFFFFFFFE; 884 for (n = 1; n < port->dd->slot_groups; n++) 885 active |= readl(port->s_active[n]); 886 887 return active != 0; 888 } 889 890 /* 891 * Wait for port to quiesce 892 * 893 * @port Pointer to port data structure 894 * @timeout Max duration to wait (ms) 895 * 896 * return value 897 * 0 Success 898 * -EBUSY Commands still active 899 */ 900 static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout) 901 { 902 unsigned long to; 903 bool active = true; 904 905 blk_mq_quiesce_queue(port->dd->queue); 906 907 to = jiffies + msecs_to_jiffies(timeout); 908 do { 909 if (test_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags) && 910 test_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags)) { 911 msleep(20); 912 continue; /* svc thd is actively issuing commands */ 913 } 914 915 msleep(100); 916 917 if (mtip_check_surprise_removal(port->dd->pdev)) 918 goto err_fault; 919 920 active = mtip_commands_active(port); 921 if (!active) 922 break; 923 } while (time_before(jiffies, to)); 924 925 blk_mq_unquiesce_queue(port->dd->queue); 926 return active ? -EBUSY : 0; 927 err_fault: 928 blk_mq_unquiesce_queue(port->dd->queue); 929 return -EFAULT; 930 } 931 932 struct mtip_int_cmd { 933 int fis_len; 934 dma_addr_t buffer; 935 int buf_len; 936 u32 opts; 937 }; 938 939 /* 940 * Execute an internal command and wait for the completion. 941 * 942 * @port Pointer to the port data structure. 943 * @fis Pointer to the FIS that describes the command. 944 * @fis_len Length in WORDS of the FIS. 945 * @buffer DMA accessible for command data. 946 * @buf_len Length, in bytes, of the data buffer. 947 * @opts Command header options, excluding the FIS length 948 * and the number of PRD entries. 949 * @timeout Time in ms to wait for the command to complete. 950 * 951 * return value 952 * 0 Command completed successfully. 953 * -EFAULT The buffer address is not correctly aligned. 954 * -EBUSY Internal command or other IO in progress. 955 * -EAGAIN Time out waiting for command to complete. 956 */ 957 static int mtip_exec_internal_command(struct mtip_port *port, 958 struct host_to_dev_fis *fis, 959 int fis_len, 960 dma_addr_t buffer, 961 int buf_len, 962 u32 opts, 963 unsigned long timeout) 964 { 965 struct mtip_cmd *int_cmd; 966 struct driver_data *dd = port->dd; 967 struct request *rq; 968 struct mtip_int_cmd icmd = { 969 .fis_len = fis_len, 970 .buffer = buffer, 971 .buf_len = buf_len, 972 .opts = opts 973 }; 974 int rv = 0; 975 976 /* Make sure the buffer is 8 byte aligned. This is asic specific. */ 977 if (buffer & 0x00000007) { 978 dev_err(&dd->pdev->dev, "SG buffer is not 8 byte aligned\n"); 979 return -EFAULT; 980 } 981 982 if (mtip_check_surprise_removal(dd->pdev)) 983 return -EFAULT; 984 985 rq = blk_mq_alloc_request(dd->queue, REQ_OP_DRV_IN, BLK_MQ_REQ_RESERVED); 986 if (IS_ERR(rq)) { 987 dbg_printk(MTIP_DRV_NAME "Unable to allocate tag for PIO cmd\n"); 988 return -EFAULT; 989 } 990 991 set_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags); 992 993 if (fis->command == ATA_CMD_SEC_ERASE_PREP) 994 set_bit(MTIP_PF_SE_ACTIVE_BIT, &port->flags); 995 996 clear_bit(MTIP_PF_DM_ACTIVE_BIT, &port->flags); 997 998 if (fis->command != ATA_CMD_STANDBYNOW1) { 999 /* wait for io to complete if non atomic */ 1000 if (mtip_quiesce_io(port, MTIP_QUIESCE_IO_TIMEOUT_MS) < 0) { 1001 dev_warn(&dd->pdev->dev, "Failed to quiesce IO\n"); 1002 blk_mq_free_request(rq); 1003 clear_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags); 1004 wake_up_interruptible(&port->svc_wait); 1005 return -EBUSY; 1006 } 1007 } 1008 1009 /* Copy the command to the command table */ 1010 int_cmd = blk_mq_rq_to_pdu(rq); 1011 int_cmd->icmd = &icmd; 1012 memcpy(int_cmd->command, fis, fis_len*4); 1013 1014 rq->timeout = timeout; 1015 1016 /* insert request and run queue */ 1017 blk_execute_rq(rq->q, NULL, rq, true); 1018 1019 if (int_cmd->status) { 1020 dev_err(&dd->pdev->dev, "Internal command [%02X] failed %d\n", 1021 fis->command, int_cmd->status); 1022 rv = -EIO; 1023 1024 if (mtip_check_surprise_removal(dd->pdev) || 1025 test_bit(MTIP_DDF_REMOVE_PENDING_BIT, 1026 &dd->dd_flag)) { 1027 dev_err(&dd->pdev->dev, 1028 "Internal command [%02X] wait returned due to SR\n", 1029 fis->command); 1030 rv = -ENXIO; 1031 goto exec_ic_exit; 1032 } 1033 mtip_device_reset(dd); /* recover from timeout issue */ 1034 rv = -EAGAIN; 1035 goto exec_ic_exit; 1036 } 1037 1038 if (readl(port->cmd_issue[MTIP_TAG_INDEX(MTIP_TAG_INTERNAL)]) 1039 & (1 << MTIP_TAG_BIT(MTIP_TAG_INTERNAL))) { 1040 rv = -ENXIO; 1041 if (!test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)) { 1042 mtip_device_reset(dd); 1043 rv = -EAGAIN; 1044 } 1045 } 1046 exec_ic_exit: 1047 /* Clear the allocated and active bits for the internal command. */ 1048 blk_mq_free_request(rq); 1049 clear_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags); 1050 if (rv >= 0 && mtip_pause_ncq(port, fis)) { 1051 /* NCQ paused */ 1052 return rv; 1053 } 1054 wake_up_interruptible(&port->svc_wait); 1055 1056 return rv; 1057 } 1058 1059 /* 1060 * Byte-swap ATA ID strings. 1061 * 1062 * ATA identify data contains strings in byte-swapped 16-bit words. 1063 * They must be swapped (on all architectures) to be usable as C strings. 1064 * This function swaps bytes in-place. 1065 * 1066 * @buf The buffer location of the string 1067 * @len The number of bytes to swap 1068 * 1069 * return value 1070 * None 1071 */ 1072 static inline void ata_swap_string(u16 *buf, unsigned int len) 1073 { 1074 int i; 1075 for (i = 0; i < (len/2); i++) 1076 be16_to_cpus(&buf[i]); 1077 } 1078 1079 static void mtip_set_timeout(struct driver_data *dd, 1080 struct host_to_dev_fis *fis, 1081 unsigned int *timeout, u8 erasemode) 1082 { 1083 switch (fis->command) { 1084 case ATA_CMD_DOWNLOAD_MICRO: 1085 *timeout = 120000; /* 2 minutes */ 1086 break; 1087 case ATA_CMD_SEC_ERASE_UNIT: 1088 case 0xFC: 1089 if (erasemode) 1090 *timeout = ((*(dd->port->identify + 90) * 2) * 60000); 1091 else 1092 *timeout = ((*(dd->port->identify + 89) * 2) * 60000); 1093 break; 1094 case ATA_CMD_STANDBYNOW1: 1095 *timeout = 120000; /* 2 minutes */ 1096 break; 1097 case 0xF7: 1098 case 0xFA: 1099 *timeout = 60000; /* 60 seconds */ 1100 break; 1101 case ATA_CMD_SMART: 1102 *timeout = 15000; /* 15 seconds */ 1103 break; 1104 default: 1105 *timeout = MTIP_IOCTL_CMD_TIMEOUT_MS; 1106 break; 1107 } 1108 } 1109 1110 /* 1111 * Request the device identity information. 1112 * 1113 * If a user space buffer is not specified, i.e. is NULL, the 1114 * identify information is still read from the drive and placed 1115 * into the identify data buffer (@e port->identify) in the 1116 * port data structure. 1117 * When the identify buffer contains valid identify information @e 1118 * port->identify_valid is non-zero. 1119 * 1120 * @port Pointer to the port structure. 1121 * @user_buffer A user space buffer where the identify data should be 1122 * copied. 1123 * 1124 * return value 1125 * 0 Command completed successfully. 1126 * -EFAULT An error occurred while coping data to the user buffer. 1127 * -1 Command failed. 1128 */ 1129 static int mtip_get_identify(struct mtip_port *port, void __user *user_buffer) 1130 { 1131 int rv = 0; 1132 struct host_to_dev_fis fis; 1133 1134 if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &port->dd->dd_flag)) 1135 return -EFAULT; 1136 1137 /* Build the FIS. */ 1138 memset(&fis, 0, sizeof(struct host_to_dev_fis)); 1139 fis.type = 0x27; 1140 fis.opts = 1 << 7; 1141 fis.command = ATA_CMD_ID_ATA; 1142 1143 /* Set the identify information as invalid. */ 1144 port->identify_valid = 0; 1145 1146 /* Clear the identify information. */ 1147 memset(port->identify, 0, sizeof(u16) * ATA_ID_WORDS); 1148 1149 /* Execute the command. */ 1150 if (mtip_exec_internal_command(port, 1151 &fis, 1152 5, 1153 port->identify_dma, 1154 sizeof(u16) * ATA_ID_WORDS, 1155 0, 1156 MTIP_INT_CMD_TIMEOUT_MS) 1157 < 0) { 1158 rv = -1; 1159 goto out; 1160 } 1161 1162 /* 1163 * Perform any necessary byte-swapping. Yes, the kernel does in fact 1164 * perform field-sensitive swapping on the string fields. 1165 * See the kernel use of ata_id_string() for proof of this. 1166 */ 1167 #ifdef __LITTLE_ENDIAN 1168 ata_swap_string(port->identify + 27, 40); /* model string*/ 1169 ata_swap_string(port->identify + 23, 8); /* firmware string*/ 1170 ata_swap_string(port->identify + 10, 20); /* serial# string*/ 1171 #else 1172 { 1173 int i; 1174 for (i = 0; i < ATA_ID_WORDS; i++) 1175 port->identify[i] = le16_to_cpu(port->identify[i]); 1176 } 1177 #endif 1178 1179 /* Check security locked state */ 1180 if (port->identify[128] & 0x4) 1181 set_bit(MTIP_DDF_SEC_LOCK_BIT, &port->dd->dd_flag); 1182 else 1183 clear_bit(MTIP_DDF_SEC_LOCK_BIT, &port->dd->dd_flag); 1184 1185 /* Set the identify buffer as valid. */ 1186 port->identify_valid = 1; 1187 1188 if (user_buffer) { 1189 if (copy_to_user( 1190 user_buffer, 1191 port->identify, 1192 ATA_ID_WORDS * sizeof(u16))) { 1193 rv = -EFAULT; 1194 goto out; 1195 } 1196 } 1197 1198 out: 1199 return rv; 1200 } 1201 1202 /* 1203 * Issue a standby immediate command to the device. 1204 * 1205 * @port Pointer to the port structure. 1206 * 1207 * return value 1208 * 0 Command was executed successfully. 1209 * -1 An error occurred while executing the command. 1210 */ 1211 static int mtip_standby_immediate(struct mtip_port *port) 1212 { 1213 int rv; 1214 struct host_to_dev_fis fis; 1215 unsigned long start; 1216 unsigned int timeout; 1217 1218 /* Build the FIS. */ 1219 memset(&fis, 0, sizeof(struct host_to_dev_fis)); 1220 fis.type = 0x27; 1221 fis.opts = 1 << 7; 1222 fis.command = ATA_CMD_STANDBYNOW1; 1223 1224 mtip_set_timeout(port->dd, &fis, &timeout, 0); 1225 1226 start = jiffies; 1227 rv = mtip_exec_internal_command(port, 1228 &fis, 1229 5, 1230 0, 1231 0, 1232 0, 1233 timeout); 1234 dbg_printk(MTIP_DRV_NAME "Time taken to complete standby cmd: %d ms\n", 1235 jiffies_to_msecs(jiffies - start)); 1236 if (rv) 1237 dev_warn(&port->dd->pdev->dev, 1238 "STANDBY IMMEDIATE command failed.\n"); 1239 1240 return rv; 1241 } 1242 1243 /* 1244 * Issue a READ LOG EXT command to the device. 1245 * 1246 * @port pointer to the port structure. 1247 * @page page number to fetch 1248 * @buffer pointer to buffer 1249 * @buffer_dma dma address corresponding to @buffer 1250 * @sectors page length to fetch, in sectors 1251 * 1252 * return value 1253 * @rv return value from mtip_exec_internal_command() 1254 */ 1255 static int mtip_read_log_page(struct mtip_port *port, u8 page, u16 *buffer, 1256 dma_addr_t buffer_dma, unsigned int sectors) 1257 { 1258 struct host_to_dev_fis fis; 1259 1260 memset(&fis, 0, sizeof(struct host_to_dev_fis)); 1261 fis.type = 0x27; 1262 fis.opts = 1 << 7; 1263 fis.command = ATA_CMD_READ_LOG_EXT; 1264 fis.sect_count = sectors & 0xFF; 1265 fis.sect_cnt_ex = (sectors >> 8) & 0xFF; 1266 fis.lba_low = page; 1267 fis.lba_mid = 0; 1268 fis.device = ATA_DEVICE_OBS; 1269 1270 memset(buffer, 0, sectors * ATA_SECT_SIZE); 1271 1272 return mtip_exec_internal_command(port, 1273 &fis, 1274 5, 1275 buffer_dma, 1276 sectors * ATA_SECT_SIZE, 1277 0, 1278 MTIP_INT_CMD_TIMEOUT_MS); 1279 } 1280 1281 /* 1282 * Issue a SMART READ DATA command to the device. 1283 * 1284 * @port pointer to the port structure. 1285 * @buffer pointer to buffer 1286 * @buffer_dma dma address corresponding to @buffer 1287 * 1288 * return value 1289 * @rv return value from mtip_exec_internal_command() 1290 */ 1291 static int mtip_get_smart_data(struct mtip_port *port, u8 *buffer, 1292 dma_addr_t buffer_dma) 1293 { 1294 struct host_to_dev_fis fis; 1295 1296 memset(&fis, 0, sizeof(struct host_to_dev_fis)); 1297 fis.type = 0x27; 1298 fis.opts = 1 << 7; 1299 fis.command = ATA_CMD_SMART; 1300 fis.features = 0xD0; 1301 fis.sect_count = 1; 1302 fis.lba_mid = 0x4F; 1303 fis.lba_hi = 0xC2; 1304 fis.device = ATA_DEVICE_OBS; 1305 1306 return mtip_exec_internal_command(port, 1307 &fis, 1308 5, 1309 buffer_dma, 1310 ATA_SECT_SIZE, 1311 0, 1312 15000); 1313 } 1314 1315 /* 1316 * Get the value of a smart attribute 1317 * 1318 * @port pointer to the port structure 1319 * @id attribute number 1320 * @attrib pointer to return attrib information corresponding to @id 1321 * 1322 * return value 1323 * -EINVAL NULL buffer passed or unsupported attribute @id. 1324 * -EPERM Identify data not valid, SMART not supported or not enabled 1325 */ 1326 static int mtip_get_smart_attr(struct mtip_port *port, unsigned int id, 1327 struct smart_attr *attrib) 1328 { 1329 int rv, i; 1330 struct smart_attr *pattr; 1331 1332 if (!attrib) 1333 return -EINVAL; 1334 1335 if (!port->identify_valid) { 1336 dev_warn(&port->dd->pdev->dev, "IDENTIFY DATA not valid\n"); 1337 return -EPERM; 1338 } 1339 if (!(port->identify[82] & 0x1)) { 1340 dev_warn(&port->dd->pdev->dev, "SMART not supported\n"); 1341 return -EPERM; 1342 } 1343 if (!(port->identify[85] & 0x1)) { 1344 dev_warn(&port->dd->pdev->dev, "SMART not enabled\n"); 1345 return -EPERM; 1346 } 1347 1348 memset(port->smart_buf, 0, ATA_SECT_SIZE); 1349 rv = mtip_get_smart_data(port, port->smart_buf, port->smart_buf_dma); 1350 if (rv) { 1351 dev_warn(&port->dd->pdev->dev, "Failed to ge SMART data\n"); 1352 return rv; 1353 } 1354 1355 pattr = (struct smart_attr *)(port->smart_buf + 2); 1356 for (i = 0; i < 29; i++, pattr++) 1357 if (pattr->attr_id == id) { 1358 memcpy(attrib, pattr, sizeof(struct smart_attr)); 1359 break; 1360 } 1361 1362 if (i == 29) { 1363 dev_warn(&port->dd->pdev->dev, 1364 "Query for invalid SMART attribute ID\n"); 1365 rv = -EINVAL; 1366 } 1367 1368 return rv; 1369 } 1370 1371 /* 1372 * Get the drive capacity. 1373 * 1374 * @dd Pointer to the device data structure. 1375 * @sectors Pointer to the variable that will receive the sector count. 1376 * 1377 * return value 1378 * 1 Capacity was returned successfully. 1379 * 0 The identify information is invalid. 1380 */ 1381 static bool mtip_hw_get_capacity(struct driver_data *dd, sector_t *sectors) 1382 { 1383 struct mtip_port *port = dd->port; 1384 u64 total, raw0, raw1, raw2, raw3; 1385 raw0 = port->identify[100]; 1386 raw1 = port->identify[101]; 1387 raw2 = port->identify[102]; 1388 raw3 = port->identify[103]; 1389 total = raw0 | raw1<<16 | raw2<<32 | raw3<<48; 1390 *sectors = total; 1391 return (bool) !!port->identify_valid; 1392 } 1393 1394 /* 1395 * Display the identify command data. 1396 * 1397 * @port Pointer to the port data structure. 1398 * 1399 * return value 1400 * None 1401 */ 1402 static void mtip_dump_identify(struct mtip_port *port) 1403 { 1404 sector_t sectors; 1405 unsigned short revid; 1406 char cbuf[42]; 1407 1408 if (!port->identify_valid) 1409 return; 1410 1411 strlcpy(cbuf, (char *)(port->identify+10), 21); 1412 dev_info(&port->dd->pdev->dev, 1413 "Serial No.: %s\n", cbuf); 1414 1415 strlcpy(cbuf, (char *)(port->identify+23), 9); 1416 dev_info(&port->dd->pdev->dev, 1417 "Firmware Ver.: %s\n", cbuf); 1418 1419 strlcpy(cbuf, (char *)(port->identify+27), 41); 1420 dev_info(&port->dd->pdev->dev, "Model: %s\n", cbuf); 1421 1422 dev_info(&port->dd->pdev->dev, "Security: %04x %s\n", 1423 port->identify[128], 1424 port->identify[128] & 0x4 ? "(LOCKED)" : ""); 1425 1426 if (mtip_hw_get_capacity(port->dd, §ors)) 1427 dev_info(&port->dd->pdev->dev, 1428 "Capacity: %llu sectors (%llu MB)\n", 1429 (u64)sectors, 1430 ((u64)sectors) * ATA_SECT_SIZE >> 20); 1431 1432 pci_read_config_word(port->dd->pdev, PCI_REVISION_ID, &revid); 1433 switch (revid & 0xFF) { 1434 case 0x1: 1435 strlcpy(cbuf, "A0", 3); 1436 break; 1437 case 0x3: 1438 strlcpy(cbuf, "A2", 3); 1439 break; 1440 default: 1441 strlcpy(cbuf, "?", 2); 1442 break; 1443 } 1444 dev_info(&port->dd->pdev->dev, 1445 "Card Type: %s\n", cbuf); 1446 } 1447 1448 /* 1449 * Map the commands scatter list into the command table. 1450 * 1451 * @command Pointer to the command. 1452 * @nents Number of scatter list entries. 1453 * 1454 * return value 1455 * None 1456 */ 1457 static inline void fill_command_sg(struct driver_data *dd, 1458 struct mtip_cmd *command, 1459 int nents) 1460 { 1461 int n; 1462 unsigned int dma_len; 1463 struct mtip_cmd_sg *command_sg; 1464 struct scatterlist *sg; 1465 1466 command_sg = command->command + AHCI_CMD_TBL_HDR_SZ; 1467 1468 for_each_sg(command->sg, sg, nents, n) { 1469 dma_len = sg_dma_len(sg); 1470 if (dma_len > 0x400000) 1471 dev_err(&dd->pdev->dev, 1472 "DMA segment length truncated\n"); 1473 command_sg->info = cpu_to_le32((dma_len-1) & 0x3FFFFF); 1474 command_sg->dba = cpu_to_le32(sg_dma_address(sg)); 1475 command_sg->dba_upper = 1476 cpu_to_le32((sg_dma_address(sg) >> 16) >> 16); 1477 command_sg++; 1478 } 1479 } 1480 1481 /* 1482 * @brief Execute a drive command. 1483 * 1484 * return value 0 The command completed successfully. 1485 * return value -1 An error occurred while executing the command. 1486 */ 1487 static int exec_drive_task(struct mtip_port *port, u8 *command) 1488 { 1489 struct host_to_dev_fis fis; 1490 struct host_to_dev_fis *reply = (port->rxfis + RX_FIS_D2H_REG); 1491 unsigned int to; 1492 1493 /* Build the FIS. */ 1494 memset(&fis, 0, sizeof(struct host_to_dev_fis)); 1495 fis.type = 0x27; 1496 fis.opts = 1 << 7; 1497 fis.command = command[0]; 1498 fis.features = command[1]; 1499 fis.sect_count = command[2]; 1500 fis.sector = command[3]; 1501 fis.cyl_low = command[4]; 1502 fis.cyl_hi = command[5]; 1503 fis.device = command[6] & ~0x10; /* Clear the dev bit*/ 1504 1505 mtip_set_timeout(port->dd, &fis, &to, 0); 1506 1507 dbg_printk(MTIP_DRV_NAME " %s: User Command: cmd %x, feat %x, nsect %x, sect %x, lcyl %x, hcyl %x, sel %x\n", 1508 __func__, 1509 command[0], 1510 command[1], 1511 command[2], 1512 command[3], 1513 command[4], 1514 command[5], 1515 command[6]); 1516 1517 /* Execute the command. */ 1518 if (mtip_exec_internal_command(port, 1519 &fis, 1520 5, 1521 0, 1522 0, 1523 0, 1524 to) < 0) { 1525 return -1; 1526 } 1527 1528 command[0] = reply->command; /* Status*/ 1529 command[1] = reply->features; /* Error*/ 1530 command[4] = reply->cyl_low; 1531 command[5] = reply->cyl_hi; 1532 1533 dbg_printk(MTIP_DRV_NAME " %s: Completion Status: stat %x, err %x , cyl_lo %x cyl_hi %x\n", 1534 __func__, 1535 command[0], 1536 command[1], 1537 command[4], 1538 command[5]); 1539 1540 return 0; 1541 } 1542 1543 /* 1544 * @brief Execute a drive command. 1545 * 1546 * @param port Pointer to the port data structure. 1547 * @param command Pointer to the user specified command parameters. 1548 * @param user_buffer Pointer to the user space buffer where read sector 1549 * data should be copied. 1550 * 1551 * return value 0 The command completed successfully. 1552 * return value -EFAULT An error occurred while copying the completion 1553 * data to the user space buffer. 1554 * return value -1 An error occurred while executing the command. 1555 */ 1556 static int exec_drive_command(struct mtip_port *port, u8 *command, 1557 void __user *user_buffer) 1558 { 1559 struct host_to_dev_fis fis; 1560 struct host_to_dev_fis *reply; 1561 u8 *buf = NULL; 1562 dma_addr_t dma_addr = 0; 1563 int rv = 0, xfer_sz = command[3]; 1564 unsigned int to; 1565 1566 if (xfer_sz) { 1567 if (!user_buffer) 1568 return -EFAULT; 1569 1570 buf = dma_alloc_coherent(&port->dd->pdev->dev, 1571 ATA_SECT_SIZE * xfer_sz, 1572 &dma_addr, 1573 GFP_KERNEL); 1574 if (!buf) { 1575 dev_err(&port->dd->pdev->dev, 1576 "Memory allocation failed (%d bytes)\n", 1577 ATA_SECT_SIZE * xfer_sz); 1578 return -ENOMEM; 1579 } 1580 } 1581 1582 /* Build the FIS. */ 1583 memset(&fis, 0, sizeof(struct host_to_dev_fis)); 1584 fis.type = 0x27; 1585 fis.opts = 1 << 7; 1586 fis.command = command[0]; 1587 fis.features = command[2]; 1588 fis.sect_count = command[3]; 1589 if (fis.command == ATA_CMD_SMART) { 1590 fis.sector = command[1]; 1591 fis.cyl_low = 0x4F; 1592 fis.cyl_hi = 0xC2; 1593 } 1594 1595 mtip_set_timeout(port->dd, &fis, &to, 0); 1596 1597 if (xfer_sz) 1598 reply = (port->rxfis + RX_FIS_PIO_SETUP); 1599 else 1600 reply = (port->rxfis + RX_FIS_D2H_REG); 1601 1602 dbg_printk(MTIP_DRV_NAME 1603 " %s: User Command: cmd %x, sect %x, " 1604 "feat %x, sectcnt %x\n", 1605 __func__, 1606 command[0], 1607 command[1], 1608 command[2], 1609 command[3]); 1610 1611 /* Execute the command. */ 1612 if (mtip_exec_internal_command(port, 1613 &fis, 1614 5, 1615 (xfer_sz ? dma_addr : 0), 1616 (xfer_sz ? ATA_SECT_SIZE * xfer_sz : 0), 1617 0, 1618 to) 1619 < 0) { 1620 rv = -EFAULT; 1621 goto exit_drive_command; 1622 } 1623 1624 /* Collect the completion status. */ 1625 command[0] = reply->command; /* Status*/ 1626 command[1] = reply->features; /* Error*/ 1627 command[2] = reply->sect_count; 1628 1629 dbg_printk(MTIP_DRV_NAME 1630 " %s: Completion Status: stat %x, " 1631 "err %x, nsect %x\n", 1632 __func__, 1633 command[0], 1634 command[1], 1635 command[2]); 1636 1637 if (xfer_sz) { 1638 if (copy_to_user(user_buffer, 1639 buf, 1640 ATA_SECT_SIZE * command[3])) { 1641 rv = -EFAULT; 1642 goto exit_drive_command; 1643 } 1644 } 1645 exit_drive_command: 1646 if (buf) 1647 dma_free_coherent(&port->dd->pdev->dev, 1648 ATA_SECT_SIZE * xfer_sz, buf, dma_addr); 1649 return rv; 1650 } 1651 1652 /* 1653 * Indicates whether a command has a single sector payload. 1654 * 1655 * @command passed to the device to perform the certain event. 1656 * @features passed to the device to perform the certain event. 1657 * 1658 * return value 1659 * 1 command is one that always has a single sector payload, 1660 * regardless of the value in the Sector Count field. 1661 * 0 otherwise 1662 * 1663 */ 1664 static unsigned int implicit_sector(unsigned char command, 1665 unsigned char features) 1666 { 1667 unsigned int rv = 0; 1668 1669 /* list of commands that have an implicit sector count of 1 */ 1670 switch (command) { 1671 case ATA_CMD_SEC_SET_PASS: 1672 case ATA_CMD_SEC_UNLOCK: 1673 case ATA_CMD_SEC_ERASE_PREP: 1674 case ATA_CMD_SEC_ERASE_UNIT: 1675 case ATA_CMD_SEC_FREEZE_LOCK: 1676 case ATA_CMD_SEC_DISABLE_PASS: 1677 case ATA_CMD_PMP_READ: 1678 case ATA_CMD_PMP_WRITE: 1679 rv = 1; 1680 break; 1681 case ATA_CMD_SET_MAX: 1682 if (features == ATA_SET_MAX_UNLOCK) 1683 rv = 1; 1684 break; 1685 case ATA_CMD_SMART: 1686 if ((features == ATA_SMART_READ_VALUES) || 1687 (features == ATA_SMART_READ_THRESHOLDS)) 1688 rv = 1; 1689 break; 1690 case ATA_CMD_CONF_OVERLAY: 1691 if ((features == ATA_DCO_IDENTIFY) || 1692 (features == ATA_DCO_SET)) 1693 rv = 1; 1694 break; 1695 } 1696 return rv; 1697 } 1698 1699 /* 1700 * Executes a taskfile 1701 * See ide_taskfile_ioctl() for derivation 1702 */ 1703 static int exec_drive_taskfile(struct driver_data *dd, 1704 void __user *buf, 1705 ide_task_request_t *req_task, 1706 int outtotal) 1707 { 1708 struct host_to_dev_fis fis; 1709 struct host_to_dev_fis *reply; 1710 u8 *outbuf = NULL; 1711 u8 *inbuf = NULL; 1712 dma_addr_t outbuf_dma = 0; 1713 dma_addr_t inbuf_dma = 0; 1714 dma_addr_t dma_buffer = 0; 1715 int err = 0; 1716 unsigned int taskin = 0; 1717 unsigned int taskout = 0; 1718 u8 nsect = 0; 1719 unsigned int timeout; 1720 unsigned int force_single_sector; 1721 unsigned int transfer_size; 1722 unsigned long task_file_data; 1723 int intotal = outtotal + req_task->out_size; 1724 int erasemode = 0; 1725 1726 taskout = req_task->out_size; 1727 taskin = req_task->in_size; 1728 /* 130560 = 512 * 0xFF*/ 1729 if (taskin > 130560 || taskout > 130560) 1730 return -EINVAL; 1731 1732 if (taskout) { 1733 outbuf = memdup_user(buf + outtotal, taskout); 1734 if (IS_ERR(outbuf)) 1735 return PTR_ERR(outbuf); 1736 1737 outbuf_dma = dma_map_single(&dd->pdev->dev, outbuf, 1738 taskout, DMA_TO_DEVICE); 1739 if (dma_mapping_error(&dd->pdev->dev, outbuf_dma)) { 1740 err = -ENOMEM; 1741 goto abort; 1742 } 1743 dma_buffer = outbuf_dma; 1744 } 1745 1746 if (taskin) { 1747 inbuf = memdup_user(buf + intotal, taskin); 1748 if (IS_ERR(inbuf)) { 1749 err = PTR_ERR(inbuf); 1750 inbuf = NULL; 1751 goto abort; 1752 } 1753 inbuf_dma = dma_map_single(&dd->pdev->dev, inbuf, 1754 taskin, DMA_FROM_DEVICE); 1755 if (dma_mapping_error(&dd->pdev->dev, inbuf_dma)) { 1756 err = -ENOMEM; 1757 goto abort; 1758 } 1759 dma_buffer = inbuf_dma; 1760 } 1761 1762 /* only supports PIO and non-data commands from this ioctl. */ 1763 switch (req_task->data_phase) { 1764 case TASKFILE_OUT: 1765 nsect = taskout / ATA_SECT_SIZE; 1766 reply = (dd->port->rxfis + RX_FIS_PIO_SETUP); 1767 break; 1768 case TASKFILE_IN: 1769 reply = (dd->port->rxfis + RX_FIS_PIO_SETUP); 1770 break; 1771 case TASKFILE_NO_DATA: 1772 reply = (dd->port->rxfis + RX_FIS_D2H_REG); 1773 break; 1774 default: 1775 err = -EINVAL; 1776 goto abort; 1777 } 1778 1779 /* Build the FIS. */ 1780 memset(&fis, 0, sizeof(struct host_to_dev_fis)); 1781 1782 fis.type = 0x27; 1783 fis.opts = 1 << 7; 1784 fis.command = req_task->io_ports[7]; 1785 fis.features = req_task->io_ports[1]; 1786 fis.sect_count = req_task->io_ports[2]; 1787 fis.lba_low = req_task->io_ports[3]; 1788 fis.lba_mid = req_task->io_ports[4]; 1789 fis.lba_hi = req_task->io_ports[5]; 1790 /* Clear the dev bit*/ 1791 fis.device = req_task->io_ports[6] & ~0x10; 1792 1793 if ((req_task->in_flags.all == 0) && (req_task->out_flags.all & 1)) { 1794 req_task->in_flags.all = 1795 IDE_TASKFILE_STD_IN_FLAGS | 1796 (IDE_HOB_STD_IN_FLAGS << 8); 1797 fis.lba_low_ex = req_task->hob_ports[3]; 1798 fis.lba_mid_ex = req_task->hob_ports[4]; 1799 fis.lba_hi_ex = req_task->hob_ports[5]; 1800 fis.features_ex = req_task->hob_ports[1]; 1801 fis.sect_cnt_ex = req_task->hob_ports[2]; 1802 1803 } else { 1804 req_task->in_flags.all = IDE_TASKFILE_STD_IN_FLAGS; 1805 } 1806 1807 force_single_sector = implicit_sector(fis.command, fis.features); 1808 1809 if ((taskin || taskout) && (!fis.sect_count)) { 1810 if (nsect) 1811 fis.sect_count = nsect; 1812 else { 1813 if (!force_single_sector) { 1814 dev_warn(&dd->pdev->dev, 1815 "data movement but " 1816 "sect_count is 0\n"); 1817 err = -EINVAL; 1818 goto abort; 1819 } 1820 } 1821 } 1822 1823 dbg_printk(MTIP_DRV_NAME 1824 " %s: cmd %x, feat %x, nsect %x," 1825 " sect/lbal %x, lcyl/lbam %x, hcyl/lbah %x," 1826 " head/dev %x\n", 1827 __func__, 1828 fis.command, 1829 fis.features, 1830 fis.sect_count, 1831 fis.lba_low, 1832 fis.lba_mid, 1833 fis.lba_hi, 1834 fis.device); 1835 1836 /* check for erase mode support during secure erase.*/ 1837 if ((fis.command == ATA_CMD_SEC_ERASE_UNIT) && outbuf && 1838 (outbuf[0] & MTIP_SEC_ERASE_MODE)) { 1839 erasemode = 1; 1840 } 1841 1842 mtip_set_timeout(dd, &fis, &timeout, erasemode); 1843 1844 /* Determine the correct transfer size.*/ 1845 if (force_single_sector) 1846 transfer_size = ATA_SECT_SIZE; 1847 else 1848 transfer_size = ATA_SECT_SIZE * fis.sect_count; 1849 1850 /* Execute the command.*/ 1851 if (mtip_exec_internal_command(dd->port, 1852 &fis, 1853 5, 1854 dma_buffer, 1855 transfer_size, 1856 0, 1857 timeout) < 0) { 1858 err = -EIO; 1859 goto abort; 1860 } 1861 1862 task_file_data = readl(dd->port->mmio+PORT_TFDATA); 1863 1864 if ((req_task->data_phase == TASKFILE_IN) && !(task_file_data & 1)) { 1865 reply = dd->port->rxfis + RX_FIS_PIO_SETUP; 1866 req_task->io_ports[7] = reply->control; 1867 } else { 1868 reply = dd->port->rxfis + RX_FIS_D2H_REG; 1869 req_task->io_ports[7] = reply->command; 1870 } 1871 1872 /* reclaim the DMA buffers.*/ 1873 if (inbuf_dma) 1874 dma_unmap_single(&dd->pdev->dev, inbuf_dma, taskin, 1875 DMA_FROM_DEVICE); 1876 if (outbuf_dma) 1877 dma_unmap_single(&dd->pdev->dev, outbuf_dma, taskout, 1878 DMA_TO_DEVICE); 1879 inbuf_dma = 0; 1880 outbuf_dma = 0; 1881 1882 /* return the ATA registers to the caller.*/ 1883 req_task->io_ports[1] = reply->features; 1884 req_task->io_ports[2] = reply->sect_count; 1885 req_task->io_ports[3] = reply->lba_low; 1886 req_task->io_ports[4] = reply->lba_mid; 1887 req_task->io_ports[5] = reply->lba_hi; 1888 req_task->io_ports[6] = reply->device; 1889 1890 if (req_task->out_flags.all & 1) { 1891 1892 req_task->hob_ports[3] = reply->lba_low_ex; 1893 req_task->hob_ports[4] = reply->lba_mid_ex; 1894 req_task->hob_ports[5] = reply->lba_hi_ex; 1895 req_task->hob_ports[1] = reply->features_ex; 1896 req_task->hob_ports[2] = reply->sect_cnt_ex; 1897 } 1898 dbg_printk(MTIP_DRV_NAME 1899 " %s: Completion: stat %x," 1900 "err %x, sect_cnt %x, lbalo %x," 1901 "lbamid %x, lbahi %x, dev %x\n", 1902 __func__, 1903 req_task->io_ports[7], 1904 req_task->io_ports[1], 1905 req_task->io_ports[2], 1906 req_task->io_ports[3], 1907 req_task->io_ports[4], 1908 req_task->io_ports[5], 1909 req_task->io_ports[6]); 1910 1911 if (taskout) { 1912 if (copy_to_user(buf + outtotal, outbuf, taskout)) { 1913 err = -EFAULT; 1914 goto abort; 1915 } 1916 } 1917 if (taskin) { 1918 if (copy_to_user(buf + intotal, inbuf, taskin)) { 1919 err = -EFAULT; 1920 goto abort; 1921 } 1922 } 1923 abort: 1924 if (inbuf_dma) 1925 dma_unmap_single(&dd->pdev->dev, inbuf_dma, taskin, 1926 DMA_FROM_DEVICE); 1927 if (outbuf_dma) 1928 dma_unmap_single(&dd->pdev->dev, outbuf_dma, taskout, 1929 DMA_TO_DEVICE); 1930 kfree(outbuf); 1931 kfree(inbuf); 1932 1933 return err; 1934 } 1935 1936 /* 1937 * Handle IOCTL calls from the Block Layer. 1938 * 1939 * This function is called by the Block Layer when it receives an IOCTL 1940 * command that it does not understand. If the IOCTL command is not supported 1941 * this function returns -ENOTTY. 1942 * 1943 * @dd Pointer to the driver data structure. 1944 * @cmd IOCTL command passed from the Block Layer. 1945 * @arg IOCTL argument passed from the Block Layer. 1946 * 1947 * return value 1948 * 0 The IOCTL completed successfully. 1949 * -ENOTTY The specified command is not supported. 1950 * -EFAULT An error occurred copying data to a user space buffer. 1951 * -EIO An error occurred while executing the command. 1952 */ 1953 static int mtip_hw_ioctl(struct driver_data *dd, unsigned int cmd, 1954 unsigned long arg) 1955 { 1956 switch (cmd) { 1957 case HDIO_GET_IDENTITY: 1958 { 1959 if (copy_to_user((void __user *)arg, dd->port->identify, 1960 sizeof(u16) * ATA_ID_WORDS)) 1961 return -EFAULT; 1962 break; 1963 } 1964 case HDIO_DRIVE_CMD: 1965 { 1966 u8 drive_command[4]; 1967 1968 /* Copy the user command info to our buffer. */ 1969 if (copy_from_user(drive_command, 1970 (void __user *) arg, 1971 sizeof(drive_command))) 1972 return -EFAULT; 1973 1974 /* Execute the drive command. */ 1975 if (exec_drive_command(dd->port, 1976 drive_command, 1977 (void __user *) (arg+4))) 1978 return -EIO; 1979 1980 /* Copy the status back to the users buffer. */ 1981 if (copy_to_user((void __user *) arg, 1982 drive_command, 1983 sizeof(drive_command))) 1984 return -EFAULT; 1985 1986 break; 1987 } 1988 case HDIO_DRIVE_TASK: 1989 { 1990 u8 drive_command[7]; 1991 1992 /* Copy the user command info to our buffer. */ 1993 if (copy_from_user(drive_command, 1994 (void __user *) arg, 1995 sizeof(drive_command))) 1996 return -EFAULT; 1997 1998 /* Execute the drive command. */ 1999 if (exec_drive_task(dd->port, drive_command)) 2000 return -EIO; 2001 2002 /* Copy the status back to the users buffer. */ 2003 if (copy_to_user((void __user *) arg, 2004 drive_command, 2005 sizeof(drive_command))) 2006 return -EFAULT; 2007 2008 break; 2009 } 2010 case HDIO_DRIVE_TASKFILE: { 2011 ide_task_request_t req_task; 2012 int ret, outtotal; 2013 2014 if (copy_from_user(&req_task, (void __user *) arg, 2015 sizeof(req_task))) 2016 return -EFAULT; 2017 2018 outtotal = sizeof(req_task); 2019 2020 ret = exec_drive_taskfile(dd, (void __user *) arg, 2021 &req_task, outtotal); 2022 2023 if (copy_to_user((void __user *) arg, &req_task, 2024 sizeof(req_task))) 2025 return -EFAULT; 2026 2027 return ret; 2028 } 2029 2030 default: 2031 return -EINVAL; 2032 } 2033 return 0; 2034 } 2035 2036 /* 2037 * Submit an IO to the hw 2038 * 2039 * This function is called by the block layer to issue an io 2040 * to the device. Upon completion, the callback function will 2041 * be called with the data parameter passed as the callback data. 2042 * 2043 * @dd Pointer to the driver data structure. 2044 * @start First sector to read. 2045 * @nsect Number of sectors to read. 2046 * @tag The tag of this read command. 2047 * @callback Pointer to the function that should be called 2048 * when the read completes. 2049 * @data Callback data passed to the callback function 2050 * when the read completes. 2051 * @dir Direction (read or write) 2052 * 2053 * return value 2054 * None 2055 */ 2056 static void mtip_hw_submit_io(struct driver_data *dd, struct request *rq, 2057 struct mtip_cmd *command, 2058 struct blk_mq_hw_ctx *hctx) 2059 { 2060 struct mtip_cmd_hdr *hdr = 2061 dd->port->command_list + sizeof(struct mtip_cmd_hdr) * rq->tag; 2062 struct host_to_dev_fis *fis; 2063 struct mtip_port *port = dd->port; 2064 int dma_dir = rq_data_dir(rq) == READ ? DMA_FROM_DEVICE : DMA_TO_DEVICE; 2065 u64 start = blk_rq_pos(rq); 2066 unsigned int nsect = blk_rq_sectors(rq); 2067 unsigned int nents; 2068 2069 /* Map the scatter list for DMA access */ 2070 nents = blk_rq_map_sg(hctx->queue, rq, command->sg); 2071 nents = dma_map_sg(&dd->pdev->dev, command->sg, nents, dma_dir); 2072 2073 prefetch(&port->flags); 2074 2075 command->scatter_ents = nents; 2076 2077 /* 2078 * The number of retries for this command before it is 2079 * reported as a failure to the upper layers. 2080 */ 2081 command->retries = MTIP_MAX_RETRIES; 2082 2083 /* Fill out fis */ 2084 fis = command->command; 2085 fis->type = 0x27; 2086 fis->opts = 1 << 7; 2087 if (dma_dir == DMA_FROM_DEVICE) 2088 fis->command = ATA_CMD_FPDMA_READ; 2089 else 2090 fis->command = ATA_CMD_FPDMA_WRITE; 2091 fis->lba_low = start & 0xFF; 2092 fis->lba_mid = (start >> 8) & 0xFF; 2093 fis->lba_hi = (start >> 16) & 0xFF; 2094 fis->lba_low_ex = (start >> 24) & 0xFF; 2095 fis->lba_mid_ex = (start >> 32) & 0xFF; 2096 fis->lba_hi_ex = (start >> 40) & 0xFF; 2097 fis->device = 1 << 6; 2098 fis->features = nsect & 0xFF; 2099 fis->features_ex = (nsect >> 8) & 0xFF; 2100 fis->sect_count = ((rq->tag << 3) | (rq->tag >> 5)); 2101 fis->sect_cnt_ex = 0; 2102 fis->control = 0; 2103 fis->res2 = 0; 2104 fis->res3 = 0; 2105 fill_command_sg(dd, command, nents); 2106 2107 if (unlikely(command->unaligned)) 2108 fis->device |= 1 << 7; 2109 2110 /* Populate the command header */ 2111 hdr->ctba = cpu_to_le32(command->command_dma & 0xFFFFFFFF); 2112 if (test_bit(MTIP_PF_HOST_CAP_64, &dd->port->flags)) 2113 hdr->ctbau = cpu_to_le32((command->command_dma >> 16) >> 16); 2114 hdr->opts = cpu_to_le32((nents << 16) | 5 | AHCI_CMD_PREFETCH); 2115 hdr->byte_count = 0; 2116 2117 command->direction = dma_dir; 2118 2119 /* 2120 * To prevent this command from being issued 2121 * if an internal command is in progress or error handling is active. 2122 */ 2123 if (unlikely(port->flags & MTIP_PF_PAUSE_IO)) { 2124 set_bit(rq->tag, port->cmds_to_issue); 2125 set_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags); 2126 return; 2127 } 2128 2129 /* Issue the command to the hardware */ 2130 mtip_issue_ncq_command(port, rq->tag); 2131 } 2132 2133 /* 2134 * Sysfs status dump. 2135 * 2136 * @dev Pointer to the device structure, passed by the kernrel. 2137 * @attr Pointer to the device_attribute structure passed by the kernel. 2138 * @buf Pointer to the char buffer that will receive the stats info. 2139 * 2140 * return value 2141 * The size, in bytes, of the data copied into buf. 2142 */ 2143 static ssize_t mtip_hw_show_status(struct device *dev, 2144 struct device_attribute *attr, 2145 char *buf) 2146 { 2147 struct driver_data *dd = dev_to_disk(dev)->private_data; 2148 int size = 0; 2149 2150 if (test_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag)) 2151 size += sprintf(buf, "%s", "thermal_shutdown\n"); 2152 else if (test_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag)) 2153 size += sprintf(buf, "%s", "write_protect\n"); 2154 else 2155 size += sprintf(buf, "%s", "online\n"); 2156 2157 return size; 2158 } 2159 2160 static DEVICE_ATTR(status, 0444, mtip_hw_show_status, NULL); 2161 2162 /* debugsfs entries */ 2163 2164 static ssize_t show_device_status(struct device_driver *drv, char *buf) 2165 { 2166 int size = 0; 2167 struct driver_data *dd, *tmp; 2168 unsigned long flags; 2169 char id_buf[42]; 2170 u16 status = 0; 2171 2172 spin_lock_irqsave(&dev_lock, flags); 2173 size += sprintf(&buf[size], "Devices Present:\n"); 2174 list_for_each_entry_safe(dd, tmp, &online_list, online_list) { 2175 if (dd->pdev) { 2176 if (dd->port && 2177 dd->port->identify && 2178 dd->port->identify_valid) { 2179 strlcpy(id_buf, 2180 (char *) (dd->port->identify + 10), 21); 2181 status = *(dd->port->identify + 141); 2182 } else { 2183 memset(id_buf, 0, 42); 2184 status = 0; 2185 } 2186 2187 if (dd->port && 2188 test_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags)) { 2189 size += sprintf(&buf[size], 2190 " device %s %s (ftl rebuild %d %%)\n", 2191 dev_name(&dd->pdev->dev), 2192 id_buf, 2193 status); 2194 } else { 2195 size += sprintf(&buf[size], 2196 " device %s %s\n", 2197 dev_name(&dd->pdev->dev), 2198 id_buf); 2199 } 2200 } 2201 } 2202 2203 size += sprintf(&buf[size], "Devices Being Removed:\n"); 2204 list_for_each_entry_safe(dd, tmp, &removing_list, remove_list) { 2205 if (dd->pdev) { 2206 if (dd->port && 2207 dd->port->identify && 2208 dd->port->identify_valid) { 2209 strlcpy(id_buf, 2210 (char *) (dd->port->identify+10), 21); 2211 status = *(dd->port->identify + 141); 2212 } else { 2213 memset(id_buf, 0, 42); 2214 status = 0; 2215 } 2216 2217 if (dd->port && 2218 test_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags)) { 2219 size += sprintf(&buf[size], 2220 " device %s %s (ftl rebuild %d %%)\n", 2221 dev_name(&dd->pdev->dev), 2222 id_buf, 2223 status); 2224 } else { 2225 size += sprintf(&buf[size], 2226 " device %s %s\n", 2227 dev_name(&dd->pdev->dev), 2228 id_buf); 2229 } 2230 } 2231 } 2232 spin_unlock_irqrestore(&dev_lock, flags); 2233 2234 return size; 2235 } 2236 2237 static ssize_t mtip_hw_read_device_status(struct file *f, char __user *ubuf, 2238 size_t len, loff_t *offset) 2239 { 2240 struct driver_data *dd = (struct driver_data *)f->private_data; 2241 int size = *offset; 2242 char *buf; 2243 int rv = 0; 2244 2245 if (!len || *offset) 2246 return 0; 2247 2248 buf = kzalloc(MTIP_DFS_MAX_BUF_SIZE, GFP_KERNEL); 2249 if (!buf) { 2250 dev_err(&dd->pdev->dev, 2251 "Memory allocation: status buffer\n"); 2252 return -ENOMEM; 2253 } 2254 2255 size += show_device_status(NULL, buf); 2256 2257 *offset = size <= len ? size : len; 2258 size = copy_to_user(ubuf, buf, *offset); 2259 if (size) 2260 rv = -EFAULT; 2261 2262 kfree(buf); 2263 return rv ? rv : *offset; 2264 } 2265 2266 static ssize_t mtip_hw_read_registers(struct file *f, char __user *ubuf, 2267 size_t len, loff_t *offset) 2268 { 2269 struct driver_data *dd = (struct driver_data *)f->private_data; 2270 char *buf; 2271 u32 group_allocated; 2272 int size = *offset; 2273 int n, rv = 0; 2274 2275 if (!len || size) 2276 return 0; 2277 2278 buf = kzalloc(MTIP_DFS_MAX_BUF_SIZE, GFP_KERNEL); 2279 if (!buf) { 2280 dev_err(&dd->pdev->dev, 2281 "Memory allocation: register buffer\n"); 2282 return -ENOMEM; 2283 } 2284 2285 size += sprintf(&buf[size], "H/ S ACTive : [ 0x"); 2286 2287 for (n = dd->slot_groups-1; n >= 0; n--) 2288 size += sprintf(&buf[size], "%08X ", 2289 readl(dd->port->s_active[n])); 2290 2291 size += sprintf(&buf[size], "]\n"); 2292 size += sprintf(&buf[size], "H/ Command Issue : [ 0x"); 2293 2294 for (n = dd->slot_groups-1; n >= 0; n--) 2295 size += sprintf(&buf[size], "%08X ", 2296 readl(dd->port->cmd_issue[n])); 2297 2298 size += sprintf(&buf[size], "]\n"); 2299 size += sprintf(&buf[size], "H/ Completed : [ 0x"); 2300 2301 for (n = dd->slot_groups-1; n >= 0; n--) 2302 size += sprintf(&buf[size], "%08X ", 2303 readl(dd->port->completed[n])); 2304 2305 size += sprintf(&buf[size], "]\n"); 2306 size += sprintf(&buf[size], "H/ PORT IRQ STAT : [ 0x%08X ]\n", 2307 readl(dd->port->mmio + PORT_IRQ_STAT)); 2308 size += sprintf(&buf[size], "H/ HOST IRQ STAT : [ 0x%08X ]\n", 2309 readl(dd->mmio + HOST_IRQ_STAT)); 2310 size += sprintf(&buf[size], "\n"); 2311 2312 size += sprintf(&buf[size], "L/ Commands in Q : [ 0x"); 2313 2314 for (n = dd->slot_groups-1; n >= 0; n--) { 2315 if (sizeof(long) > sizeof(u32)) 2316 group_allocated = 2317 dd->port->cmds_to_issue[n/2] >> (32*(n&1)); 2318 else 2319 group_allocated = dd->port->cmds_to_issue[n]; 2320 size += sprintf(&buf[size], "%08X ", group_allocated); 2321 } 2322 size += sprintf(&buf[size], "]\n"); 2323 2324 *offset = size <= len ? size : len; 2325 size = copy_to_user(ubuf, buf, *offset); 2326 if (size) 2327 rv = -EFAULT; 2328 2329 kfree(buf); 2330 return rv ? rv : *offset; 2331 } 2332 2333 static ssize_t mtip_hw_read_flags(struct file *f, char __user *ubuf, 2334 size_t len, loff_t *offset) 2335 { 2336 struct driver_data *dd = (struct driver_data *)f->private_data; 2337 char *buf; 2338 int size = *offset; 2339 int rv = 0; 2340 2341 if (!len || size) 2342 return 0; 2343 2344 buf = kzalloc(MTIP_DFS_MAX_BUF_SIZE, GFP_KERNEL); 2345 if (!buf) { 2346 dev_err(&dd->pdev->dev, 2347 "Memory allocation: flag buffer\n"); 2348 return -ENOMEM; 2349 } 2350 2351 size += sprintf(&buf[size], "Flag-port : [ %08lX ]\n", 2352 dd->port->flags); 2353 size += sprintf(&buf[size], "Flag-dd : [ %08lX ]\n", 2354 dd->dd_flag); 2355 2356 *offset = size <= len ? size : len; 2357 size = copy_to_user(ubuf, buf, *offset); 2358 if (size) 2359 rv = -EFAULT; 2360 2361 kfree(buf); 2362 return rv ? rv : *offset; 2363 } 2364 2365 static const struct file_operations mtip_device_status_fops = { 2366 .owner = THIS_MODULE, 2367 .open = simple_open, 2368 .read = mtip_hw_read_device_status, 2369 .llseek = no_llseek, 2370 }; 2371 2372 static const struct file_operations mtip_regs_fops = { 2373 .owner = THIS_MODULE, 2374 .open = simple_open, 2375 .read = mtip_hw_read_registers, 2376 .llseek = no_llseek, 2377 }; 2378 2379 static const struct file_operations mtip_flags_fops = { 2380 .owner = THIS_MODULE, 2381 .open = simple_open, 2382 .read = mtip_hw_read_flags, 2383 .llseek = no_llseek, 2384 }; 2385 2386 /* 2387 * Create the sysfs related attributes. 2388 * 2389 * @dd Pointer to the driver data structure. 2390 * @kobj Pointer to the kobj for the block device. 2391 * 2392 * return value 2393 * 0 Operation completed successfully. 2394 * -EINVAL Invalid parameter. 2395 */ 2396 static int mtip_hw_sysfs_init(struct driver_data *dd, struct kobject *kobj) 2397 { 2398 if (!kobj || !dd) 2399 return -EINVAL; 2400 2401 if (sysfs_create_file(kobj, &dev_attr_status.attr)) 2402 dev_warn(&dd->pdev->dev, 2403 "Error creating 'status' sysfs entry\n"); 2404 return 0; 2405 } 2406 2407 /* 2408 * Remove the sysfs related attributes. 2409 * 2410 * @dd Pointer to the driver data structure. 2411 * @kobj Pointer to the kobj for the block device. 2412 * 2413 * return value 2414 * 0 Operation completed successfully. 2415 * -EINVAL Invalid parameter. 2416 */ 2417 static int mtip_hw_sysfs_exit(struct driver_data *dd, struct kobject *kobj) 2418 { 2419 if (!kobj || !dd) 2420 return -EINVAL; 2421 2422 sysfs_remove_file(kobj, &dev_attr_status.attr); 2423 2424 return 0; 2425 } 2426 2427 static int mtip_hw_debugfs_init(struct driver_data *dd) 2428 { 2429 if (!dfs_parent) 2430 return -1; 2431 2432 dd->dfs_node = debugfs_create_dir(dd->disk->disk_name, dfs_parent); 2433 if (IS_ERR_OR_NULL(dd->dfs_node)) { 2434 dev_warn(&dd->pdev->dev, 2435 "Error creating node %s under debugfs\n", 2436 dd->disk->disk_name); 2437 dd->dfs_node = NULL; 2438 return -1; 2439 } 2440 2441 debugfs_create_file("flags", 0444, dd->dfs_node, dd, &mtip_flags_fops); 2442 debugfs_create_file("registers", 0444, dd->dfs_node, dd, 2443 &mtip_regs_fops); 2444 2445 return 0; 2446 } 2447 2448 static void mtip_hw_debugfs_exit(struct driver_data *dd) 2449 { 2450 debugfs_remove_recursive(dd->dfs_node); 2451 } 2452 2453 /* 2454 * Perform any init/resume time hardware setup 2455 * 2456 * @dd Pointer to the driver data structure. 2457 * 2458 * return value 2459 * None 2460 */ 2461 static inline void hba_setup(struct driver_data *dd) 2462 { 2463 u32 hwdata; 2464 hwdata = readl(dd->mmio + HOST_HSORG); 2465 2466 /* interrupt bug workaround: use only 1 IS bit.*/ 2467 writel(hwdata | 2468 HSORG_DISABLE_SLOTGRP_INTR | 2469 HSORG_DISABLE_SLOTGRP_PXIS, 2470 dd->mmio + HOST_HSORG); 2471 } 2472 2473 static int mtip_device_unaligned_constrained(struct driver_data *dd) 2474 { 2475 return (dd->pdev->device == P420M_DEVICE_ID ? 1 : 0); 2476 } 2477 2478 /* 2479 * Detect the details of the product, and store anything needed 2480 * into the driver data structure. This includes product type and 2481 * version and number of slot groups. 2482 * 2483 * @dd Pointer to the driver data structure. 2484 * 2485 * return value 2486 * None 2487 */ 2488 static void mtip_detect_product(struct driver_data *dd) 2489 { 2490 u32 hwdata; 2491 unsigned int rev, slotgroups; 2492 2493 /* 2494 * HBA base + 0xFC [15:0] - vendor-specific hardware interface 2495 * info register: 2496 * [15:8] hardware/software interface rev# 2497 * [ 3] asic-style interface 2498 * [ 2:0] number of slot groups, minus 1 (only valid for asic-style). 2499 */ 2500 hwdata = readl(dd->mmio + HOST_HSORG); 2501 2502 dd->product_type = MTIP_PRODUCT_UNKNOWN; 2503 dd->slot_groups = 1; 2504 2505 if (hwdata & 0x8) { 2506 dd->product_type = MTIP_PRODUCT_ASICFPGA; 2507 rev = (hwdata & HSORG_HWREV) >> 8; 2508 slotgroups = (hwdata & HSORG_SLOTGROUPS) + 1; 2509 dev_info(&dd->pdev->dev, 2510 "ASIC-FPGA design, HS rev 0x%x, " 2511 "%i slot groups [%i slots]\n", 2512 rev, 2513 slotgroups, 2514 slotgroups * 32); 2515 2516 if (slotgroups > MTIP_MAX_SLOT_GROUPS) { 2517 dev_warn(&dd->pdev->dev, 2518 "Warning: driver only supports " 2519 "%i slot groups.\n", MTIP_MAX_SLOT_GROUPS); 2520 slotgroups = MTIP_MAX_SLOT_GROUPS; 2521 } 2522 dd->slot_groups = slotgroups; 2523 return; 2524 } 2525 2526 dev_warn(&dd->pdev->dev, "Unrecognized product id\n"); 2527 } 2528 2529 /* 2530 * Blocking wait for FTL rebuild to complete 2531 * 2532 * @dd Pointer to the DRIVER_DATA structure. 2533 * 2534 * return value 2535 * 0 FTL rebuild completed successfully 2536 * -EFAULT FTL rebuild error/timeout/interruption 2537 */ 2538 static int mtip_ftl_rebuild_poll(struct driver_data *dd) 2539 { 2540 unsigned long timeout, cnt = 0, start; 2541 2542 dev_warn(&dd->pdev->dev, 2543 "FTL rebuild in progress. Polling for completion.\n"); 2544 2545 start = jiffies; 2546 timeout = jiffies + msecs_to_jiffies(MTIP_FTL_REBUILD_TIMEOUT_MS); 2547 2548 do { 2549 if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, 2550 &dd->dd_flag))) 2551 return -EFAULT; 2552 if (mtip_check_surprise_removal(dd->pdev)) 2553 return -EFAULT; 2554 2555 if (mtip_get_identify(dd->port, NULL) < 0) 2556 return -EFAULT; 2557 2558 if (*(dd->port->identify + MTIP_FTL_REBUILD_OFFSET) == 2559 MTIP_FTL_REBUILD_MAGIC) { 2560 ssleep(1); 2561 /* Print message every 3 minutes */ 2562 if (cnt++ >= 180) { 2563 dev_warn(&dd->pdev->dev, 2564 "FTL rebuild in progress (%d secs).\n", 2565 jiffies_to_msecs(jiffies - start) / 1000); 2566 cnt = 0; 2567 } 2568 } else { 2569 dev_warn(&dd->pdev->dev, 2570 "FTL rebuild complete (%d secs).\n", 2571 jiffies_to_msecs(jiffies - start) / 1000); 2572 mtip_block_initialize(dd); 2573 return 0; 2574 } 2575 } while (time_before(jiffies, timeout)); 2576 2577 /* Check for timeout */ 2578 dev_err(&dd->pdev->dev, 2579 "Timed out waiting for FTL rebuild to complete (%d secs).\n", 2580 jiffies_to_msecs(jiffies - start) / 1000); 2581 return -EFAULT; 2582 } 2583 2584 static void mtip_softirq_done_fn(struct request *rq) 2585 { 2586 struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq); 2587 struct driver_data *dd = rq->q->queuedata; 2588 2589 /* Unmap the DMA scatter list entries */ 2590 dma_unmap_sg(&dd->pdev->dev, cmd->sg, cmd->scatter_ents, 2591 cmd->direction); 2592 2593 if (unlikely(cmd->unaligned)) 2594 atomic_inc(&dd->port->cmd_slot_unal); 2595 2596 blk_mq_end_request(rq, cmd->status); 2597 } 2598 2599 static bool mtip_abort_cmd(struct request *req, void *data, bool reserved) 2600 { 2601 struct mtip_cmd *cmd = blk_mq_rq_to_pdu(req); 2602 struct driver_data *dd = data; 2603 2604 dbg_printk(MTIP_DRV_NAME " Aborting request, tag = %d\n", req->tag); 2605 2606 clear_bit(req->tag, dd->port->cmds_to_issue); 2607 cmd->status = BLK_STS_IOERR; 2608 mtip_softirq_done_fn(req); 2609 return true; 2610 } 2611 2612 static bool mtip_queue_cmd(struct request *req, void *data, bool reserved) 2613 { 2614 struct driver_data *dd = data; 2615 2616 set_bit(req->tag, dd->port->cmds_to_issue); 2617 blk_abort_request(req); 2618 return true; 2619 } 2620 2621 /* 2622 * service thread to issue queued commands 2623 * 2624 * @data Pointer to the driver data structure. 2625 * 2626 * return value 2627 * 0 2628 */ 2629 2630 static int mtip_service_thread(void *data) 2631 { 2632 struct driver_data *dd = (struct driver_data *)data; 2633 unsigned long slot, slot_start, slot_wrap, to; 2634 unsigned int num_cmd_slots = dd->slot_groups * 32; 2635 struct mtip_port *port = dd->port; 2636 2637 while (1) { 2638 if (kthread_should_stop() || 2639 test_bit(MTIP_PF_SVC_THD_STOP_BIT, &port->flags)) 2640 goto st_out; 2641 clear_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags); 2642 2643 /* 2644 * the condition is to check neither an internal command is 2645 * is in progress nor error handling is active 2646 */ 2647 wait_event_interruptible(port->svc_wait, (port->flags) && 2648 (port->flags & MTIP_PF_SVC_THD_WORK)); 2649 2650 if (kthread_should_stop() || 2651 test_bit(MTIP_PF_SVC_THD_STOP_BIT, &port->flags)) 2652 goto st_out; 2653 2654 if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, 2655 &dd->dd_flag))) 2656 goto st_out; 2657 2658 set_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags); 2659 2660 restart_eh: 2661 /* Demux bits: start with error handling */ 2662 if (test_bit(MTIP_PF_EH_ACTIVE_BIT, &port->flags)) { 2663 mtip_handle_tfe(dd); 2664 clear_bit(MTIP_PF_EH_ACTIVE_BIT, &port->flags); 2665 } 2666 2667 if (test_bit(MTIP_PF_EH_ACTIVE_BIT, &port->flags)) 2668 goto restart_eh; 2669 2670 if (test_bit(MTIP_PF_TO_ACTIVE_BIT, &port->flags)) { 2671 to = jiffies + msecs_to_jiffies(5000); 2672 2673 do { 2674 mdelay(100); 2675 } while (atomic_read(&dd->irq_workers_active) != 0 && 2676 time_before(jiffies, to)); 2677 2678 if (atomic_read(&dd->irq_workers_active) != 0) 2679 dev_warn(&dd->pdev->dev, 2680 "Completion workers still active!"); 2681 2682 blk_mq_quiesce_queue(dd->queue); 2683 2684 blk_mq_tagset_busy_iter(&dd->tags, mtip_queue_cmd, dd); 2685 2686 set_bit(MTIP_PF_ISSUE_CMDS_BIT, &dd->port->flags); 2687 2688 if (mtip_device_reset(dd)) 2689 blk_mq_tagset_busy_iter(&dd->tags, 2690 mtip_abort_cmd, dd); 2691 2692 clear_bit(MTIP_PF_TO_ACTIVE_BIT, &dd->port->flags); 2693 2694 blk_mq_unquiesce_queue(dd->queue); 2695 } 2696 2697 if (test_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags)) { 2698 slot = 1; 2699 /* used to restrict the loop to one iteration */ 2700 slot_start = num_cmd_slots; 2701 slot_wrap = 0; 2702 while (1) { 2703 slot = find_next_bit(port->cmds_to_issue, 2704 num_cmd_slots, slot); 2705 if (slot_wrap == 1) { 2706 if ((slot_start >= slot) || 2707 (slot >= num_cmd_slots)) 2708 break; 2709 } 2710 if (unlikely(slot_start == num_cmd_slots)) 2711 slot_start = slot; 2712 2713 if (unlikely(slot == num_cmd_slots)) { 2714 slot = 1; 2715 slot_wrap = 1; 2716 continue; 2717 } 2718 2719 /* Issue the command to the hardware */ 2720 mtip_issue_ncq_command(port, slot); 2721 2722 clear_bit(slot, port->cmds_to_issue); 2723 } 2724 2725 clear_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags); 2726 } 2727 2728 if (test_bit(MTIP_PF_REBUILD_BIT, &port->flags)) { 2729 if (mtip_ftl_rebuild_poll(dd) == 0) 2730 clear_bit(MTIP_PF_REBUILD_BIT, &port->flags); 2731 } 2732 } 2733 2734 st_out: 2735 return 0; 2736 } 2737 2738 /* 2739 * DMA region teardown 2740 * 2741 * @dd Pointer to driver_data structure 2742 * 2743 * return value 2744 * None 2745 */ 2746 static void mtip_dma_free(struct driver_data *dd) 2747 { 2748 struct mtip_port *port = dd->port; 2749 2750 if (port->block1) 2751 dma_free_coherent(&dd->pdev->dev, BLOCK_DMA_ALLOC_SZ, 2752 port->block1, port->block1_dma); 2753 2754 if (port->command_list) { 2755 dma_free_coherent(&dd->pdev->dev, AHCI_CMD_TBL_SZ, 2756 port->command_list, port->command_list_dma); 2757 } 2758 } 2759 2760 /* 2761 * DMA region setup 2762 * 2763 * @dd Pointer to driver_data structure 2764 * 2765 * return value 2766 * -ENOMEM Not enough free DMA region space to initialize driver 2767 */ 2768 static int mtip_dma_alloc(struct driver_data *dd) 2769 { 2770 struct mtip_port *port = dd->port; 2771 2772 /* Allocate dma memory for RX Fis, Identify, and Sector Bufffer */ 2773 port->block1 = 2774 dma_alloc_coherent(&dd->pdev->dev, BLOCK_DMA_ALLOC_SZ, 2775 &port->block1_dma, GFP_KERNEL); 2776 if (!port->block1) 2777 return -ENOMEM; 2778 2779 /* Allocate dma memory for command list */ 2780 port->command_list = 2781 dma_alloc_coherent(&dd->pdev->dev, AHCI_CMD_TBL_SZ, 2782 &port->command_list_dma, GFP_KERNEL); 2783 if (!port->command_list) { 2784 dma_free_coherent(&dd->pdev->dev, BLOCK_DMA_ALLOC_SZ, 2785 port->block1, port->block1_dma); 2786 port->block1 = NULL; 2787 port->block1_dma = 0; 2788 return -ENOMEM; 2789 } 2790 2791 /* Setup all pointers into first DMA region */ 2792 port->rxfis = port->block1 + AHCI_RX_FIS_OFFSET; 2793 port->rxfis_dma = port->block1_dma + AHCI_RX_FIS_OFFSET; 2794 port->identify = port->block1 + AHCI_IDFY_OFFSET; 2795 port->identify_dma = port->block1_dma + AHCI_IDFY_OFFSET; 2796 port->log_buf = port->block1 + AHCI_SECTBUF_OFFSET; 2797 port->log_buf_dma = port->block1_dma + AHCI_SECTBUF_OFFSET; 2798 port->smart_buf = port->block1 + AHCI_SMARTBUF_OFFSET; 2799 port->smart_buf_dma = port->block1_dma + AHCI_SMARTBUF_OFFSET; 2800 2801 return 0; 2802 } 2803 2804 static int mtip_hw_get_identify(struct driver_data *dd) 2805 { 2806 struct smart_attr attr242; 2807 unsigned char *buf; 2808 int rv; 2809 2810 if (mtip_get_identify(dd->port, NULL) < 0) 2811 return -EFAULT; 2812 2813 if (*(dd->port->identify + MTIP_FTL_REBUILD_OFFSET) == 2814 MTIP_FTL_REBUILD_MAGIC) { 2815 set_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags); 2816 return MTIP_FTL_REBUILD_MAGIC; 2817 } 2818 mtip_dump_identify(dd->port); 2819 2820 /* check write protect, over temp and rebuild statuses */ 2821 rv = mtip_read_log_page(dd->port, ATA_LOG_SATA_NCQ, 2822 dd->port->log_buf, 2823 dd->port->log_buf_dma, 1); 2824 if (rv) { 2825 dev_warn(&dd->pdev->dev, 2826 "Error in READ LOG EXT (10h) command\n"); 2827 /* non-critical error, don't fail the load */ 2828 } else { 2829 buf = (unsigned char *)dd->port->log_buf; 2830 if (buf[259] & 0x1) { 2831 dev_info(&dd->pdev->dev, 2832 "Write protect bit is set.\n"); 2833 set_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag); 2834 } 2835 if (buf[288] == 0xF7) { 2836 dev_info(&dd->pdev->dev, 2837 "Exceeded Tmax, drive in thermal shutdown.\n"); 2838 set_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag); 2839 } 2840 if (buf[288] == 0xBF) { 2841 dev_info(&dd->pdev->dev, 2842 "Drive indicates rebuild has failed.\n"); 2843 set_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag); 2844 } 2845 } 2846 2847 /* get write protect progess */ 2848 memset(&attr242, 0, sizeof(struct smart_attr)); 2849 if (mtip_get_smart_attr(dd->port, 242, &attr242)) 2850 dev_warn(&dd->pdev->dev, 2851 "Unable to check write protect progress\n"); 2852 else 2853 dev_info(&dd->pdev->dev, 2854 "Write protect progress: %u%% (%u blocks)\n", 2855 attr242.cur, le32_to_cpu(attr242.data)); 2856 2857 return rv; 2858 } 2859 2860 /* 2861 * Called once for each card. 2862 * 2863 * @dd Pointer to the driver data structure. 2864 * 2865 * return value 2866 * 0 on success, else an error code. 2867 */ 2868 static int mtip_hw_init(struct driver_data *dd) 2869 { 2870 int i; 2871 int rv; 2872 unsigned long timeout, timetaken; 2873 2874 dd->mmio = pcim_iomap_table(dd->pdev)[MTIP_ABAR]; 2875 2876 mtip_detect_product(dd); 2877 if (dd->product_type == MTIP_PRODUCT_UNKNOWN) { 2878 rv = -EIO; 2879 goto out1; 2880 } 2881 2882 hba_setup(dd); 2883 2884 dd->port = kzalloc_node(sizeof(struct mtip_port), GFP_KERNEL, 2885 dd->numa_node); 2886 if (!dd->port) { 2887 dev_err(&dd->pdev->dev, 2888 "Memory allocation: port structure\n"); 2889 return -ENOMEM; 2890 } 2891 2892 /* Continue workqueue setup */ 2893 for (i = 0; i < MTIP_MAX_SLOT_GROUPS; i++) 2894 dd->work[i].port = dd->port; 2895 2896 /* Enable unaligned IO constraints for some devices */ 2897 if (mtip_device_unaligned_constrained(dd)) 2898 dd->unal_qdepth = MTIP_MAX_UNALIGNED_SLOTS; 2899 else 2900 dd->unal_qdepth = 0; 2901 2902 atomic_set(&dd->port->cmd_slot_unal, dd->unal_qdepth); 2903 2904 /* Spinlock to prevent concurrent issue */ 2905 for (i = 0; i < MTIP_MAX_SLOT_GROUPS; i++) 2906 spin_lock_init(&dd->port->cmd_issue_lock[i]); 2907 2908 /* Set the port mmio base address. */ 2909 dd->port->mmio = dd->mmio + PORT_OFFSET; 2910 dd->port->dd = dd; 2911 2912 /* DMA allocations */ 2913 rv = mtip_dma_alloc(dd); 2914 if (rv < 0) 2915 goto out1; 2916 2917 /* Setup the pointers to the extended s_active and CI registers. */ 2918 for (i = 0; i < dd->slot_groups; i++) { 2919 dd->port->s_active[i] = 2920 dd->port->mmio + i*0x80 + PORT_SCR_ACT; 2921 dd->port->cmd_issue[i] = 2922 dd->port->mmio + i*0x80 + PORT_COMMAND_ISSUE; 2923 dd->port->completed[i] = 2924 dd->port->mmio + i*0x80 + PORT_SDBV; 2925 } 2926 2927 timetaken = jiffies; 2928 timeout = jiffies + msecs_to_jiffies(30000); 2929 while (((readl(dd->port->mmio + PORT_SCR_STAT) & 0x0F) != 0x03) && 2930 time_before(jiffies, timeout)) { 2931 mdelay(100); 2932 } 2933 if (unlikely(mtip_check_surprise_removal(dd->pdev))) { 2934 timetaken = jiffies - timetaken; 2935 dev_warn(&dd->pdev->dev, 2936 "Surprise removal detected at %u ms\n", 2937 jiffies_to_msecs(timetaken)); 2938 rv = -ENODEV; 2939 goto out2 ; 2940 } 2941 if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag))) { 2942 timetaken = jiffies - timetaken; 2943 dev_warn(&dd->pdev->dev, 2944 "Removal detected at %u ms\n", 2945 jiffies_to_msecs(timetaken)); 2946 rv = -EFAULT; 2947 goto out2; 2948 } 2949 2950 /* Conditionally reset the HBA. */ 2951 if (!(readl(dd->mmio + HOST_CAP) & HOST_CAP_NZDMA)) { 2952 if (mtip_hba_reset(dd) < 0) { 2953 dev_err(&dd->pdev->dev, 2954 "Card did not reset within timeout\n"); 2955 rv = -EIO; 2956 goto out2; 2957 } 2958 } else { 2959 /* Clear any pending interrupts on the HBA */ 2960 writel(readl(dd->mmio + HOST_IRQ_STAT), 2961 dd->mmio + HOST_IRQ_STAT); 2962 } 2963 2964 mtip_init_port(dd->port); 2965 mtip_start_port(dd->port); 2966 2967 /* Setup the ISR and enable interrupts. */ 2968 rv = request_irq(dd->pdev->irq, mtip_irq_handler, IRQF_SHARED, 2969 dev_driver_string(&dd->pdev->dev), dd); 2970 if (rv) { 2971 dev_err(&dd->pdev->dev, 2972 "Unable to allocate IRQ %d\n", dd->pdev->irq); 2973 goto out2; 2974 } 2975 irq_set_affinity_hint(dd->pdev->irq, get_cpu_mask(dd->isr_binding)); 2976 2977 /* Enable interrupts on the HBA. */ 2978 writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN, 2979 dd->mmio + HOST_CTL); 2980 2981 init_waitqueue_head(&dd->port->svc_wait); 2982 2983 if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)) { 2984 rv = -EFAULT; 2985 goto out3; 2986 } 2987 2988 return rv; 2989 2990 out3: 2991 /* Disable interrupts on the HBA. */ 2992 writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN, 2993 dd->mmio + HOST_CTL); 2994 2995 /* Release the IRQ. */ 2996 irq_set_affinity_hint(dd->pdev->irq, NULL); 2997 free_irq(dd->pdev->irq, dd); 2998 2999 out2: 3000 mtip_deinit_port(dd->port); 3001 mtip_dma_free(dd); 3002 3003 out1: 3004 /* Free the memory allocated for the for structure. */ 3005 kfree(dd->port); 3006 3007 return rv; 3008 } 3009 3010 static int mtip_standby_drive(struct driver_data *dd) 3011 { 3012 int rv = 0; 3013 3014 if (dd->sr || !dd->port) 3015 return -ENODEV; 3016 /* 3017 * Send standby immediate (E0h) to the drive so that it 3018 * saves its state. 3019 */ 3020 if (!test_bit(MTIP_PF_REBUILD_BIT, &dd->port->flags) && 3021 !test_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag) && 3022 !test_bit(MTIP_DDF_SEC_LOCK_BIT, &dd->dd_flag)) { 3023 rv = mtip_standby_immediate(dd->port); 3024 if (rv) 3025 dev_warn(&dd->pdev->dev, 3026 "STANDBY IMMEDIATE failed\n"); 3027 } 3028 return rv; 3029 } 3030 3031 /* 3032 * Called to deinitialize an interface. 3033 * 3034 * @dd Pointer to the driver data structure. 3035 * 3036 * return value 3037 * 0 3038 */ 3039 static int mtip_hw_exit(struct driver_data *dd) 3040 { 3041 if (!dd->sr) { 3042 /* de-initialize the port. */ 3043 mtip_deinit_port(dd->port); 3044 3045 /* Disable interrupts on the HBA. */ 3046 writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN, 3047 dd->mmio + HOST_CTL); 3048 } 3049 3050 /* Release the IRQ. */ 3051 irq_set_affinity_hint(dd->pdev->irq, NULL); 3052 free_irq(dd->pdev->irq, dd); 3053 msleep(1000); 3054 3055 /* Free dma regions */ 3056 mtip_dma_free(dd); 3057 3058 /* Free the memory allocated for the for structure. */ 3059 kfree(dd->port); 3060 dd->port = NULL; 3061 3062 return 0; 3063 } 3064 3065 /* 3066 * Issue a Standby Immediate command to the device. 3067 * 3068 * This function is called by the Block Layer just before the 3069 * system powers off during a shutdown. 3070 * 3071 * @dd Pointer to the driver data structure. 3072 * 3073 * return value 3074 * 0 3075 */ 3076 static int mtip_hw_shutdown(struct driver_data *dd) 3077 { 3078 /* 3079 * Send standby immediate (E0h) to the drive so that it 3080 * saves its state. 3081 */ 3082 mtip_standby_drive(dd); 3083 3084 return 0; 3085 } 3086 3087 /* 3088 * Suspend function 3089 * 3090 * This function is called by the Block Layer just before the 3091 * system hibernates. 3092 * 3093 * @dd Pointer to the driver data structure. 3094 * 3095 * return value 3096 * 0 Suspend was successful 3097 * -EFAULT Suspend was not successful 3098 */ 3099 static int mtip_hw_suspend(struct driver_data *dd) 3100 { 3101 /* 3102 * Send standby immediate (E0h) to the drive 3103 * so that it saves its state. 3104 */ 3105 if (mtip_standby_drive(dd) != 0) { 3106 dev_err(&dd->pdev->dev, 3107 "Failed standby-immediate command\n"); 3108 return -EFAULT; 3109 } 3110 3111 /* Disable interrupts on the HBA.*/ 3112 writel(readl(dd->mmio + HOST_CTL) & ~HOST_IRQ_EN, 3113 dd->mmio + HOST_CTL); 3114 mtip_deinit_port(dd->port); 3115 3116 return 0; 3117 } 3118 3119 /* 3120 * Resume function 3121 * 3122 * This function is called by the Block Layer as the 3123 * system resumes. 3124 * 3125 * @dd Pointer to the driver data structure. 3126 * 3127 * return value 3128 * 0 Resume was successful 3129 * -EFAULT Resume was not successful 3130 */ 3131 static int mtip_hw_resume(struct driver_data *dd) 3132 { 3133 /* Perform any needed hardware setup steps */ 3134 hba_setup(dd); 3135 3136 /* Reset the HBA */ 3137 if (mtip_hba_reset(dd) != 0) { 3138 dev_err(&dd->pdev->dev, 3139 "Unable to reset the HBA\n"); 3140 return -EFAULT; 3141 } 3142 3143 /* 3144 * Enable the port, DMA engine, and FIS reception specific 3145 * h/w in controller. 3146 */ 3147 mtip_init_port(dd->port); 3148 mtip_start_port(dd->port); 3149 3150 /* Enable interrupts on the HBA.*/ 3151 writel(readl(dd->mmio + HOST_CTL) | HOST_IRQ_EN, 3152 dd->mmio + HOST_CTL); 3153 3154 return 0; 3155 } 3156 3157 /* 3158 * Helper function for reusing disk name 3159 * upon hot insertion. 3160 */ 3161 static int rssd_disk_name_format(char *prefix, 3162 int index, 3163 char *buf, 3164 int buflen) 3165 { 3166 const int base = 'z' - 'a' + 1; 3167 char *begin = buf + strlen(prefix); 3168 char *end = buf + buflen; 3169 char *p; 3170 int unit; 3171 3172 p = end - 1; 3173 *p = '\0'; 3174 unit = base; 3175 do { 3176 if (p == begin) 3177 return -EINVAL; 3178 *--p = 'a' + (index % unit); 3179 index = (index / unit) - 1; 3180 } while (index >= 0); 3181 3182 memmove(begin, p, end - p); 3183 memcpy(buf, prefix, strlen(prefix)); 3184 3185 return 0; 3186 } 3187 3188 /* 3189 * Block layer IOCTL handler. 3190 * 3191 * @dev Pointer to the block_device structure. 3192 * @mode ignored 3193 * @cmd IOCTL command passed from the user application. 3194 * @arg Argument passed from the user application. 3195 * 3196 * return value 3197 * 0 IOCTL completed successfully. 3198 * -ENOTTY IOCTL not supported or invalid driver data 3199 * structure pointer. 3200 */ 3201 static int mtip_block_ioctl(struct block_device *dev, 3202 fmode_t mode, 3203 unsigned cmd, 3204 unsigned long arg) 3205 { 3206 struct driver_data *dd = dev->bd_disk->private_data; 3207 3208 if (!capable(CAP_SYS_ADMIN)) 3209 return -EACCES; 3210 3211 if (!dd) 3212 return -ENOTTY; 3213 3214 if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag))) 3215 return -ENOTTY; 3216 3217 switch (cmd) { 3218 case BLKFLSBUF: 3219 return -ENOTTY; 3220 default: 3221 return mtip_hw_ioctl(dd, cmd, arg); 3222 } 3223 } 3224 3225 #ifdef CONFIG_COMPAT 3226 /* 3227 * Block layer compat IOCTL handler. 3228 * 3229 * @dev Pointer to the block_device structure. 3230 * @mode ignored 3231 * @cmd IOCTL command passed from the user application. 3232 * @arg Argument passed from the user application. 3233 * 3234 * return value 3235 * 0 IOCTL completed successfully. 3236 * -ENOTTY IOCTL not supported or invalid driver data 3237 * structure pointer. 3238 */ 3239 static int mtip_block_compat_ioctl(struct block_device *dev, 3240 fmode_t mode, 3241 unsigned cmd, 3242 unsigned long arg) 3243 { 3244 struct driver_data *dd = dev->bd_disk->private_data; 3245 3246 if (!capable(CAP_SYS_ADMIN)) 3247 return -EACCES; 3248 3249 if (!dd) 3250 return -ENOTTY; 3251 3252 if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag))) 3253 return -ENOTTY; 3254 3255 switch (cmd) { 3256 case BLKFLSBUF: 3257 return -ENOTTY; 3258 case HDIO_DRIVE_TASKFILE: { 3259 struct mtip_compat_ide_task_request_s __user *compat_req_task; 3260 ide_task_request_t req_task; 3261 int compat_tasksize, outtotal, ret; 3262 3263 compat_tasksize = 3264 sizeof(struct mtip_compat_ide_task_request_s); 3265 3266 compat_req_task = 3267 (struct mtip_compat_ide_task_request_s __user *) arg; 3268 3269 if (copy_from_user(&req_task, (void __user *) arg, 3270 compat_tasksize - (2 * sizeof(compat_long_t)))) 3271 return -EFAULT; 3272 3273 if (get_user(req_task.out_size, &compat_req_task->out_size)) 3274 return -EFAULT; 3275 3276 if (get_user(req_task.in_size, &compat_req_task->in_size)) 3277 return -EFAULT; 3278 3279 outtotal = sizeof(struct mtip_compat_ide_task_request_s); 3280 3281 ret = exec_drive_taskfile(dd, (void __user *) arg, 3282 &req_task, outtotal); 3283 3284 if (copy_to_user((void __user *) arg, &req_task, 3285 compat_tasksize - 3286 (2 * sizeof(compat_long_t)))) 3287 return -EFAULT; 3288 3289 if (put_user(req_task.out_size, &compat_req_task->out_size)) 3290 return -EFAULT; 3291 3292 if (put_user(req_task.in_size, &compat_req_task->in_size)) 3293 return -EFAULT; 3294 3295 return ret; 3296 } 3297 default: 3298 return mtip_hw_ioctl(dd, cmd, arg); 3299 } 3300 } 3301 #endif 3302 3303 /* 3304 * Obtain the geometry of the device. 3305 * 3306 * You may think that this function is obsolete, but some applications, 3307 * fdisk for example still used CHS values. This function describes the 3308 * device as having 224 heads and 56 sectors per cylinder. These values are 3309 * chosen so that each cylinder is aligned on a 4KB boundary. Since a 3310 * partition is described in terms of a start and end cylinder this means 3311 * that each partition is also 4KB aligned. Non-aligned partitions adversely 3312 * affects performance. 3313 * 3314 * @dev Pointer to the block_device strucutre. 3315 * @geo Pointer to a hd_geometry structure. 3316 * 3317 * return value 3318 * 0 Operation completed successfully. 3319 * -ENOTTY An error occurred while reading the drive capacity. 3320 */ 3321 static int mtip_block_getgeo(struct block_device *dev, 3322 struct hd_geometry *geo) 3323 { 3324 struct driver_data *dd = dev->bd_disk->private_data; 3325 sector_t capacity; 3326 3327 if (!dd) 3328 return -ENOTTY; 3329 3330 if (!(mtip_hw_get_capacity(dd, &capacity))) { 3331 dev_warn(&dd->pdev->dev, 3332 "Could not get drive capacity.\n"); 3333 return -ENOTTY; 3334 } 3335 3336 geo->heads = 224; 3337 geo->sectors = 56; 3338 sector_div(capacity, (geo->heads * geo->sectors)); 3339 geo->cylinders = capacity; 3340 return 0; 3341 } 3342 3343 static int mtip_block_open(struct block_device *dev, fmode_t mode) 3344 { 3345 struct driver_data *dd; 3346 3347 if (dev && dev->bd_disk) { 3348 dd = (struct driver_data *) dev->bd_disk->private_data; 3349 3350 if (dd) { 3351 if (test_bit(MTIP_DDF_REMOVAL_BIT, 3352 &dd->dd_flag)) { 3353 return -ENODEV; 3354 } 3355 return 0; 3356 } 3357 } 3358 return -ENODEV; 3359 } 3360 3361 static void mtip_block_release(struct gendisk *disk, fmode_t mode) 3362 { 3363 } 3364 3365 /* 3366 * Block device operation function. 3367 * 3368 * This structure contains pointers to the functions required by the block 3369 * layer. 3370 */ 3371 static const struct block_device_operations mtip_block_ops = { 3372 .open = mtip_block_open, 3373 .release = mtip_block_release, 3374 .ioctl = mtip_block_ioctl, 3375 #ifdef CONFIG_COMPAT 3376 .compat_ioctl = mtip_block_compat_ioctl, 3377 #endif 3378 .getgeo = mtip_block_getgeo, 3379 .owner = THIS_MODULE 3380 }; 3381 3382 static inline bool is_se_active(struct driver_data *dd) 3383 { 3384 if (unlikely(test_bit(MTIP_PF_SE_ACTIVE_BIT, &dd->port->flags))) { 3385 if (dd->port->ic_pause_timer) { 3386 unsigned long to = dd->port->ic_pause_timer + 3387 msecs_to_jiffies(1000); 3388 if (time_after(jiffies, to)) { 3389 clear_bit(MTIP_PF_SE_ACTIVE_BIT, 3390 &dd->port->flags); 3391 clear_bit(MTIP_DDF_SEC_LOCK_BIT, &dd->dd_flag); 3392 dd->port->ic_pause_timer = 0; 3393 wake_up_interruptible(&dd->port->svc_wait); 3394 return false; 3395 } 3396 } 3397 return true; 3398 } 3399 return false; 3400 } 3401 3402 static inline bool is_stopped(struct driver_data *dd, struct request *rq) 3403 { 3404 if (likely(!(dd->dd_flag & MTIP_DDF_STOP_IO))) 3405 return false; 3406 3407 if (test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag)) 3408 return true; 3409 if (test_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag)) 3410 return true; 3411 if (test_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag) && 3412 rq_data_dir(rq)) 3413 return true; 3414 if (test_bit(MTIP_DDF_SEC_LOCK_BIT, &dd->dd_flag)) 3415 return true; 3416 if (test_bit(MTIP_DDF_REBUILD_FAILED_BIT, &dd->dd_flag)) 3417 return true; 3418 3419 return false; 3420 } 3421 3422 static bool mtip_check_unal_depth(struct blk_mq_hw_ctx *hctx, 3423 struct request *rq) 3424 { 3425 struct driver_data *dd = hctx->queue->queuedata; 3426 struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq); 3427 3428 if (rq_data_dir(rq) == READ || !dd->unal_qdepth) 3429 return false; 3430 3431 /* 3432 * If unaligned depth must be limited on this controller, mark it 3433 * as unaligned if the IO isn't on a 4k boundary (start of length). 3434 */ 3435 if (blk_rq_sectors(rq) <= 64) { 3436 if ((blk_rq_pos(rq) & 7) || (blk_rq_sectors(rq) & 7)) 3437 cmd->unaligned = 1; 3438 } 3439 3440 if (cmd->unaligned && atomic_dec_if_positive(&dd->port->cmd_slot_unal) >= 0) 3441 return true; 3442 3443 return false; 3444 } 3445 3446 static blk_status_t mtip_issue_reserved_cmd(struct blk_mq_hw_ctx *hctx, 3447 struct request *rq) 3448 { 3449 struct driver_data *dd = hctx->queue->queuedata; 3450 struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq); 3451 struct mtip_int_cmd *icmd = cmd->icmd; 3452 struct mtip_cmd_hdr *hdr = 3453 dd->port->command_list + sizeof(struct mtip_cmd_hdr) * rq->tag; 3454 struct mtip_cmd_sg *command_sg; 3455 3456 if (mtip_commands_active(dd->port)) 3457 return BLK_STS_DEV_RESOURCE; 3458 3459 hdr->ctba = cpu_to_le32(cmd->command_dma & 0xFFFFFFFF); 3460 if (test_bit(MTIP_PF_HOST_CAP_64, &dd->port->flags)) 3461 hdr->ctbau = cpu_to_le32((cmd->command_dma >> 16) >> 16); 3462 /* Populate the SG list */ 3463 hdr->opts = cpu_to_le32(icmd->opts | icmd->fis_len); 3464 if (icmd->buf_len) { 3465 command_sg = cmd->command + AHCI_CMD_TBL_HDR_SZ; 3466 3467 command_sg->info = cpu_to_le32((icmd->buf_len-1) & 0x3FFFFF); 3468 command_sg->dba = cpu_to_le32(icmd->buffer & 0xFFFFFFFF); 3469 command_sg->dba_upper = 3470 cpu_to_le32((icmd->buffer >> 16) >> 16); 3471 3472 hdr->opts |= cpu_to_le32((1 << 16)); 3473 } 3474 3475 /* Populate the command header */ 3476 hdr->byte_count = 0; 3477 3478 blk_mq_start_request(rq); 3479 mtip_issue_non_ncq_command(dd->port, rq->tag); 3480 return 0; 3481 } 3482 3483 static blk_status_t mtip_queue_rq(struct blk_mq_hw_ctx *hctx, 3484 const struct blk_mq_queue_data *bd) 3485 { 3486 struct driver_data *dd = hctx->queue->queuedata; 3487 struct request *rq = bd->rq; 3488 struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq); 3489 3490 if (blk_rq_is_passthrough(rq)) 3491 return mtip_issue_reserved_cmd(hctx, rq); 3492 3493 if (unlikely(mtip_check_unal_depth(hctx, rq))) 3494 return BLK_STS_DEV_RESOURCE; 3495 3496 if (is_se_active(dd) || is_stopped(dd, rq)) 3497 return BLK_STS_IOERR; 3498 3499 blk_mq_start_request(rq); 3500 3501 mtip_hw_submit_io(dd, rq, cmd, hctx); 3502 return BLK_STS_OK; 3503 } 3504 3505 static void mtip_free_cmd(struct blk_mq_tag_set *set, struct request *rq, 3506 unsigned int hctx_idx) 3507 { 3508 struct driver_data *dd = set->driver_data; 3509 struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq); 3510 3511 if (!cmd->command) 3512 return; 3513 3514 dma_free_coherent(&dd->pdev->dev, CMD_DMA_ALLOC_SZ, cmd->command, 3515 cmd->command_dma); 3516 } 3517 3518 static int mtip_init_cmd(struct blk_mq_tag_set *set, struct request *rq, 3519 unsigned int hctx_idx, unsigned int numa_node) 3520 { 3521 struct driver_data *dd = set->driver_data; 3522 struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq); 3523 3524 cmd->command = dma_alloc_coherent(&dd->pdev->dev, CMD_DMA_ALLOC_SZ, 3525 &cmd->command_dma, GFP_KERNEL); 3526 if (!cmd->command) 3527 return -ENOMEM; 3528 3529 sg_init_table(cmd->sg, MTIP_MAX_SG); 3530 return 0; 3531 } 3532 3533 static enum blk_eh_timer_return mtip_cmd_timeout(struct request *req, 3534 bool reserved) 3535 { 3536 struct driver_data *dd = req->q->queuedata; 3537 3538 if (reserved) { 3539 struct mtip_cmd *cmd = blk_mq_rq_to_pdu(req); 3540 3541 cmd->status = BLK_STS_TIMEOUT; 3542 blk_mq_complete_request(req); 3543 return BLK_EH_DONE; 3544 } 3545 3546 if (test_bit(req->tag, dd->port->cmds_to_issue)) 3547 goto exit_handler; 3548 3549 if (test_and_set_bit(MTIP_PF_TO_ACTIVE_BIT, &dd->port->flags)) 3550 goto exit_handler; 3551 3552 wake_up_interruptible(&dd->port->svc_wait); 3553 exit_handler: 3554 return BLK_EH_RESET_TIMER; 3555 } 3556 3557 static const struct blk_mq_ops mtip_mq_ops = { 3558 .queue_rq = mtip_queue_rq, 3559 .init_request = mtip_init_cmd, 3560 .exit_request = mtip_free_cmd, 3561 .complete = mtip_softirq_done_fn, 3562 .timeout = mtip_cmd_timeout, 3563 }; 3564 3565 /* 3566 * Block layer initialization function. 3567 * 3568 * This function is called once by the PCI layer for each P320 3569 * device that is connected to the system. 3570 * 3571 * @dd Pointer to the driver data structure. 3572 * 3573 * return value 3574 * 0 on success else an error code. 3575 */ 3576 static int mtip_block_initialize(struct driver_data *dd) 3577 { 3578 int rv = 0, wait_for_rebuild = 0; 3579 sector_t capacity; 3580 unsigned int index = 0; 3581 struct kobject *kobj; 3582 3583 if (dd->disk) 3584 goto skip_create_disk; /* hw init done, before rebuild */ 3585 3586 if (mtip_hw_init(dd)) { 3587 rv = -EINVAL; 3588 goto protocol_init_error; 3589 } 3590 3591 dd->disk = alloc_disk_node(MTIP_MAX_MINORS, dd->numa_node); 3592 if (dd->disk == NULL) { 3593 dev_err(&dd->pdev->dev, 3594 "Unable to allocate gendisk structure\n"); 3595 rv = -EINVAL; 3596 goto alloc_disk_error; 3597 } 3598 3599 rv = ida_alloc(&rssd_index_ida, GFP_KERNEL); 3600 if (rv < 0) 3601 goto ida_get_error; 3602 index = rv; 3603 3604 rv = rssd_disk_name_format("rssd", 3605 index, 3606 dd->disk->disk_name, 3607 DISK_NAME_LEN); 3608 if (rv) 3609 goto disk_index_error; 3610 3611 dd->disk->major = dd->major; 3612 dd->disk->first_minor = index * MTIP_MAX_MINORS; 3613 dd->disk->minors = MTIP_MAX_MINORS; 3614 dd->disk->fops = &mtip_block_ops; 3615 dd->disk->private_data = dd; 3616 dd->index = index; 3617 3618 mtip_hw_debugfs_init(dd); 3619 3620 memset(&dd->tags, 0, sizeof(dd->tags)); 3621 dd->tags.ops = &mtip_mq_ops; 3622 dd->tags.nr_hw_queues = 1; 3623 dd->tags.queue_depth = MTIP_MAX_COMMAND_SLOTS; 3624 dd->tags.reserved_tags = 1; 3625 dd->tags.cmd_size = sizeof(struct mtip_cmd); 3626 dd->tags.numa_node = dd->numa_node; 3627 dd->tags.flags = BLK_MQ_F_SHOULD_MERGE; 3628 dd->tags.driver_data = dd; 3629 dd->tags.timeout = MTIP_NCQ_CMD_TIMEOUT_MS; 3630 3631 rv = blk_mq_alloc_tag_set(&dd->tags); 3632 if (rv) { 3633 dev_err(&dd->pdev->dev, 3634 "Unable to allocate request queue\n"); 3635 goto block_queue_alloc_tag_error; 3636 } 3637 3638 /* Allocate the request queue. */ 3639 dd->queue = blk_mq_init_queue(&dd->tags); 3640 if (IS_ERR(dd->queue)) { 3641 dev_err(&dd->pdev->dev, 3642 "Unable to allocate request queue\n"); 3643 rv = -ENOMEM; 3644 goto block_queue_alloc_init_error; 3645 } 3646 3647 dd->disk->queue = dd->queue; 3648 dd->queue->queuedata = dd; 3649 3650 skip_create_disk: 3651 /* Initialize the protocol layer. */ 3652 wait_for_rebuild = mtip_hw_get_identify(dd); 3653 if (wait_for_rebuild < 0) { 3654 dev_err(&dd->pdev->dev, 3655 "Protocol layer initialization failed\n"); 3656 rv = -EINVAL; 3657 goto init_hw_cmds_error; 3658 } 3659 3660 /* 3661 * if rebuild pending, start the service thread, and delay the block 3662 * queue creation and device_add_disk() 3663 */ 3664 if (wait_for_rebuild == MTIP_FTL_REBUILD_MAGIC) 3665 goto start_service_thread; 3666 3667 /* Set device limits. */ 3668 blk_queue_flag_set(QUEUE_FLAG_NONROT, dd->queue); 3669 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, dd->queue); 3670 blk_queue_max_segments(dd->queue, MTIP_MAX_SG); 3671 blk_queue_physical_block_size(dd->queue, 4096); 3672 blk_queue_max_hw_sectors(dd->queue, 0xffff); 3673 blk_queue_max_segment_size(dd->queue, 0x400000); 3674 dma_set_max_seg_size(&dd->pdev->dev, 0x400000); 3675 blk_queue_io_min(dd->queue, 4096); 3676 3677 /* Set the capacity of the device in 512 byte sectors. */ 3678 if (!(mtip_hw_get_capacity(dd, &capacity))) { 3679 dev_warn(&dd->pdev->dev, 3680 "Could not read drive capacity\n"); 3681 rv = -EIO; 3682 goto read_capacity_error; 3683 } 3684 set_capacity(dd->disk, capacity); 3685 3686 /* Enable the block device and add it to /dev */ 3687 device_add_disk(&dd->pdev->dev, dd->disk, NULL); 3688 3689 dd->bdev = bdget_disk(dd->disk, 0); 3690 /* 3691 * Now that the disk is active, initialize any sysfs attributes 3692 * managed by the protocol layer. 3693 */ 3694 kobj = kobject_get(&disk_to_dev(dd->disk)->kobj); 3695 if (kobj) { 3696 mtip_hw_sysfs_init(dd, kobj); 3697 kobject_put(kobj); 3698 } 3699 3700 if (dd->mtip_svc_handler) { 3701 set_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag); 3702 return rv; /* service thread created for handling rebuild */ 3703 } 3704 3705 start_service_thread: 3706 dd->mtip_svc_handler = kthread_create_on_node(mtip_service_thread, 3707 dd, dd->numa_node, 3708 "mtip_svc_thd_%02d", index); 3709 3710 if (IS_ERR(dd->mtip_svc_handler)) { 3711 dev_err(&dd->pdev->dev, "service thread failed to start\n"); 3712 dd->mtip_svc_handler = NULL; 3713 rv = -EFAULT; 3714 goto kthread_run_error; 3715 } 3716 wake_up_process(dd->mtip_svc_handler); 3717 if (wait_for_rebuild == MTIP_FTL_REBUILD_MAGIC) 3718 rv = wait_for_rebuild; 3719 3720 return rv; 3721 3722 kthread_run_error: 3723 bdput(dd->bdev); 3724 dd->bdev = NULL; 3725 3726 /* Delete our gendisk. This also removes the device from /dev */ 3727 del_gendisk(dd->disk); 3728 3729 read_capacity_error: 3730 init_hw_cmds_error: 3731 blk_cleanup_queue(dd->queue); 3732 block_queue_alloc_init_error: 3733 blk_mq_free_tag_set(&dd->tags); 3734 block_queue_alloc_tag_error: 3735 mtip_hw_debugfs_exit(dd); 3736 disk_index_error: 3737 ida_free(&rssd_index_ida, index); 3738 3739 ida_get_error: 3740 put_disk(dd->disk); 3741 3742 alloc_disk_error: 3743 mtip_hw_exit(dd); /* De-initialize the protocol layer. */ 3744 3745 protocol_init_error: 3746 return rv; 3747 } 3748 3749 static bool mtip_no_dev_cleanup(struct request *rq, void *data, bool reserv) 3750 { 3751 struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq); 3752 3753 cmd->status = BLK_STS_IOERR; 3754 blk_mq_complete_request(rq); 3755 return true; 3756 } 3757 3758 /* 3759 * Block layer deinitialization function. 3760 * 3761 * Called by the PCI layer as each P320 device is removed. 3762 * 3763 * @dd Pointer to the driver data structure. 3764 * 3765 * return value 3766 * 0 3767 */ 3768 static int mtip_block_remove(struct driver_data *dd) 3769 { 3770 struct kobject *kobj; 3771 3772 mtip_hw_debugfs_exit(dd); 3773 3774 if (dd->mtip_svc_handler) { 3775 set_bit(MTIP_PF_SVC_THD_STOP_BIT, &dd->port->flags); 3776 wake_up_interruptible(&dd->port->svc_wait); 3777 kthread_stop(dd->mtip_svc_handler); 3778 } 3779 3780 /* Clean up the sysfs attributes, if created */ 3781 if (test_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag)) { 3782 kobj = kobject_get(&disk_to_dev(dd->disk)->kobj); 3783 if (kobj) { 3784 mtip_hw_sysfs_exit(dd, kobj); 3785 kobject_put(kobj); 3786 } 3787 } 3788 3789 if (!dd->sr) { 3790 /* 3791 * Explicitly wait here for IOs to quiesce, 3792 * as mtip_standby_drive usually won't wait for IOs. 3793 */ 3794 if (!mtip_quiesce_io(dd->port, MTIP_QUIESCE_IO_TIMEOUT_MS)) 3795 mtip_standby_drive(dd); 3796 } 3797 else 3798 dev_info(&dd->pdev->dev, "device %s surprise removal\n", 3799 dd->disk->disk_name); 3800 3801 blk_freeze_queue_start(dd->queue); 3802 blk_mq_quiesce_queue(dd->queue); 3803 blk_mq_tagset_busy_iter(&dd->tags, mtip_no_dev_cleanup, dd); 3804 blk_mq_unquiesce_queue(dd->queue); 3805 3806 /* 3807 * Delete our gendisk structure. This also removes the device 3808 * from /dev 3809 */ 3810 if (dd->bdev) { 3811 bdput(dd->bdev); 3812 dd->bdev = NULL; 3813 } 3814 if (dd->disk) { 3815 if (test_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag)) 3816 del_gendisk(dd->disk); 3817 if (dd->disk->queue) { 3818 blk_cleanup_queue(dd->queue); 3819 blk_mq_free_tag_set(&dd->tags); 3820 dd->queue = NULL; 3821 } 3822 put_disk(dd->disk); 3823 } 3824 dd->disk = NULL; 3825 3826 ida_free(&rssd_index_ida, dd->index); 3827 3828 /* De-initialize the protocol layer. */ 3829 mtip_hw_exit(dd); 3830 3831 return 0; 3832 } 3833 3834 /* 3835 * Function called by the PCI layer when just before the 3836 * machine shuts down. 3837 * 3838 * If a protocol layer shutdown function is present it will be called 3839 * by this function. 3840 * 3841 * @dd Pointer to the driver data structure. 3842 * 3843 * return value 3844 * 0 3845 */ 3846 static int mtip_block_shutdown(struct driver_data *dd) 3847 { 3848 mtip_hw_shutdown(dd); 3849 3850 /* Delete our gendisk structure, and cleanup the blk queue. */ 3851 if (dd->disk) { 3852 dev_info(&dd->pdev->dev, 3853 "Shutting down %s ...\n", dd->disk->disk_name); 3854 3855 if (test_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag)) 3856 del_gendisk(dd->disk); 3857 if (dd->disk->queue) { 3858 blk_cleanup_queue(dd->queue); 3859 blk_mq_free_tag_set(&dd->tags); 3860 } 3861 put_disk(dd->disk); 3862 dd->disk = NULL; 3863 dd->queue = NULL; 3864 } 3865 3866 ida_free(&rssd_index_ida, dd->index); 3867 return 0; 3868 } 3869 3870 static int mtip_block_suspend(struct driver_data *dd) 3871 { 3872 dev_info(&dd->pdev->dev, 3873 "Suspending %s ...\n", dd->disk->disk_name); 3874 mtip_hw_suspend(dd); 3875 return 0; 3876 } 3877 3878 static int mtip_block_resume(struct driver_data *dd) 3879 { 3880 dev_info(&dd->pdev->dev, "Resuming %s ...\n", 3881 dd->disk->disk_name); 3882 mtip_hw_resume(dd); 3883 return 0; 3884 } 3885 3886 static void drop_cpu(int cpu) 3887 { 3888 cpu_use[cpu]--; 3889 } 3890 3891 static int get_least_used_cpu_on_node(int node) 3892 { 3893 int cpu, least_used_cpu, least_cnt; 3894 const struct cpumask *node_mask; 3895 3896 node_mask = cpumask_of_node(node); 3897 least_used_cpu = cpumask_first(node_mask); 3898 least_cnt = cpu_use[least_used_cpu]; 3899 cpu = least_used_cpu; 3900 3901 for_each_cpu(cpu, node_mask) { 3902 if (cpu_use[cpu] < least_cnt) { 3903 least_used_cpu = cpu; 3904 least_cnt = cpu_use[cpu]; 3905 } 3906 } 3907 cpu_use[least_used_cpu]++; 3908 return least_used_cpu; 3909 } 3910 3911 /* Helper for selecting a node in round robin mode */ 3912 static inline int mtip_get_next_rr_node(void) 3913 { 3914 static int next_node = NUMA_NO_NODE; 3915 3916 if (next_node == NUMA_NO_NODE) { 3917 next_node = first_online_node; 3918 return next_node; 3919 } 3920 3921 next_node = next_online_node(next_node); 3922 if (next_node == MAX_NUMNODES) 3923 next_node = first_online_node; 3924 return next_node; 3925 } 3926 3927 static DEFINE_HANDLER(0); 3928 static DEFINE_HANDLER(1); 3929 static DEFINE_HANDLER(2); 3930 static DEFINE_HANDLER(3); 3931 static DEFINE_HANDLER(4); 3932 static DEFINE_HANDLER(5); 3933 static DEFINE_HANDLER(6); 3934 static DEFINE_HANDLER(7); 3935 3936 static void mtip_disable_link_opts(struct driver_data *dd, struct pci_dev *pdev) 3937 { 3938 int pos; 3939 unsigned short pcie_dev_ctrl; 3940 3941 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP); 3942 if (pos) { 3943 pci_read_config_word(pdev, 3944 pos + PCI_EXP_DEVCTL, 3945 &pcie_dev_ctrl); 3946 if (pcie_dev_ctrl & (1 << 11) || 3947 pcie_dev_ctrl & (1 << 4)) { 3948 dev_info(&dd->pdev->dev, 3949 "Disabling ERO/No-Snoop on bridge device %04x:%04x\n", 3950 pdev->vendor, pdev->device); 3951 pcie_dev_ctrl &= ~(PCI_EXP_DEVCTL_NOSNOOP_EN | 3952 PCI_EXP_DEVCTL_RELAX_EN); 3953 pci_write_config_word(pdev, 3954 pos + PCI_EXP_DEVCTL, 3955 pcie_dev_ctrl); 3956 } 3957 } 3958 } 3959 3960 static void mtip_fix_ero_nosnoop(struct driver_data *dd, struct pci_dev *pdev) 3961 { 3962 /* 3963 * This workaround is specific to AMD/ATI chipset with a PCI upstream 3964 * device with device id 0x5aXX 3965 */ 3966 if (pdev->bus && pdev->bus->self) { 3967 if (pdev->bus->self->vendor == PCI_VENDOR_ID_ATI && 3968 ((pdev->bus->self->device & 0xff00) == 0x5a00)) { 3969 mtip_disable_link_opts(dd, pdev->bus->self); 3970 } else { 3971 /* Check further up the topology */ 3972 struct pci_dev *parent_dev = pdev->bus->self; 3973 if (parent_dev->bus && 3974 parent_dev->bus->parent && 3975 parent_dev->bus->parent->self && 3976 parent_dev->bus->parent->self->vendor == 3977 PCI_VENDOR_ID_ATI && 3978 (parent_dev->bus->parent->self->device & 3979 0xff00) == 0x5a00) { 3980 mtip_disable_link_opts(dd, 3981 parent_dev->bus->parent->self); 3982 } 3983 } 3984 } 3985 } 3986 3987 /* 3988 * Called for each supported PCI device detected. 3989 * 3990 * This function allocates the private data structure, enables the 3991 * PCI device and then calls the block layer initialization function. 3992 * 3993 * return value 3994 * 0 on success else an error code. 3995 */ 3996 static int mtip_pci_probe(struct pci_dev *pdev, 3997 const struct pci_device_id *ent) 3998 { 3999 int rv = 0; 4000 struct driver_data *dd = NULL; 4001 char cpu_list[256]; 4002 const struct cpumask *node_mask; 4003 int cpu, i = 0, j = 0; 4004 int my_node = NUMA_NO_NODE; 4005 unsigned long flags; 4006 4007 /* Allocate memory for this devices private data. */ 4008 my_node = pcibus_to_node(pdev->bus); 4009 if (my_node != NUMA_NO_NODE) { 4010 if (!node_online(my_node)) 4011 my_node = mtip_get_next_rr_node(); 4012 } else { 4013 dev_info(&pdev->dev, "Kernel not reporting proximity, choosing a node\n"); 4014 my_node = mtip_get_next_rr_node(); 4015 } 4016 dev_info(&pdev->dev, "NUMA node %d (closest: %d,%d, probe on %d:%d)\n", 4017 my_node, pcibus_to_node(pdev->bus), dev_to_node(&pdev->dev), 4018 cpu_to_node(raw_smp_processor_id()), raw_smp_processor_id()); 4019 4020 dd = kzalloc_node(sizeof(struct driver_data), GFP_KERNEL, my_node); 4021 if (dd == NULL) { 4022 dev_err(&pdev->dev, 4023 "Unable to allocate memory for driver data\n"); 4024 return -ENOMEM; 4025 } 4026 4027 /* Attach the private data to this PCI device. */ 4028 pci_set_drvdata(pdev, dd); 4029 4030 rv = pcim_enable_device(pdev); 4031 if (rv < 0) { 4032 dev_err(&pdev->dev, "Unable to enable device\n"); 4033 goto iomap_err; 4034 } 4035 4036 /* Map BAR5 to memory. */ 4037 rv = pcim_iomap_regions(pdev, 1 << MTIP_ABAR, MTIP_DRV_NAME); 4038 if (rv < 0) { 4039 dev_err(&pdev->dev, "Unable to map regions\n"); 4040 goto iomap_err; 4041 } 4042 4043 rv = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 4044 if (rv) { 4045 dev_warn(&pdev->dev, "64-bit DMA enable failed\n"); 4046 goto setmask_err; 4047 } 4048 4049 /* Copy the info we may need later into the private data structure. */ 4050 dd->major = mtip_major; 4051 dd->instance = instance; 4052 dd->pdev = pdev; 4053 dd->numa_node = my_node; 4054 4055 INIT_LIST_HEAD(&dd->online_list); 4056 INIT_LIST_HEAD(&dd->remove_list); 4057 4058 memset(dd->workq_name, 0, 32); 4059 snprintf(dd->workq_name, 31, "mtipq%d", dd->instance); 4060 4061 dd->isr_workq = create_workqueue(dd->workq_name); 4062 if (!dd->isr_workq) { 4063 dev_warn(&pdev->dev, "Can't create wq %d\n", dd->instance); 4064 rv = -ENOMEM; 4065 goto setmask_err; 4066 } 4067 4068 memset(cpu_list, 0, sizeof(cpu_list)); 4069 4070 node_mask = cpumask_of_node(dd->numa_node); 4071 if (!cpumask_empty(node_mask)) { 4072 for_each_cpu(cpu, node_mask) 4073 { 4074 snprintf(&cpu_list[j], 256 - j, "%d ", cpu); 4075 j = strlen(cpu_list); 4076 } 4077 4078 dev_info(&pdev->dev, "Node %d on package %d has %d cpu(s): %s\n", 4079 dd->numa_node, 4080 topology_physical_package_id(cpumask_first(node_mask)), 4081 nr_cpus_node(dd->numa_node), 4082 cpu_list); 4083 } else 4084 dev_dbg(&pdev->dev, "mtip32xx: node_mask empty\n"); 4085 4086 dd->isr_binding = get_least_used_cpu_on_node(dd->numa_node); 4087 dev_info(&pdev->dev, "Initial IRQ binding node:cpu %d:%d\n", 4088 cpu_to_node(dd->isr_binding), dd->isr_binding); 4089 4090 /* first worker context always runs in ISR */ 4091 dd->work[0].cpu_binding = dd->isr_binding; 4092 dd->work[1].cpu_binding = get_least_used_cpu_on_node(dd->numa_node); 4093 dd->work[2].cpu_binding = get_least_used_cpu_on_node(dd->numa_node); 4094 dd->work[3].cpu_binding = dd->work[0].cpu_binding; 4095 dd->work[4].cpu_binding = dd->work[1].cpu_binding; 4096 dd->work[5].cpu_binding = dd->work[2].cpu_binding; 4097 dd->work[6].cpu_binding = dd->work[2].cpu_binding; 4098 dd->work[7].cpu_binding = dd->work[1].cpu_binding; 4099 4100 /* Log the bindings */ 4101 for_each_present_cpu(cpu) { 4102 memset(cpu_list, 0, sizeof(cpu_list)); 4103 for (i = 0, j = 0; i < MTIP_MAX_SLOT_GROUPS; i++) { 4104 if (dd->work[i].cpu_binding == cpu) { 4105 snprintf(&cpu_list[j], 256 - j, "%d ", i); 4106 j = strlen(cpu_list); 4107 } 4108 } 4109 if (j) 4110 dev_info(&pdev->dev, "CPU %d: WQs %s\n", cpu, cpu_list); 4111 } 4112 4113 INIT_WORK(&dd->work[0].work, mtip_workq_sdbf0); 4114 INIT_WORK(&dd->work[1].work, mtip_workq_sdbf1); 4115 INIT_WORK(&dd->work[2].work, mtip_workq_sdbf2); 4116 INIT_WORK(&dd->work[3].work, mtip_workq_sdbf3); 4117 INIT_WORK(&dd->work[4].work, mtip_workq_sdbf4); 4118 INIT_WORK(&dd->work[5].work, mtip_workq_sdbf5); 4119 INIT_WORK(&dd->work[6].work, mtip_workq_sdbf6); 4120 INIT_WORK(&dd->work[7].work, mtip_workq_sdbf7); 4121 4122 pci_set_master(pdev); 4123 rv = pci_enable_msi(pdev); 4124 if (rv) { 4125 dev_warn(&pdev->dev, 4126 "Unable to enable MSI interrupt.\n"); 4127 goto msi_initialize_err; 4128 } 4129 4130 mtip_fix_ero_nosnoop(dd, pdev); 4131 4132 /* Initialize the block layer. */ 4133 rv = mtip_block_initialize(dd); 4134 if (rv < 0) { 4135 dev_err(&pdev->dev, 4136 "Unable to initialize block layer\n"); 4137 goto block_initialize_err; 4138 } 4139 4140 /* 4141 * Increment the instance count so that each device has a unique 4142 * instance number. 4143 */ 4144 instance++; 4145 if (rv != MTIP_FTL_REBUILD_MAGIC) 4146 set_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag); 4147 else 4148 rv = 0; /* device in rebuild state, return 0 from probe */ 4149 4150 /* Add to online list even if in ftl rebuild */ 4151 spin_lock_irqsave(&dev_lock, flags); 4152 list_add(&dd->online_list, &online_list); 4153 spin_unlock_irqrestore(&dev_lock, flags); 4154 4155 goto done; 4156 4157 block_initialize_err: 4158 pci_disable_msi(pdev); 4159 4160 msi_initialize_err: 4161 if (dd->isr_workq) { 4162 flush_workqueue(dd->isr_workq); 4163 destroy_workqueue(dd->isr_workq); 4164 drop_cpu(dd->work[0].cpu_binding); 4165 drop_cpu(dd->work[1].cpu_binding); 4166 drop_cpu(dd->work[2].cpu_binding); 4167 } 4168 setmask_err: 4169 pcim_iounmap_regions(pdev, 1 << MTIP_ABAR); 4170 4171 iomap_err: 4172 kfree(dd); 4173 pci_set_drvdata(pdev, NULL); 4174 return rv; 4175 done: 4176 return rv; 4177 } 4178 4179 /* 4180 * Called for each probed device when the device is removed or the 4181 * driver is unloaded. 4182 * 4183 * return value 4184 * None 4185 */ 4186 static void mtip_pci_remove(struct pci_dev *pdev) 4187 { 4188 struct driver_data *dd = pci_get_drvdata(pdev); 4189 unsigned long flags, to; 4190 4191 set_bit(MTIP_DDF_REMOVAL_BIT, &dd->dd_flag); 4192 4193 spin_lock_irqsave(&dev_lock, flags); 4194 list_del_init(&dd->online_list); 4195 list_add(&dd->remove_list, &removing_list); 4196 spin_unlock_irqrestore(&dev_lock, flags); 4197 4198 mtip_check_surprise_removal(pdev); 4199 synchronize_irq(dd->pdev->irq); 4200 4201 /* Spin until workers are done */ 4202 to = jiffies + msecs_to_jiffies(4000); 4203 do { 4204 msleep(20); 4205 } while (atomic_read(&dd->irq_workers_active) != 0 && 4206 time_before(jiffies, to)); 4207 4208 if (!dd->sr) 4209 fsync_bdev(dd->bdev); 4210 4211 if (atomic_read(&dd->irq_workers_active) != 0) { 4212 dev_warn(&dd->pdev->dev, 4213 "Completion workers still active!\n"); 4214 } 4215 4216 blk_set_queue_dying(dd->queue); 4217 set_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag); 4218 4219 /* Clean up the block layer. */ 4220 mtip_block_remove(dd); 4221 4222 if (dd->isr_workq) { 4223 flush_workqueue(dd->isr_workq); 4224 destroy_workqueue(dd->isr_workq); 4225 drop_cpu(dd->work[0].cpu_binding); 4226 drop_cpu(dd->work[1].cpu_binding); 4227 drop_cpu(dd->work[2].cpu_binding); 4228 } 4229 4230 pci_disable_msi(pdev); 4231 4232 spin_lock_irqsave(&dev_lock, flags); 4233 list_del_init(&dd->remove_list); 4234 spin_unlock_irqrestore(&dev_lock, flags); 4235 4236 kfree(dd); 4237 4238 pcim_iounmap_regions(pdev, 1 << MTIP_ABAR); 4239 pci_set_drvdata(pdev, NULL); 4240 } 4241 4242 /* 4243 * Called for each probed device when the device is suspended. 4244 * 4245 * return value 4246 * 0 Success 4247 * <0 Error 4248 */ 4249 static int mtip_pci_suspend(struct pci_dev *pdev, pm_message_t mesg) 4250 { 4251 int rv = 0; 4252 struct driver_data *dd = pci_get_drvdata(pdev); 4253 4254 if (!dd) { 4255 dev_err(&pdev->dev, 4256 "Driver private datastructure is NULL\n"); 4257 return -EFAULT; 4258 } 4259 4260 set_bit(MTIP_DDF_RESUME_BIT, &dd->dd_flag); 4261 4262 /* Disable ports & interrupts then send standby immediate */ 4263 rv = mtip_block_suspend(dd); 4264 if (rv < 0) { 4265 dev_err(&pdev->dev, 4266 "Failed to suspend controller\n"); 4267 return rv; 4268 } 4269 4270 /* 4271 * Save the pci config space to pdev structure & 4272 * disable the device 4273 */ 4274 pci_save_state(pdev); 4275 pci_disable_device(pdev); 4276 4277 /* Move to Low power state*/ 4278 pci_set_power_state(pdev, PCI_D3hot); 4279 4280 return rv; 4281 } 4282 4283 /* 4284 * Called for each probed device when the device is resumed. 4285 * 4286 * return value 4287 * 0 Success 4288 * <0 Error 4289 */ 4290 static int mtip_pci_resume(struct pci_dev *pdev) 4291 { 4292 int rv = 0; 4293 struct driver_data *dd; 4294 4295 dd = pci_get_drvdata(pdev); 4296 if (!dd) { 4297 dev_err(&pdev->dev, 4298 "Driver private datastructure is NULL\n"); 4299 return -EFAULT; 4300 } 4301 4302 /* Move the device to active State */ 4303 pci_set_power_state(pdev, PCI_D0); 4304 4305 /* Restore PCI configuration space */ 4306 pci_restore_state(pdev); 4307 4308 /* Enable the PCI device*/ 4309 rv = pcim_enable_device(pdev); 4310 if (rv < 0) { 4311 dev_err(&pdev->dev, 4312 "Failed to enable card during resume\n"); 4313 goto err; 4314 } 4315 pci_set_master(pdev); 4316 4317 /* 4318 * Calls hbaReset, initPort, & startPort function 4319 * then enables interrupts 4320 */ 4321 rv = mtip_block_resume(dd); 4322 if (rv < 0) 4323 dev_err(&pdev->dev, "Unable to resume\n"); 4324 4325 err: 4326 clear_bit(MTIP_DDF_RESUME_BIT, &dd->dd_flag); 4327 4328 return rv; 4329 } 4330 4331 /* 4332 * Shutdown routine 4333 * 4334 * return value 4335 * None 4336 */ 4337 static void mtip_pci_shutdown(struct pci_dev *pdev) 4338 { 4339 struct driver_data *dd = pci_get_drvdata(pdev); 4340 if (dd) 4341 mtip_block_shutdown(dd); 4342 } 4343 4344 /* Table of device ids supported by this driver. */ 4345 static const struct pci_device_id mtip_pci_tbl[] = { 4346 { PCI_DEVICE(PCI_VENDOR_ID_MICRON, P320H_DEVICE_ID) }, 4347 { PCI_DEVICE(PCI_VENDOR_ID_MICRON, P320M_DEVICE_ID) }, 4348 { PCI_DEVICE(PCI_VENDOR_ID_MICRON, P320S_DEVICE_ID) }, 4349 { PCI_DEVICE(PCI_VENDOR_ID_MICRON, P325M_DEVICE_ID) }, 4350 { PCI_DEVICE(PCI_VENDOR_ID_MICRON, P420H_DEVICE_ID) }, 4351 { PCI_DEVICE(PCI_VENDOR_ID_MICRON, P420M_DEVICE_ID) }, 4352 { PCI_DEVICE(PCI_VENDOR_ID_MICRON, P425M_DEVICE_ID) }, 4353 { 0 } 4354 }; 4355 4356 /* Structure that describes the PCI driver functions. */ 4357 static struct pci_driver mtip_pci_driver = { 4358 .name = MTIP_DRV_NAME, 4359 .id_table = mtip_pci_tbl, 4360 .probe = mtip_pci_probe, 4361 .remove = mtip_pci_remove, 4362 .suspend = mtip_pci_suspend, 4363 .resume = mtip_pci_resume, 4364 .shutdown = mtip_pci_shutdown, 4365 }; 4366 4367 MODULE_DEVICE_TABLE(pci, mtip_pci_tbl); 4368 4369 /* 4370 * Module initialization function. 4371 * 4372 * Called once when the module is loaded. This function allocates a major 4373 * block device number to the Cyclone devices and registers the PCI layer 4374 * of the driver. 4375 * 4376 * Return value 4377 * 0 on success else error code. 4378 */ 4379 static int __init mtip_init(void) 4380 { 4381 int error; 4382 4383 pr_info(MTIP_DRV_NAME " Version " MTIP_DRV_VERSION "\n"); 4384 4385 spin_lock_init(&dev_lock); 4386 4387 INIT_LIST_HEAD(&online_list); 4388 INIT_LIST_HEAD(&removing_list); 4389 4390 /* Allocate a major block device number to use with this driver. */ 4391 error = register_blkdev(0, MTIP_DRV_NAME); 4392 if (error <= 0) { 4393 pr_err("Unable to register block device (%d)\n", 4394 error); 4395 return -EBUSY; 4396 } 4397 mtip_major = error; 4398 4399 dfs_parent = debugfs_create_dir("rssd", NULL); 4400 if (IS_ERR_OR_NULL(dfs_parent)) { 4401 pr_warn("Error creating debugfs parent\n"); 4402 dfs_parent = NULL; 4403 } 4404 if (dfs_parent) { 4405 dfs_device_status = debugfs_create_file("device_status", 4406 0444, dfs_parent, NULL, 4407 &mtip_device_status_fops); 4408 if (IS_ERR_OR_NULL(dfs_device_status)) { 4409 pr_err("Error creating device_status node\n"); 4410 dfs_device_status = NULL; 4411 } 4412 } 4413 4414 /* Register our PCI operations. */ 4415 error = pci_register_driver(&mtip_pci_driver); 4416 if (error) { 4417 debugfs_remove(dfs_parent); 4418 unregister_blkdev(mtip_major, MTIP_DRV_NAME); 4419 } 4420 4421 return error; 4422 } 4423 4424 /* 4425 * Module de-initialization function. 4426 * 4427 * Called once when the module is unloaded. This function deallocates 4428 * the major block device number allocated by mtip_init() and 4429 * unregisters the PCI layer of the driver. 4430 * 4431 * Return value 4432 * none 4433 */ 4434 static void __exit mtip_exit(void) 4435 { 4436 /* Release the allocated major block device number. */ 4437 unregister_blkdev(mtip_major, MTIP_DRV_NAME); 4438 4439 /* Unregister the PCI driver. */ 4440 pci_unregister_driver(&mtip_pci_driver); 4441 4442 debugfs_remove_recursive(dfs_parent); 4443 } 4444 4445 MODULE_AUTHOR("Micron Technology, Inc"); 4446 MODULE_DESCRIPTION("Micron RealSSD PCIe Block Driver"); 4447 MODULE_LICENSE("GPL"); 4448 MODULE_VERSION(MTIP_DRV_VERSION); 4449 4450 module_init(mtip_init); 4451 module_exit(mtip_exit); 4452