drivers/firmware/qemu_fw_cfg.c

   1 /*
   2  * drivers/firmware/qemu_fw_cfg.c
   3  *
   4  * Copyright 2015 Carnegie Mellon University
   5  *
   6  * Expose entries from QEMU's firmware configuration (fw_cfg) device in
   7  * sysfs (read-only, under "/sys/firmware/qemu_fw_cfg/...").
   8  *
   9  * The fw_cfg device may be instantiated via either an ACPI node (on x86
  10  * and select subsets of aarch64), a Device Tree node (on arm), or using
  11  * a kernel module (or command line) parameter with the following syntax:
  12  *
  13  *      [qemu_fw_cfg.]ioport=<size>@<base>[:<ctrl_off>:<data_off>[:<dma_off>]]
  14  * or
  15  *      [qemu_fw_cfg.]mmio=<size>@<base>[:<ctrl_off>:<data_off>[:<dma_off>]]
  16  *
  17  * where:
  18  *      <size>     := size of ioport or mmio range
  19  *      <base>     := physical base address of ioport or mmio range
  20  *      <ctrl_off> := (optional) offset of control register
  21  *      <data_off> := (optional) offset of data register
  22  *      <dma_off> := (optional) offset of dma register
  23  *
  24  * e.g.:
  25  *      qemu_fw_cfg.ioport=12@0x510:0:1:4       (the default on x86)
  26  * or
  27  *      qemu_fw_cfg.mmio=16@0x9020000:8:0:16    (the default on arm)
  28  */
  29
  30 #include <linux/module.h>
  31 #include <linux/mod_devicetable.h>
  32 #include <linux/platform_device.h>
  33 #include <linux/acpi.h>
  34 #include <linux/slab.h>
  35 #include <linux/io.h>
  36 #include <linux/ioport.h>
  37 #include <uapi/linux/qemu_fw_cfg.h>
  38 #include <linux/delay.h>
  39 #include <linux/crash_dump.h>
  40 #include <linux/crash_core.h>
  41
  42 MODULE_AUTHOR("Gabriel L. Somlo <somlo@cmu.edu>");
  43 MODULE_DESCRIPTION("QEMU fw_cfg sysfs support");
  44 MODULE_LICENSE("GPL");
  45
  46 /* fw_cfg revision attribute, in /sys/firmware/qemu_fw_cfg top-level dir. */
  47 static u32 fw_cfg_rev;
  48
  49 /* fw_cfg device i/o register addresses */
  50 static bool fw_cfg_is_mmio;
  51 static phys_addr_t fw_cfg_p_base;
  52 static resource_size_t fw_cfg_p_size;
  53 static void __iomem *fw_cfg_dev_base;
  54 static void __iomem *fw_cfg_reg_ctrl;
  55 static void __iomem *fw_cfg_reg_data;
  56 static void __iomem *fw_cfg_reg_dma;
  57
  58 /* atomic access to fw_cfg device (potentially slow i/o, so using mutex) */
  59 static DEFINE_MUTEX(fw_cfg_dev_lock);
  60
  61 /* pick appropriate endianness for selector key */
  62 static void fw_cfg_sel_endianness(u16 key)
  63 {
  64         if (fw_cfg_is_mmio)
  65                 iowrite16be(key, fw_cfg_reg_ctrl);
  66         else
  67                 iowrite16(key, fw_cfg_reg_ctrl);
  68 }
  69
  70 #ifdef CONFIG_CRASH_CORE
  71 static inline bool fw_cfg_dma_enabled(void)
  72 {
  73         return (fw_cfg_rev & FW_CFG_VERSION_DMA) && fw_cfg_reg_dma;
  74 }
  75
  76 /* qemu fw_cfg device is sync today, but spec says it may become async */
  77 static void fw_cfg_wait_for_control(struct fw_cfg_dma_access *d)
  78 {
  79         for (;;) {
  80                 u32 ctrl = be32_to_cpu(READ_ONCE(d->control));
  81
  82                 /* do not reorder the read to d->control */
  83                 rmb();
  84                 if ((ctrl & ~FW_CFG_DMA_CTL_ERROR) == 0)
  85                         return;
  86
  87                 cpu_relax();
  88         }
  89 }
  90
  91 static ssize_t fw_cfg_dma_transfer(void *address, u32 length, u32 control)
  92 {
  93         phys_addr_t dma;
  94         struct fw_cfg_dma_access *d = NULL;
  95         ssize_t ret = length;
  96
  97         d = kmalloc(sizeof(*d), GFP_KERNEL);
  98         if (!d) {
  99                 ret = -ENOMEM;
 100                 goto end;
 101         }
 102
 103         /* fw_cfg device does not need IOMMU protection, so use physical addresses */
 104         *d = (struct fw_cfg_dma_access) {
 105                 .address = cpu_to_be64(address ? virt_to_phys(address) : 0),
 106                 .length = cpu_to_be32(length),
 107                 .control = cpu_to_be32(control)
 108         };
 109
 110         dma = virt_to_phys(d);
 111
 112         iowrite32be((u64)dma >> 32, fw_cfg_reg_dma);
 113         /* force memory to sync before notifying device via MMIO */
 114         wmb();
 115         iowrite32be(dma, fw_cfg_reg_dma + 4);
 116
 117         fw_cfg_wait_for_control(d);
 118
 119         if (be32_to_cpu(READ_ONCE(d->control)) & FW_CFG_DMA_CTL_ERROR) {
 120                 ret = -EIO;
 121         }
 122
 123 end:
 124         kfree(d);
 125
 126         return ret;
 127 }
 128 #endif
 129
 130 /* read chunk of given fw_cfg blob (caller responsible for sanity-check) */
 131 static ssize_t fw_cfg_read_blob(u16 key,
 132                                 void *buf, loff_t pos, size_t count)
 133 {
 134         u32 glk = -1U;
 135         acpi_status status;
 136
 137         /* If we have ACPI, ensure mutual exclusion against any potential
 138          * device access by the firmware, e.g. via AML methods:
 139          */
 140         status = acpi_acquire_global_lock(ACPI_WAIT_FOREVER, &glk);
 141         if (ACPI_FAILURE(status) && status != AE_NOT_CONFIGURED) {
 142                 /* Should never get here */
 143                 WARN(1, "fw_cfg_read_blob: Failed to lock ACPI!\n");
 144                 memset(buf, 0, count);
 145                 return -EINVAL;
 146         }
 147
 148         mutex_lock(&fw_cfg_dev_lock);
 149         fw_cfg_sel_endianness(key);
 150         while (pos-- > 0)
 151                 ioread8(fw_cfg_reg_data);
 152         ioread8_rep(fw_cfg_reg_data, buf, count);
 153         mutex_unlock(&fw_cfg_dev_lock);
 154
 155         acpi_release_global_lock(glk);
 156         return count;
 157 }
 158
 159 #ifdef CONFIG_CRASH_CORE
 160 /* write chunk of given fw_cfg blob (caller responsible for sanity-check) */
 161 static ssize_t fw_cfg_write_blob(u16 key,
 162                                  void *buf, loff_t pos, size_t count)
 163 {
 164         u32 glk = -1U;
 165         acpi_status status;
 166         ssize_t ret = count;
 167
 168         /* If we have ACPI, ensure mutual exclusion against any potential
 169          * device access by the firmware, e.g. via AML methods:
 170          */
 171         status = acpi_acquire_global_lock(ACPI_WAIT_FOREVER, &glk);
 172         if (ACPI_FAILURE(status) && status != AE_NOT_CONFIGURED) {
 173                 /* Should never get here */
 174                 WARN(1, "%s: Failed to lock ACPI!\n", __func__);
 175                 return -EINVAL;
 176         }
 177
 178         mutex_lock(&fw_cfg_dev_lock);
 179         if (pos == 0) {
 180                 ret = fw_cfg_dma_transfer(buf, count, key << 16
 181                                           | FW_CFG_DMA_CTL_SELECT
 182                                           | FW_CFG_DMA_CTL_WRITE);
 183         } else {
 184                 fw_cfg_sel_endianness(key);
 185                 ret = fw_cfg_dma_transfer(NULL, pos, FW_CFG_DMA_CTL_SKIP);
 186                 if (ret < 0)
 187                         goto end;
 188                 ret = fw_cfg_dma_transfer(buf, count, FW_CFG_DMA_CTL_WRITE);
 189         }
 190
 191 end:
 192         mutex_unlock(&fw_cfg_dev_lock);
 193
 194         acpi_release_global_lock(glk);
 195
 196         return ret;
 197 }
 198 #endif /* CONFIG_CRASH_CORE */
 199
 200 /* clean up fw_cfg device i/o */
 201 static void fw_cfg_io_cleanup(void)
 202 {
 203         if (fw_cfg_is_mmio) {
 204                 iounmap(fw_cfg_dev_base);
 205                 release_mem_region(fw_cfg_p_base, fw_cfg_p_size);
 206         } else {
 207                 ioport_unmap(fw_cfg_dev_base);
 208                 release_region(fw_cfg_p_base, fw_cfg_p_size);
 209         }
 210 }
 211
 212 /* arch-specific ctrl & data register offsets are not available in ACPI, DT */
 213 #if !(defined(FW_CFG_CTRL_OFF) && defined(FW_CFG_DATA_OFF))
 214 # if (defined(CONFIG_ARM) || defined(CONFIG_ARM64))
 215 #  define FW_CFG_CTRL_OFF 0x08
 216 #  define FW_CFG_DATA_OFF 0x00
 217 #  define FW_CFG_DMA_OFF 0x10
 218 # elif defined(CONFIG_PARISC)   /* parisc */
 219 #  define FW_CFG_CTRL_OFF 0x00
 220 #  define FW_CFG_DATA_OFF 0x04
 221 # elif (defined(CONFIG_PPC_PMAC) || defined(CONFIG_SPARC32)) /* ppc/mac,sun4m */
 222 #  define FW_CFG_CTRL_OFF 0x00
 223 #  define FW_CFG_DATA_OFF 0x02
 224 # elif (defined(CONFIG_X86) || defined(CONFIG_SPARC64)) /* x86, sun4u */
 225 #  define FW_CFG_CTRL_OFF 0x00
 226 #  define FW_CFG_DATA_OFF 0x01
 227 #  define FW_CFG_DMA_OFF 0x04
 228 # else
 229 #  error "QEMU FW_CFG not available on this architecture!"
 230 # endif
 231 #endif
 232
 233 /* initialize fw_cfg device i/o from platform data */
 234 static int fw_cfg_do_platform_probe(struct platform_device *pdev)
 235 {
 236         char sig[FW_CFG_SIG_SIZE];
 237         struct resource *range, *ctrl, *data, *dma;
 238
 239         /* acquire i/o range details */
 240         fw_cfg_is_mmio = false;
 241         range = platform_get_resource(pdev, IORESOURCE_IO, 0);
 242         if (!range) {
 243                 fw_cfg_is_mmio = true;
 244                 range = platform_get_resource(pdev, IORESOURCE_MEM, 0);
 245                 if (!range)
 246                         return -EINVAL;
 247         }
 248         fw_cfg_p_base = range->start;
 249         fw_cfg_p_size = resource_size(range);
 250
 251         if (fw_cfg_is_mmio) {
 252                 if (!request_mem_region(fw_cfg_p_base,
 253                                         fw_cfg_p_size, "fw_cfg_mem"))
 254                         return -EBUSY;
 255                 fw_cfg_dev_base = ioremap(fw_cfg_p_base, fw_cfg_p_size);
 256                 if (!fw_cfg_dev_base) {
 257                         release_mem_region(fw_cfg_p_base, fw_cfg_p_size);
 258                         return -EFAULT;
 259                 }
 260         } else {
 261                 if (!request_region(fw_cfg_p_base,
 262                                     fw_cfg_p_size, "fw_cfg_io"))
 263                         return -EBUSY;
 264                 fw_cfg_dev_base = ioport_map(fw_cfg_p_base, fw_cfg_p_size);
 265                 if (!fw_cfg_dev_base) {
 266                         release_region(fw_cfg_p_base, fw_cfg_p_size);
 267                         return -EFAULT;
 268                 }
 269         }
 270
 271         /* were custom register offsets provided (e.g. on the command line)? */
 272         ctrl = platform_get_resource_byname(pdev, IORESOURCE_REG, "ctrl");
 273         data = platform_get_resource_byname(pdev, IORESOURCE_REG, "data");
 274         dma = platform_get_resource_byname(pdev, IORESOURCE_REG, "dma");
 275         if (ctrl && data) {
 276                 fw_cfg_reg_ctrl = fw_cfg_dev_base + ctrl->start;
 277                 fw_cfg_reg_data = fw_cfg_dev_base + data->start;
 278         } else {
 279                 /* use architecture-specific offsets */
 280                 fw_cfg_reg_ctrl = fw_cfg_dev_base + FW_CFG_CTRL_OFF;
 281                 fw_cfg_reg_data = fw_cfg_dev_base + FW_CFG_DATA_OFF;
 282         }
 283
 284         if (dma)
 285                 fw_cfg_reg_dma = fw_cfg_dev_base + dma->start;
 286 #ifdef FW_CFG_DMA_OFF
 287         else
 288                 fw_cfg_reg_dma = fw_cfg_dev_base + FW_CFG_DMA_OFF;
 289 #endif
 290
 291         /* verify fw_cfg device signature */
 292         if (fw_cfg_read_blob(FW_CFG_SIGNATURE, sig,
 293                                 0, FW_CFG_SIG_SIZE) < 0 ||
 294                 memcmp(sig, "QEMU", FW_CFG_SIG_SIZE) != 0) {
 295                 fw_cfg_io_cleanup();
 296                 return -ENODEV;
 297         }
 298
 299         return 0;
 300 }
 301
 302 static ssize_t fw_cfg_showrev(struct kobject *k, struct attribute *a, char *buf)
 303 {
 304         return sprintf(buf, "%u\n", fw_cfg_rev);
 305 }
 306
 307 static const struct {
 308         struct attribute attr;
 309         ssize_t (*show)(struct kobject *k, struct attribute *a, char *buf);
 310 } fw_cfg_rev_attr = {
 311         .attr = { .name = "rev", .mode = S_IRUSR },
 312         .show = fw_cfg_showrev,
 313 };
 314
 315 /* fw_cfg_sysfs_entry type */
 316 struct fw_cfg_sysfs_entry {
 317         struct kobject kobj;
 318         u32 size;
 319         u16 select;
 320         char name[FW_CFG_MAX_FILE_PATH];
 321         struct list_head list;
 322 };
 323
 324 #ifdef CONFIG_CRASH_CORE
 325 static ssize_t fw_cfg_write_vmcoreinfo(const struct fw_cfg_file *f)
 326 {
 327         static struct fw_cfg_vmcoreinfo *data;
 328         ssize_t ret;
 329
 330         data = kmalloc(sizeof(struct fw_cfg_vmcoreinfo), GFP_KERNEL);
 331         if (!data)
 332                 return -ENOMEM;
 333
 334         *data = (struct fw_cfg_vmcoreinfo) {
 335                 .guest_format = cpu_to_le16(FW_CFG_VMCOREINFO_FORMAT_ELF),
 336                 .size = cpu_to_le32(VMCOREINFO_NOTE_SIZE),
 337                 .paddr = cpu_to_le64(paddr_vmcoreinfo_note())
 338         };
 339         /* spare ourself reading host format support for now since we
 340          * don't know what else to format - host may ignore ours
 341          */
 342         ret = fw_cfg_write_blob(be16_to_cpu(f->select), data,
 343                                 0, sizeof(struct fw_cfg_vmcoreinfo));
 344
 345         kfree(data);
 346         return ret;
 347 }
 348 #endif /* CONFIG_CRASH_CORE */
 349
 350 /* get fw_cfg_sysfs_entry from kobject member */
 351 static inline struct fw_cfg_sysfs_entry *to_entry(struct kobject *kobj)
 352 {
 353         return container_of(kobj, struct fw_cfg_sysfs_entry, kobj);
 354 }
 355
 356 /* fw_cfg_sysfs_attribute type */
 357 struct fw_cfg_sysfs_attribute {
 358         struct attribute attr;
 359         ssize_t (*show)(struct fw_cfg_sysfs_entry *entry, char *buf);
 360 };
 361
 362 /* get fw_cfg_sysfs_attribute from attribute member */
 363 static inline struct fw_cfg_sysfs_attribute *to_attr(struct attribute *attr)
 364 {
 365         return container_of(attr, struct fw_cfg_sysfs_attribute, attr);
 366 }
 367
 368 /* global cache of fw_cfg_sysfs_entry objects */
 369 static LIST_HEAD(fw_cfg_entry_cache);
 370
 371 /* kobjects removed lazily by kernel, mutual exclusion needed */
 372 static DEFINE_SPINLOCK(fw_cfg_cache_lock);
 373
 374 static inline void fw_cfg_sysfs_cache_enlist(struct fw_cfg_sysfs_entry *entry)
 375 {
 376         spin_lock(&fw_cfg_cache_lock);
 377         list_add_tail(&entry->list, &fw_cfg_entry_cache);
 378         spin_unlock(&fw_cfg_cache_lock);
 379 }
 380
 381 static inline void fw_cfg_sysfs_cache_delist(struct fw_cfg_sysfs_entry *entry)
 382 {
 383         spin_lock(&fw_cfg_cache_lock);
 384         list_del(&entry->list);
 385         spin_unlock(&fw_cfg_cache_lock);
 386 }
 387
 388 static void fw_cfg_sysfs_cache_cleanup(void)
 389 {
 390         struct fw_cfg_sysfs_entry *entry, *next;
 391
 392         list_for_each_entry_safe(entry, next, &fw_cfg_entry_cache, list) {
 393                 /* will end up invoking fw_cfg_sysfs_cache_delist()
 394                  * via each object's release() method (i.e. destructor)
 395                  */
 396                 kobject_put(&entry->kobj);
 397         }
 398 }
 399
 400 /* default_attrs: per-entry attributes and show methods */
 401
 402 #define FW_CFG_SYSFS_ATTR(_attr) \
 403 struct fw_cfg_sysfs_attribute fw_cfg_sysfs_attr_##_attr = { \
 404         .attr = { .name = __stringify(_attr), .mode = S_IRUSR }, \
 405         .show = fw_cfg_sysfs_show_##_attr, \
 406 }
 407
 408 static ssize_t fw_cfg_sysfs_show_size(struct fw_cfg_sysfs_entry *e, char *buf)
 409 {
 410         return sprintf(buf, "%u\n", e->size);
 411 }
 412
 413 static ssize_t fw_cfg_sysfs_show_key(struct fw_cfg_sysfs_entry *e, char *buf)
 414 {
 415         return sprintf(buf, "%u\n", e->select);
 416 }
 417
 418 static ssize_t fw_cfg_sysfs_show_name(struct fw_cfg_sysfs_entry *e, char *buf)
 419 {
 420         return sprintf(buf, "%s\n", e->name);
 421 }
 422
 423 static FW_CFG_SYSFS_ATTR(size);
 424 static FW_CFG_SYSFS_ATTR(key);
 425 static FW_CFG_SYSFS_ATTR(name);
 426
 427 static struct attribute *fw_cfg_sysfs_entry_attrs[] = {
 428         &fw_cfg_sysfs_attr_size.attr,
 429         &fw_cfg_sysfs_attr_key.attr,
 430         &fw_cfg_sysfs_attr_name.attr,
 431         NULL,
 432 };
 433
 434 /* sysfs_ops: find fw_cfg_[entry, attribute] and call appropriate show method */
 435 static ssize_t fw_cfg_sysfs_attr_show(struct kobject *kobj, struct attribute *a,
 436                                       char *buf)
 437 {
 438         struct fw_cfg_sysfs_entry *entry = to_entry(kobj);
 439         struct fw_cfg_sysfs_attribute *attr = to_attr(a);
 440
 441         return attr->show(entry, buf);
 442 }
 443
 444 static const struct sysfs_ops fw_cfg_sysfs_attr_ops = {
 445         .show = fw_cfg_sysfs_attr_show,
 446 };
 447
 448 /* release: destructor, to be called via kobject_put() */
 449 static void fw_cfg_sysfs_release_entry(struct kobject *kobj)
 450 {
 451         struct fw_cfg_sysfs_entry *entry = to_entry(kobj);
 452
 453         fw_cfg_sysfs_cache_delist(entry);
 454         kfree(entry);
 455 }
 456
 457 /* kobj_type: ties together all properties required to register an entry */
 458 static struct kobj_type fw_cfg_sysfs_entry_ktype = {
 459         .default_attrs = fw_cfg_sysfs_entry_attrs,
 460         .sysfs_ops = &fw_cfg_sysfs_attr_ops,
 461         .release = fw_cfg_sysfs_release_entry,
 462 };
 463
 464 /* raw-read method and attribute */
 465 static ssize_t fw_cfg_sysfs_read_raw(struct file *filp, struct kobject *kobj,
 466                                      struct bin_attribute *bin_attr,
 467                                      char *buf, loff_t pos, size_t count)
 468 {
 469         struct fw_cfg_sysfs_entry *entry = to_entry(kobj);
 470
 471         if (pos > entry->size)
 472                 return -EINVAL;
 473
 474         if (count > entry->size - pos)
 475                 count = entry->size - pos;
 476
 477         return fw_cfg_read_blob(entry->select, buf, pos, count);
 478 }
 479
 480 static struct bin_attribute fw_cfg_sysfs_attr_raw = {
 481         .attr = { .name = "raw", .mode = S_IRUSR },
 482         .read = fw_cfg_sysfs_read_raw,
 483 };
 484
 485 /*
 486  * Create a kset subdirectory matching each '/' delimited dirname token
 487  * in 'name', starting with sysfs kset/folder 'dir'; At the end, create
 488  * a symlink directed at the given 'target'.
 489  * NOTE: We do this on a best-effort basis, since 'name' is not guaranteed
 490  * to be a well-behaved path name. Whenever a symlink vs. kset directory
 491  * name collision occurs, the kernel will issue big scary warnings while
 492  * refusing to add the offending link or directory. We follow up with our
 493  * own, slightly less scary error messages explaining the situation :)
 494  */
 495 static int fw_cfg_build_symlink(struct kset *dir,
 496                                 struct kobject *target, const char *name)
 497 {
 498         int ret;
 499         struct kset *subdir;
 500         struct kobject *ko;
 501         char *name_copy, *p, *tok;
 502
 503         if (!dir || !target || !name || !*name)
 504                 return -EINVAL;
 505
 506         /* clone a copy of name for parsing */
 507         name_copy = p = kstrdup(name, GFP_KERNEL);
 508         if (!name_copy)
 509                 return -ENOMEM;
 510
 511         /* create folders for each dirname token, then symlink for basename */
 512         while ((tok = strsep(&p, "/")) && *tok) {
 513
 514                 /* last (basename) token? If so, add symlink here */
 515                 if (!p || !*p) {
 516                         ret = sysfs_create_link(&dir->kobj, target, tok);
 517                         break;
 518                 }
 519
 520                 /* does the current dir contain an item named after tok ? */
 521                 ko = kset_find_obj(dir, tok);
 522                 if (ko) {
 523                         /* drop reference added by kset_find_obj */
 524                         kobject_put(ko);
 525
 526                         /* ko MUST be a kset - we're about to use it as one ! */
 527                         if (ko->ktype != dir->kobj.ktype) {
 528                                 ret = -EINVAL;
 529                                 break;
 530                         }
 531
 532                         /* descend into already existing subdirectory */
 533                         dir = to_kset(ko);
 534                 } else {
 535                         /* create new subdirectory kset */
 536                         subdir = kzalloc(sizeof(struct kset), GFP_KERNEL);
 537                         if (!subdir) {
 538                                 ret = -ENOMEM;
 539                                 break;
 540                         }
 541                         subdir->kobj.kset = dir;
 542                         subdir->kobj.ktype = dir->kobj.ktype;
 543                         ret = kobject_set_name(&subdir->kobj, "%s", tok);
 544                         if (ret) {
 545                                 kfree(subdir);
 546                                 break;
 547                         }
 548                         ret = kset_register(subdir);
 549                         if (ret) {
 550                                 kfree(subdir);
 551                                 break;
 552                         }
 553
 554                         /* descend into newly created subdirectory */
 555                         dir = subdir;
 556                 }
 557         }
 558
 559         /* we're done with cloned copy of name */
 560         kfree(name_copy);
 561         return ret;
 562 }
 563
 564 /* recursively unregister fw_cfg/by_name/ kset directory tree */
 565 static void fw_cfg_kset_unregister_recursive(struct kset *kset)
 566 {
 567         struct kobject *k, *next;
 568
 569         list_for_each_entry_safe(k, next, &kset->list, entry)
 570                 /* all set members are ksets too, but check just in case... */
 571                 if (k->ktype == kset->kobj.ktype)
 572                         fw_cfg_kset_unregister_recursive(to_kset(k));
 573
 574         /* symlinks are cleanly and automatically removed with the directory */
 575         kset_unregister(kset);
 576 }
 577
 578 /* kobjects & kset representing top-level, by_key, and by_name folders */
 579 static struct kobject *fw_cfg_top_ko;
 580 static struct kobject *fw_cfg_sel_ko;
 581 static struct kset *fw_cfg_fname_kset;
 582
 583 /* register an individual fw_cfg file */
 584 static int fw_cfg_register_file(const struct fw_cfg_file *f)
 585 {
 586         int err;
 587         struct fw_cfg_sysfs_entry *entry;
 588
 589 #ifdef CONFIG_CRASH_CORE
 590         if (fw_cfg_dma_enabled() &&
 591                 strcmp(f->name, FW_CFG_VMCOREINFO_FILENAME) == 0 &&
 592                 !is_kdump_kernel()) {
 593                 if (fw_cfg_write_vmcoreinfo(f) < 0)
 594                         pr_warn("fw_cfg: failed to write vmcoreinfo");
 595         }
 596 #endif
 597
 598         /* allocate new entry */
 599         entry = kzalloc(sizeof(*entry), GFP_KERNEL);
 600         if (!entry)
 601                 return -ENOMEM;
 602
 603         /* set file entry information */
 604         entry->size = be32_to_cpu(f->size);
 605         entry->select = be16_to_cpu(f->select);
 606         memcpy(entry->name, f->name, FW_CFG_MAX_FILE_PATH);
 607
 608         /* register entry under "/sys/firmware/qemu_fw_cfg/by_key/" */
 609         err = kobject_init_and_add(&entry->kobj, &fw_cfg_sysfs_entry_ktype,
 610                                    fw_cfg_sel_ko, "%d", entry->select);
 611         if (err) {
 612                 kobject_put(&entry->kobj);
 613                 return err;
 614         }
 615
 616         /* add raw binary content access */
 617         err = sysfs_create_bin_file(&entry->kobj, &fw_cfg_sysfs_attr_raw);
 618         if (err)
 619                 goto err_add_raw;
 620
 621         /* try adding "/sys/firmware/qemu_fw_cfg/by_name/" symlink */
 622         fw_cfg_build_symlink(fw_cfg_fname_kset, &entry->kobj, entry->name);
 623
 624         /* success, add entry to global cache */
 625         fw_cfg_sysfs_cache_enlist(entry);
 626         return 0;
 627
 628 err_add_raw:
 629         kobject_del(&entry->kobj);
 630         kfree(entry);
 631         return err;
 632 }
 633
 634 /* iterate over all fw_cfg directory entries, registering each one */
 635 static int fw_cfg_register_dir_entries(void)
 636 {
 637         int ret = 0;
 638         __be32 files_count;
 639         u32 count, i;
 640         struct fw_cfg_file *dir;
 641         size_t dir_size;
 642
 643         ret = fw_cfg_read_blob(FW_CFG_FILE_DIR, &files_count,
 644                         0, sizeof(files_count));
 645         if (ret < 0)
 646                 return ret;
 647
 648         count = be32_to_cpu(files_count);
 649         dir_size = count * sizeof(struct fw_cfg_file);
 650
 651         dir = kmalloc(dir_size, GFP_KERNEL);
 652         if (!dir)
 653                 return -ENOMEM;
 654
 655         ret = fw_cfg_read_blob(FW_CFG_FILE_DIR, dir,
 656                         sizeof(files_count), dir_size);
 657         if (ret < 0)
 658                 goto end;
 659
 660         for (i = 0; i < count; i++) {
 661                 ret = fw_cfg_register_file(&dir[i]);
 662                 if (ret)
 663                         break;
 664         }
 665
 666 end:
 667         kfree(dir);
 668         return ret;
 669 }
 670
 671 /* unregister top-level or by_key folder */
 672 static inline void fw_cfg_kobj_cleanup(struct kobject *kobj)
 673 {
 674         kobject_del(kobj);
 675         kobject_put(kobj);
 676 }
 677
 678 static int fw_cfg_sysfs_probe(struct platform_device *pdev)
 679 {
 680         int err;
 681         __le32 rev;
 682
 683         /* NOTE: If we supported multiple fw_cfg devices, we'd first create
 684          * a subdirectory named after e.g. pdev->id, then hang per-device
 685          * by_key (and by_name) subdirectories underneath it. However, only
 686          * one fw_cfg device exist system-wide, so if one was already found
 687          * earlier, we might as well stop here.
 688          */
 689         if (fw_cfg_sel_ko)
 690                 return -EBUSY;
 691
 692         /* create by_key and by_name subdirs of /sys/firmware/qemu_fw_cfg/ */
 693         err = -ENOMEM;
 694         fw_cfg_sel_ko = kobject_create_and_add("by_key", fw_cfg_top_ko);
 695         if (!fw_cfg_sel_ko)
 696                 goto err_sel;
 697         fw_cfg_fname_kset = kset_create_and_add("by_name", NULL, fw_cfg_top_ko);
 698         if (!fw_cfg_fname_kset)
 699                 goto err_name;
 700
 701         /* initialize fw_cfg device i/o from platform data */
 702         err = fw_cfg_do_platform_probe(pdev);
 703         if (err)
 704                 goto err_probe;
 705
 706         /* get revision number, add matching top-level attribute */
 707         err = fw_cfg_read_blob(FW_CFG_ID, &rev, 0, sizeof(rev));
 708         if (err < 0)
 709                 goto err_probe;
 710
 711         fw_cfg_rev = le32_to_cpu(rev);
 712         err = sysfs_create_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr);
 713         if (err)
 714                 goto err_rev;
 715
 716         /* process fw_cfg file directory entry, registering each file */
 717         err = fw_cfg_register_dir_entries();
 718         if (err)
 719                 goto err_dir;
 720
 721         /* success */
 722         pr_debug("fw_cfg: loaded.\n");
 723         return 0;
 724
 725 err_dir:
 726         fw_cfg_sysfs_cache_cleanup();
 727         sysfs_remove_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr);
 728 err_rev:
 729         fw_cfg_io_cleanup();
 730 err_probe:
 731         fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset);
 732 err_name:
 733         fw_cfg_kobj_cleanup(fw_cfg_sel_ko);
 734 err_sel:
 735         return err;
 736 }
 737
 738 static int fw_cfg_sysfs_remove(struct platform_device *pdev)
 739 {
 740         pr_debug("fw_cfg: unloading.\n");
 741         fw_cfg_sysfs_cache_cleanup();
 742         sysfs_remove_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr);
 743         fw_cfg_io_cleanup();
 744         fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset);
 745         fw_cfg_kobj_cleanup(fw_cfg_sel_ko);
 746         return 0;
 747 }
 748
 749 static const struct of_device_id fw_cfg_sysfs_mmio_match[] = {
 750         { .compatible = "qemu,fw-cfg-mmio", },
 751         {},
 752 };
 753 MODULE_DEVICE_TABLE(of, fw_cfg_sysfs_mmio_match);
 754
 755 #ifdef CONFIG_ACPI
 756 static const struct acpi_device_id fw_cfg_sysfs_acpi_match[] = {
 757         { FW_CFG_ACPI_DEVICE_ID, },
 758         {},
 759 };
 760 MODULE_DEVICE_TABLE(acpi, fw_cfg_sysfs_acpi_match);
 761 #endif
 762
 763 static struct platform_driver fw_cfg_sysfs_driver = {
 764         .probe = fw_cfg_sysfs_probe,
 765         .remove = fw_cfg_sysfs_remove,
 766         .driver = {
 767                 .name = "fw_cfg",
 768                 .of_match_table = fw_cfg_sysfs_mmio_match,
 769                 .acpi_match_table = ACPI_PTR(fw_cfg_sysfs_acpi_match),
 770         },
 771 };
 772
 773 #ifdef CONFIG_FW_CFG_SYSFS_CMDLINE
 774
 775 static struct platform_device *fw_cfg_cmdline_dev;
 776
 777 /* this probably belongs in e.g. include/linux/types.h,
 778  * but right now we are the only ones doing it...
 779  */
 780 #ifdef CONFIG_PHYS_ADDR_T_64BIT
 781 #define __PHYS_ADDR_PREFIX "ll"
 782 #else
 783 #define __PHYS_ADDR_PREFIX ""
 784 #endif
 785
 786 /* use special scanf/printf modifier for phys_addr_t, resource_size_t */
 787 #define PH_ADDR_SCAN_FMT "@%" __PHYS_ADDR_PREFIX "i%n" \
 788                          ":%" __PHYS_ADDR_PREFIX "i" \
 789                          ":%" __PHYS_ADDR_PREFIX "i%n" \
 790                          ":%" __PHYS_ADDR_PREFIX "i%n"
 791
 792 #define PH_ADDR_PR_1_FMT "0x%" __PHYS_ADDR_PREFIX "x@" \
 793                          "0x%" __PHYS_ADDR_PREFIX "x"
 794
 795 #define PH_ADDR_PR_3_FMT PH_ADDR_PR_1_FMT \
 796                          ":%" __PHYS_ADDR_PREFIX "u" \
 797                          ":%" __PHYS_ADDR_PREFIX "u"
 798
 799 #define PH_ADDR_PR_4_FMT PH_ADDR_PR_3_FMT \
 800                          ":%" __PHYS_ADDR_PREFIX "u"
 801
 802 static int fw_cfg_cmdline_set(const char *arg, const struct kernel_param *kp)
 803 {
 804         struct resource res[4] = {};
 805         char *str;
 806         phys_addr_t base;
 807         resource_size_t size, ctrl_off, data_off, dma_off;
 808         int processed, consumed = 0;
 809
 810         /* only one fw_cfg device can exist system-wide, so if one
 811          * was processed on the command line already, we might as
 812          * well stop here.
 813          */
 814         if (fw_cfg_cmdline_dev) {
 815                 /* avoid leaking previously registered device */
 816                 platform_device_unregister(fw_cfg_cmdline_dev);
 817                 return -EINVAL;
 818         }
 819
 820         /* consume "<size>" portion of command line argument */
 821         size = memparse(arg, &str);
 822
 823         /* get "@<base>[:<ctrl_off>:<data_off>[:<dma_off>]]" chunks */
 824         processed = sscanf(str, PH_ADDR_SCAN_FMT,
 825                            &base, &consumed,
 826                            &ctrl_off, &data_off, &consumed,
 827                            &dma_off, &consumed);
 828
 829         /* sscanf() must process precisely 1, 3 or 4 chunks:
 830          * <base> is mandatory, optionally followed by <ctrl_off>
 831          * and <data_off>, and <dma_off>;
 832          * there must be no extra characters after the last chunk,
 833          * so str[consumed] must be '\0'.
 834          */
 835         if (str[consumed] ||
 836             (processed != 1 && processed != 3 && processed != 4))
 837                 return -EINVAL;
 838
 839         res[0].start = base;
 840         res[0].end = base + size - 1;
 841         res[0].flags = !strcmp(kp->name, "mmio") ? IORESOURCE_MEM :
 842                                                    IORESOURCE_IO;
 843
 844         /* insert register offsets, if provided */
 845         if (processed > 1) {
 846                 res[1].name = "ctrl";
 847                 res[1].start = ctrl_off;
 848                 res[1].flags = IORESOURCE_REG;
 849                 res[2].name = "data";
 850                 res[2].start = data_off;
 851                 res[2].flags = IORESOURCE_REG;
 852         }
 853         if (processed > 3) {
 854                 res[3].name = "dma";
 855                 res[3].start = dma_off;
 856                 res[3].flags = IORESOURCE_REG;
 857         }
 858
 859         /* "processed" happens to nicely match the number of resources
 860          * we need to pass in to this platform device.
 861          */
 862         fw_cfg_cmdline_dev = platform_device_register_simple("fw_cfg",
 863                                         PLATFORM_DEVID_NONE, res, processed);
 864
 865         return PTR_ERR_OR_ZERO(fw_cfg_cmdline_dev);
 866 }
 867
 868 static int fw_cfg_cmdline_get(char *buf, const struct kernel_param *kp)
 869 {
 870         /* stay silent if device was not configured via the command
 871          * line, or if the parameter name (ioport/mmio) doesn't match
 872          * the device setting
 873          */
 874         if (!fw_cfg_cmdline_dev ||
 875             (!strcmp(kp->name, "mmio") ^
 876              (fw_cfg_cmdline_dev->resource[0].flags == IORESOURCE_MEM)))
 877                 return 0;
 878
 879         switch (fw_cfg_cmdline_dev->num_resources) {
 880         case 1:
 881                 return snprintf(buf, PAGE_SIZE, PH_ADDR_PR_1_FMT,
 882                                 resource_size(&fw_cfg_cmdline_dev->resource[0]),
 883                                 fw_cfg_cmdline_dev->resource[0].start);
 884         case 3:
 885                 return snprintf(buf, PAGE_SIZE, PH_ADDR_PR_3_FMT,
 886                                 resource_size(&fw_cfg_cmdline_dev->resource[0]),
 887                                 fw_cfg_cmdline_dev->resource[0].start,
 888                                 fw_cfg_cmdline_dev->resource[1].start,
 889                                 fw_cfg_cmdline_dev->resource[2].start);
 890         case 4:
 891                 return snprintf(buf, PAGE_SIZE, PH_ADDR_PR_4_FMT,
 892                                 resource_size(&fw_cfg_cmdline_dev->resource[0]),
 893                                 fw_cfg_cmdline_dev->resource[0].start,
 894                                 fw_cfg_cmdline_dev->resource[1].start,
 895                                 fw_cfg_cmdline_dev->resource[2].start,
 896                                 fw_cfg_cmdline_dev->resource[3].start);
 897         }
 898
 899         /* Should never get here */
 900         WARN(1, "Unexpected number of resources: %d\n",
 901                 fw_cfg_cmdline_dev->num_resources);
 902         return 0;
 903 }
 904
 905 static const struct kernel_param_ops fw_cfg_cmdline_param_ops = {
 906         .set = fw_cfg_cmdline_set,
 907         .get = fw_cfg_cmdline_get,
 908 };
 909
 910 device_param_cb(ioport, &fw_cfg_cmdline_param_ops, NULL, S_IRUSR);
 911 device_param_cb(mmio, &fw_cfg_cmdline_param_ops, NULL, S_IRUSR);
 912
 913 #endif /* CONFIG_FW_CFG_SYSFS_CMDLINE */
 914
 915 static int __init fw_cfg_sysfs_init(void)
 916 {
 917         int ret;
 918
 919         /* create /sys/firmware/qemu_fw_cfg/ top level directory */
 920         fw_cfg_top_ko = kobject_create_and_add("qemu_fw_cfg", firmware_kobj);
 921         if (!fw_cfg_top_ko)
 922                 return -ENOMEM;
 923
 924         ret = platform_driver_register(&fw_cfg_sysfs_driver);
 925         if (ret)
 926                 fw_cfg_kobj_cleanup(fw_cfg_top_ko);
 927
 928         return ret;
 929 }
 930
 931 static void __exit fw_cfg_sysfs_exit(void)
 932 {
 933         platform_driver_unregister(&fw_cfg_sysfs_driver);
 934
 935 #ifdef CONFIG_FW_CFG_SYSFS_CMDLINE
 936         platform_device_unregister(fw_cfg_cmdline_dev);
 937 #endif
 938
 939         /* clean up /sys/firmware/qemu_fw_cfg/ */
 940         fw_cfg_kobj_cleanup(fw_cfg_top_ko);
 941 }
 942
 943 module_init(fw_cfg_sysfs_init);
 944 module_exit(fw_cfg_sysfs_exit);