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 kobj_attribute *a,
303 			      char *buf)
304 {
305 	return sprintf(buf, "%u\n", fw_cfg_rev);
306 }
307 
308 static const struct kobj_attribute fw_cfg_rev_attr = {
309 	.attr = { .name = "rev", .mode = S_IRUSR },
310 	.show = fw_cfg_showrev,
311 };
312 
313 /* fw_cfg_sysfs_entry type */
314 struct fw_cfg_sysfs_entry {
315 	struct kobject kobj;
316 	u32 size;
317 	u16 select;
318 	char name[FW_CFG_MAX_FILE_PATH];
319 	struct list_head list;
320 };
321 
322 #ifdef CONFIG_CRASH_CORE
323 static ssize_t fw_cfg_write_vmcoreinfo(const struct fw_cfg_file *f)
324 {
325 	static struct fw_cfg_vmcoreinfo *data;
326 	ssize_t ret;
327 
328 	data = kmalloc(sizeof(struct fw_cfg_vmcoreinfo), GFP_KERNEL);
329 	if (!data)
330 		return -ENOMEM;
331 
332 	*data = (struct fw_cfg_vmcoreinfo) {
333 		.guest_format = cpu_to_le16(FW_CFG_VMCOREINFO_FORMAT_ELF),
334 		.size = cpu_to_le32(VMCOREINFO_NOTE_SIZE),
335 		.paddr = cpu_to_le64(paddr_vmcoreinfo_note())
336 	};
337 	/* spare ourself reading host format support for now since we
338 	 * don't know what else to format - host may ignore ours
339 	 */
340 	ret = fw_cfg_write_blob(be16_to_cpu(f->select), data,
341 				0, sizeof(struct fw_cfg_vmcoreinfo));
342 
343 	kfree(data);
344 	return ret;
345 }
346 #endif /* CONFIG_CRASH_CORE */
347 
348 /* get fw_cfg_sysfs_entry from kobject member */
349 static inline struct fw_cfg_sysfs_entry *to_entry(struct kobject *kobj)
350 {
351 	return container_of(kobj, struct fw_cfg_sysfs_entry, kobj);
352 }
353 
354 /* fw_cfg_sysfs_attribute type */
355 struct fw_cfg_sysfs_attribute {
356 	struct attribute attr;
357 	ssize_t (*show)(struct fw_cfg_sysfs_entry *entry, char *buf);
358 };
359 
360 /* get fw_cfg_sysfs_attribute from attribute member */
361 static inline struct fw_cfg_sysfs_attribute *to_attr(struct attribute *attr)
362 {
363 	return container_of(attr, struct fw_cfg_sysfs_attribute, attr);
364 }
365 
366 /* global cache of fw_cfg_sysfs_entry objects */
367 static LIST_HEAD(fw_cfg_entry_cache);
368 
369 /* kobjects removed lazily by kernel, mutual exclusion needed */
370 static DEFINE_SPINLOCK(fw_cfg_cache_lock);
371 
372 static inline void fw_cfg_sysfs_cache_enlist(struct fw_cfg_sysfs_entry *entry)
373 {
374 	spin_lock(&fw_cfg_cache_lock);
375 	list_add_tail(&entry->list, &fw_cfg_entry_cache);
376 	spin_unlock(&fw_cfg_cache_lock);
377 }
378 
379 static inline void fw_cfg_sysfs_cache_delist(struct fw_cfg_sysfs_entry *entry)
380 {
381 	spin_lock(&fw_cfg_cache_lock);
382 	list_del(&entry->list);
383 	spin_unlock(&fw_cfg_cache_lock);
384 }
385 
386 static void fw_cfg_sysfs_cache_cleanup(void)
387 {
388 	struct fw_cfg_sysfs_entry *entry, *next;
389 
390 	list_for_each_entry_safe(entry, next, &fw_cfg_entry_cache, list) {
391 		fw_cfg_sysfs_cache_delist(entry);
392 		kobject_put(&entry->kobj);
393 	}
394 }
395 
396 /* default_attrs: per-entry attributes and show methods */
397 
398 #define FW_CFG_SYSFS_ATTR(_attr) \
399 struct fw_cfg_sysfs_attribute fw_cfg_sysfs_attr_##_attr = { \
400 	.attr = { .name = __stringify(_attr), .mode = S_IRUSR }, \
401 	.show = fw_cfg_sysfs_show_##_attr, \
402 }
403 
404 static ssize_t fw_cfg_sysfs_show_size(struct fw_cfg_sysfs_entry *e, char *buf)
405 {
406 	return sprintf(buf, "%u\n", e->size);
407 }
408 
409 static ssize_t fw_cfg_sysfs_show_key(struct fw_cfg_sysfs_entry *e, char *buf)
410 {
411 	return sprintf(buf, "%u\n", e->select);
412 }
413 
414 static ssize_t fw_cfg_sysfs_show_name(struct fw_cfg_sysfs_entry *e, char *buf)
415 {
416 	return sprintf(buf, "%s\n", e->name);
417 }
418 
419 static FW_CFG_SYSFS_ATTR(size);
420 static FW_CFG_SYSFS_ATTR(key);
421 static FW_CFG_SYSFS_ATTR(name);
422 
423 static struct attribute *fw_cfg_sysfs_entry_attrs[] = {
424 	&fw_cfg_sysfs_attr_size.attr,
425 	&fw_cfg_sysfs_attr_key.attr,
426 	&fw_cfg_sysfs_attr_name.attr,
427 	NULL,
428 };
429 
430 /* sysfs_ops: find fw_cfg_[entry, attribute] and call appropriate show method */
431 static ssize_t fw_cfg_sysfs_attr_show(struct kobject *kobj, struct attribute *a,
432 				      char *buf)
433 {
434 	struct fw_cfg_sysfs_entry *entry = to_entry(kobj);
435 	struct fw_cfg_sysfs_attribute *attr = to_attr(a);
436 
437 	return attr->show(entry, buf);
438 }
439 
440 static const struct sysfs_ops fw_cfg_sysfs_attr_ops = {
441 	.show = fw_cfg_sysfs_attr_show,
442 };
443 
444 /* release: destructor, to be called via kobject_put() */
445 static void fw_cfg_sysfs_release_entry(struct kobject *kobj)
446 {
447 	struct fw_cfg_sysfs_entry *entry = to_entry(kobj);
448 
449 	kfree(entry);
450 }
451 
452 /* kobj_type: ties together all properties required to register an entry */
453 static struct kobj_type fw_cfg_sysfs_entry_ktype = {
454 	.default_attrs = fw_cfg_sysfs_entry_attrs,
455 	.sysfs_ops = &fw_cfg_sysfs_attr_ops,
456 	.release = fw_cfg_sysfs_release_entry,
457 };
458 
459 /* raw-read method and attribute */
460 static ssize_t fw_cfg_sysfs_read_raw(struct file *filp, struct kobject *kobj,
461 				     struct bin_attribute *bin_attr,
462 				     char *buf, loff_t pos, size_t count)
463 {
464 	struct fw_cfg_sysfs_entry *entry = to_entry(kobj);
465 
466 	if (pos > entry->size)
467 		return -EINVAL;
468 
469 	if (count > entry->size - pos)
470 		count = entry->size - pos;
471 
472 	return fw_cfg_read_blob(entry->select, buf, pos, count);
473 }
474 
475 static struct bin_attribute fw_cfg_sysfs_attr_raw = {
476 	.attr = { .name = "raw", .mode = S_IRUSR },
477 	.read = fw_cfg_sysfs_read_raw,
478 };
479 
480 /*
481  * Create a kset subdirectory matching each '/' delimited dirname token
482  * in 'name', starting with sysfs kset/folder 'dir'; At the end, create
483  * a symlink directed at the given 'target'.
484  * NOTE: We do this on a best-effort basis, since 'name' is not guaranteed
485  * to be a well-behaved path name. Whenever a symlink vs. kset directory
486  * name collision occurs, the kernel will issue big scary warnings while
487  * refusing to add the offending link or directory. We follow up with our
488  * own, slightly less scary error messages explaining the situation :)
489  */
490 static int fw_cfg_build_symlink(struct kset *dir,
491 				struct kobject *target, const char *name)
492 {
493 	int ret;
494 	struct kset *subdir;
495 	struct kobject *ko;
496 	char *name_copy, *p, *tok;
497 
498 	if (!dir || !target || !name || !*name)
499 		return -EINVAL;
500 
501 	/* clone a copy of name for parsing */
502 	name_copy = p = kstrdup(name, GFP_KERNEL);
503 	if (!name_copy)
504 		return -ENOMEM;
505 
506 	/* create folders for each dirname token, then symlink for basename */
507 	while ((tok = strsep(&p, "/")) && *tok) {
508 
509 		/* last (basename) token? If so, add symlink here */
510 		if (!p || !*p) {
511 			ret = sysfs_create_link(&dir->kobj, target, tok);
512 			break;
513 		}
514 
515 		/* does the current dir contain an item named after tok ? */
516 		ko = kset_find_obj(dir, tok);
517 		if (ko) {
518 			/* drop reference added by kset_find_obj */
519 			kobject_put(ko);
520 
521 			/* ko MUST be a kset - we're about to use it as one ! */
522 			if (ko->ktype != dir->kobj.ktype) {
523 				ret = -EINVAL;
524 				break;
525 			}
526 
527 			/* descend into already existing subdirectory */
528 			dir = to_kset(ko);
529 		} else {
530 			/* create new subdirectory kset */
531 			subdir = kzalloc(sizeof(struct kset), GFP_KERNEL);
532 			if (!subdir) {
533 				ret = -ENOMEM;
534 				break;
535 			}
536 			subdir->kobj.kset = dir;
537 			subdir->kobj.ktype = dir->kobj.ktype;
538 			ret = kobject_set_name(&subdir->kobj, "%s", tok);
539 			if (ret) {
540 				kfree(subdir);
541 				break;
542 			}
543 			ret = kset_register(subdir);
544 			if (ret) {
545 				kfree(subdir);
546 				break;
547 			}
548 
549 			/* descend into newly created subdirectory */
550 			dir = subdir;
551 		}
552 	}
553 
554 	/* we're done with cloned copy of name */
555 	kfree(name_copy);
556 	return ret;
557 }
558 
559 /* recursively unregister fw_cfg/by_name/ kset directory tree */
560 static void fw_cfg_kset_unregister_recursive(struct kset *kset)
561 {
562 	struct kobject *k, *next;
563 
564 	list_for_each_entry_safe(k, next, &kset->list, entry)
565 		/* all set members are ksets too, but check just in case... */
566 		if (k->ktype == kset->kobj.ktype)
567 			fw_cfg_kset_unregister_recursive(to_kset(k));
568 
569 	/* symlinks are cleanly and automatically removed with the directory */
570 	kset_unregister(kset);
571 }
572 
573 /* kobjects & kset representing top-level, by_key, and by_name folders */
574 static struct kobject *fw_cfg_top_ko;
575 static struct kobject *fw_cfg_sel_ko;
576 static struct kset *fw_cfg_fname_kset;
577 
578 /* register an individual fw_cfg file */
579 static int fw_cfg_register_file(const struct fw_cfg_file *f)
580 {
581 	int err;
582 	struct fw_cfg_sysfs_entry *entry;
583 
584 #ifdef CONFIG_CRASH_CORE
585 	if (fw_cfg_dma_enabled() &&
586 		strcmp(f->name, FW_CFG_VMCOREINFO_FILENAME) == 0 &&
587 		!is_kdump_kernel()) {
588 		if (fw_cfg_write_vmcoreinfo(f) < 0)
589 			pr_warn("fw_cfg: failed to write vmcoreinfo");
590 	}
591 #endif
592 
593 	/* allocate new entry */
594 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
595 	if (!entry)
596 		return -ENOMEM;
597 
598 	/* set file entry information */
599 	entry->size = be32_to_cpu(f->size);
600 	entry->select = be16_to_cpu(f->select);
601 	strscpy(entry->name, f->name, FW_CFG_MAX_FILE_PATH);
602 
603 	/* register entry under "/sys/firmware/qemu_fw_cfg/by_key/" */
604 	err = kobject_init_and_add(&entry->kobj, &fw_cfg_sysfs_entry_ktype,
605 				   fw_cfg_sel_ko, "%d", entry->select);
606 	if (err)
607 		goto err_put_entry;
608 
609 	/* add raw binary content access */
610 	err = sysfs_create_bin_file(&entry->kobj, &fw_cfg_sysfs_attr_raw);
611 	if (err)
612 		goto err_del_entry;
613 
614 	/* try adding "/sys/firmware/qemu_fw_cfg/by_name/" symlink */
615 	fw_cfg_build_symlink(fw_cfg_fname_kset, &entry->kobj, entry->name);
616 
617 	/* success, add entry to global cache */
618 	fw_cfg_sysfs_cache_enlist(entry);
619 	return 0;
620 
621 err_del_entry:
622 	kobject_del(&entry->kobj);
623 err_put_entry:
624 	kobject_put(&entry->kobj);
625 	return err;
626 }
627 
628 /* iterate over all fw_cfg directory entries, registering each one */
629 static int fw_cfg_register_dir_entries(void)
630 {
631 	int ret = 0;
632 	__be32 files_count;
633 	u32 count, i;
634 	struct fw_cfg_file *dir;
635 	size_t dir_size;
636 
637 	ret = fw_cfg_read_blob(FW_CFG_FILE_DIR, &files_count,
638 			0, sizeof(files_count));
639 	if (ret < 0)
640 		return ret;
641 
642 	count = be32_to_cpu(files_count);
643 	dir_size = count * sizeof(struct fw_cfg_file);
644 
645 	dir = kmalloc(dir_size, GFP_KERNEL);
646 	if (!dir)
647 		return -ENOMEM;
648 
649 	ret = fw_cfg_read_blob(FW_CFG_FILE_DIR, dir,
650 			sizeof(files_count), dir_size);
651 	if (ret < 0)
652 		goto end;
653 
654 	for (i = 0; i < count; i++) {
655 		ret = fw_cfg_register_file(&dir[i]);
656 		if (ret)
657 			break;
658 	}
659 
660 end:
661 	kfree(dir);
662 	return ret;
663 }
664 
665 /* unregister top-level or by_key folder */
666 static inline void fw_cfg_kobj_cleanup(struct kobject *kobj)
667 {
668 	kobject_del(kobj);
669 	kobject_put(kobj);
670 }
671 
672 static int fw_cfg_sysfs_probe(struct platform_device *pdev)
673 {
674 	int err;
675 	__le32 rev;
676 
677 	/* NOTE: If we supported multiple fw_cfg devices, we'd first create
678 	 * a subdirectory named after e.g. pdev->id, then hang per-device
679 	 * by_key (and by_name) subdirectories underneath it. However, only
680 	 * one fw_cfg device exist system-wide, so if one was already found
681 	 * earlier, we might as well stop here.
682 	 */
683 	if (fw_cfg_sel_ko)
684 		return -EBUSY;
685 
686 	/* create by_key and by_name subdirs of /sys/firmware/qemu_fw_cfg/ */
687 	err = -ENOMEM;
688 	fw_cfg_sel_ko = kobject_create_and_add("by_key", fw_cfg_top_ko);
689 	if (!fw_cfg_sel_ko)
690 		goto err_sel;
691 	fw_cfg_fname_kset = kset_create_and_add("by_name", NULL, fw_cfg_top_ko);
692 	if (!fw_cfg_fname_kset)
693 		goto err_name;
694 
695 	/* initialize fw_cfg device i/o from platform data */
696 	err = fw_cfg_do_platform_probe(pdev);
697 	if (err)
698 		goto err_probe;
699 
700 	/* get revision number, add matching top-level attribute */
701 	err = fw_cfg_read_blob(FW_CFG_ID, &rev, 0, sizeof(rev));
702 	if (err < 0)
703 		goto err_probe;
704 
705 	fw_cfg_rev = le32_to_cpu(rev);
706 	err = sysfs_create_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr);
707 	if (err)
708 		goto err_rev;
709 
710 	/* process fw_cfg file directory entry, registering each file */
711 	err = fw_cfg_register_dir_entries();
712 	if (err)
713 		goto err_dir;
714 
715 	/* success */
716 	pr_debug("fw_cfg: loaded.\n");
717 	return 0;
718 
719 err_dir:
720 	fw_cfg_sysfs_cache_cleanup();
721 	sysfs_remove_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr);
722 err_rev:
723 	fw_cfg_io_cleanup();
724 err_probe:
725 	fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset);
726 err_name:
727 	fw_cfg_kobj_cleanup(fw_cfg_sel_ko);
728 err_sel:
729 	return err;
730 }
731 
732 static int fw_cfg_sysfs_remove(struct platform_device *pdev)
733 {
734 	pr_debug("fw_cfg: unloading.\n");
735 	fw_cfg_sysfs_cache_cleanup();
736 	sysfs_remove_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr);
737 	fw_cfg_io_cleanup();
738 	fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset);
739 	fw_cfg_kobj_cleanup(fw_cfg_sel_ko);
740 	return 0;
741 }
742 
743 static const struct of_device_id fw_cfg_sysfs_mmio_match[] = {
744 	{ .compatible = "qemu,fw-cfg-mmio", },
745 	{},
746 };
747 MODULE_DEVICE_TABLE(of, fw_cfg_sysfs_mmio_match);
748 
749 #ifdef CONFIG_ACPI
750 static const struct acpi_device_id fw_cfg_sysfs_acpi_match[] = {
751 	{ FW_CFG_ACPI_DEVICE_ID, },
752 	{},
753 };
754 MODULE_DEVICE_TABLE(acpi, fw_cfg_sysfs_acpi_match);
755 #endif
756 
757 static struct platform_driver fw_cfg_sysfs_driver = {
758 	.probe = fw_cfg_sysfs_probe,
759 	.remove = fw_cfg_sysfs_remove,
760 	.driver = {
761 		.name = "fw_cfg",
762 		.of_match_table = fw_cfg_sysfs_mmio_match,
763 		.acpi_match_table = ACPI_PTR(fw_cfg_sysfs_acpi_match),
764 	},
765 };
766 
767 #ifdef CONFIG_FW_CFG_SYSFS_CMDLINE
768 
769 static struct platform_device *fw_cfg_cmdline_dev;
770 
771 /* this probably belongs in e.g. include/linux/types.h,
772  * but right now we are the only ones doing it...
773  */
774 #ifdef CONFIG_PHYS_ADDR_T_64BIT
775 #define __PHYS_ADDR_PREFIX "ll"
776 #else
777 #define __PHYS_ADDR_PREFIX ""
778 #endif
779 
780 /* use special scanf/printf modifier for phys_addr_t, resource_size_t */
781 #define PH_ADDR_SCAN_FMT "@%" __PHYS_ADDR_PREFIX "i%n" \
782 			 ":%" __PHYS_ADDR_PREFIX "i" \
783 			 ":%" __PHYS_ADDR_PREFIX "i%n" \
784 			 ":%" __PHYS_ADDR_PREFIX "i%n"
785 
786 #define PH_ADDR_PR_1_FMT "0x%" __PHYS_ADDR_PREFIX "x@" \
787 			 "0x%" __PHYS_ADDR_PREFIX "x"
788 
789 #define PH_ADDR_PR_3_FMT PH_ADDR_PR_1_FMT \
790 			 ":%" __PHYS_ADDR_PREFIX "u" \
791 			 ":%" __PHYS_ADDR_PREFIX "u"
792 
793 #define PH_ADDR_PR_4_FMT PH_ADDR_PR_3_FMT \
794 			 ":%" __PHYS_ADDR_PREFIX "u"
795 
796 static int fw_cfg_cmdline_set(const char *arg, const struct kernel_param *kp)
797 {
798 	struct resource res[4] = {};
799 	char *str;
800 	phys_addr_t base;
801 	resource_size_t size, ctrl_off, data_off, dma_off;
802 	int processed, consumed = 0;
803 
804 	/* only one fw_cfg device can exist system-wide, so if one
805 	 * was processed on the command line already, we might as
806 	 * well stop here.
807 	 */
808 	if (fw_cfg_cmdline_dev) {
809 		/* avoid leaking previously registered device */
810 		platform_device_unregister(fw_cfg_cmdline_dev);
811 		return -EINVAL;
812 	}
813 
814 	/* consume "<size>" portion of command line argument */
815 	size = memparse(arg, &str);
816 
817 	/* get "@<base>[:<ctrl_off>:<data_off>[:<dma_off>]]" chunks */
818 	processed = sscanf(str, PH_ADDR_SCAN_FMT,
819 			   &base, &consumed,
820 			   &ctrl_off, &data_off, &consumed,
821 			   &dma_off, &consumed);
822 
823 	/* sscanf() must process precisely 1, 3 or 4 chunks:
824 	 * <base> is mandatory, optionally followed by <ctrl_off>
825 	 * and <data_off>, and <dma_off>;
826 	 * there must be no extra characters after the last chunk,
827 	 * so str[consumed] must be '\0'.
828 	 */
829 	if (str[consumed] ||
830 	    (processed != 1 && processed != 3 && processed != 4))
831 		return -EINVAL;
832 
833 	res[0].start = base;
834 	res[0].end = base + size - 1;
835 	res[0].flags = !strcmp(kp->name, "mmio") ? IORESOURCE_MEM :
836 						   IORESOURCE_IO;
837 
838 	/* insert register offsets, if provided */
839 	if (processed > 1) {
840 		res[1].name = "ctrl";
841 		res[1].start = ctrl_off;
842 		res[1].flags = IORESOURCE_REG;
843 		res[2].name = "data";
844 		res[2].start = data_off;
845 		res[2].flags = IORESOURCE_REG;
846 	}
847 	if (processed > 3) {
848 		res[3].name = "dma";
849 		res[3].start = dma_off;
850 		res[3].flags = IORESOURCE_REG;
851 	}
852 
853 	/* "processed" happens to nicely match the number of resources
854 	 * we need to pass in to this platform device.
855 	 */
856 	fw_cfg_cmdline_dev = platform_device_register_simple("fw_cfg",
857 					PLATFORM_DEVID_NONE, res, processed);
858 
859 	return PTR_ERR_OR_ZERO(fw_cfg_cmdline_dev);
860 }
861 
862 static int fw_cfg_cmdline_get(char *buf, const struct kernel_param *kp)
863 {
864 	/* stay silent if device was not configured via the command
865 	 * line, or if the parameter name (ioport/mmio) doesn't match
866 	 * the device setting
867 	 */
868 	if (!fw_cfg_cmdline_dev ||
869 	    (!strcmp(kp->name, "mmio") ^
870 	     (fw_cfg_cmdline_dev->resource[0].flags == IORESOURCE_MEM)))
871 		return 0;
872 
873 	switch (fw_cfg_cmdline_dev->num_resources) {
874 	case 1:
875 		return snprintf(buf, PAGE_SIZE, PH_ADDR_PR_1_FMT,
876 				resource_size(&fw_cfg_cmdline_dev->resource[0]),
877 				fw_cfg_cmdline_dev->resource[0].start);
878 	case 3:
879 		return snprintf(buf, PAGE_SIZE, PH_ADDR_PR_3_FMT,
880 				resource_size(&fw_cfg_cmdline_dev->resource[0]),
881 				fw_cfg_cmdline_dev->resource[0].start,
882 				fw_cfg_cmdline_dev->resource[1].start,
883 				fw_cfg_cmdline_dev->resource[2].start);
884 	case 4:
885 		return snprintf(buf, PAGE_SIZE, PH_ADDR_PR_4_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 				fw_cfg_cmdline_dev->resource[3].start);
891 	}
892 
893 	/* Should never get here */
894 	WARN(1, "Unexpected number of resources: %d\n",
895 		fw_cfg_cmdline_dev->num_resources);
896 	return 0;
897 }
898 
899 static const struct kernel_param_ops fw_cfg_cmdline_param_ops = {
900 	.set = fw_cfg_cmdline_set,
901 	.get = fw_cfg_cmdline_get,
902 };
903 
904 device_param_cb(ioport, &fw_cfg_cmdline_param_ops, NULL, S_IRUSR);
905 device_param_cb(mmio, &fw_cfg_cmdline_param_ops, NULL, S_IRUSR);
906 
907 #endif /* CONFIG_FW_CFG_SYSFS_CMDLINE */
908 
909 static int __init fw_cfg_sysfs_init(void)
910 {
911 	int ret;
912 
913 	/* create /sys/firmware/qemu_fw_cfg/ top level directory */
914 	fw_cfg_top_ko = kobject_create_and_add("qemu_fw_cfg", firmware_kobj);
915 	if (!fw_cfg_top_ko)
916 		return -ENOMEM;
917 
918 	ret = platform_driver_register(&fw_cfg_sysfs_driver);
919 	if (ret)
920 		fw_cfg_kobj_cleanup(fw_cfg_top_ko);
921 
922 	return ret;
923 }
924 
925 static void __exit fw_cfg_sysfs_exit(void)
926 {
927 	platform_driver_unregister(&fw_cfg_sysfs_driver);
928 
929 #ifdef CONFIG_FW_CFG_SYSFS_CMDLINE
930 	platform_device_unregister(fw_cfg_cmdline_dev);
931 #endif
932 
933 	/* clean up /sys/firmware/qemu_fw_cfg/ */
934 	fw_cfg_kobj_cleanup(fw_cfg_top_ko);
935 }
936 
937 module_init(fw_cfg_sysfs_init);
938 module_exit(fw_cfg_sysfs_exit);
939