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
42MODULE_AUTHOR("Gabriel L. Somlo <somlo@cmu.edu>");
43MODULE_DESCRIPTION("QEMU fw_cfg sysfs support");
44MODULE_LICENSE("GPL");
45
46/* fw_cfg revision attribute, in /sys/firmware/qemu_fw_cfg top-level dir. */
47static u32 fw_cfg_rev;
48
49/* fw_cfg device i/o register addresses */
50static bool fw_cfg_is_mmio;
51static phys_addr_t fw_cfg_p_base;
52static resource_size_t fw_cfg_p_size;
53static void __iomem *fw_cfg_dev_base;
54static void __iomem *fw_cfg_reg_ctrl;
55static void __iomem *fw_cfg_reg_data;
56static void __iomem *fw_cfg_reg_dma;
57
58/* atomic access to fw_cfg device (potentially slow i/o, so using mutex) */
59static DEFINE_MUTEX(fw_cfg_dev_lock);
60
61/* pick appropriate endianness for selector key */
62static 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
71static 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 */
77static 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
91static 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
123end:
124	kfree(d);
125
126	return ret;
127}
128#endif
129
130/* read chunk of given fw_cfg blob (caller responsible for sanity-check) */
131static 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) */
161static 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
191end:
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 */
201static 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 */
234static 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
302static 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
308static 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 */
314struct 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
323static 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 */
349static 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 */
355struct 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 */
361static 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 */
367static LIST_HEAD(fw_cfg_entry_cache);
368
369/* kobjects removed lazily by kernel, mutual exclusion needed */
370static DEFINE_SPINLOCK(fw_cfg_cache_lock);
371
372static 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
379static 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
386static 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) \
399struct 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
404static 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
409static 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
414static 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
419static FW_CFG_SYSFS_ATTR(size);
420static FW_CFG_SYSFS_ATTR(key);
421static FW_CFG_SYSFS_ATTR(name);
422
423static 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 */
431static 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
440static 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() */
445static 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 */
453static 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 */
460static 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
475static 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 */
490static 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 */
560static 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 */
574static struct kobject *fw_cfg_top_ko;
575static struct kobject *fw_cfg_sel_ko;
576static struct kset *fw_cfg_fname_kset;
577
578/* register an individual fw_cfg file */
579static 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
621err_del_entry:
622	kobject_del(&entry->kobj);
623err_put_entry:
624	kobject_put(&entry->kobj);
625	return err;
626}
627
628/* iterate over all fw_cfg directory entries, registering each one */
629static 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
660end:
661	kfree(dir);
662	return ret;
663}
664
665/* unregister top-level or by_key folder */
666static inline void fw_cfg_kobj_cleanup(struct kobject *kobj)
667{
668	kobject_del(kobj);
669	kobject_put(kobj);
670}
671
672static 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
719err_dir:
720	fw_cfg_sysfs_cache_cleanup();
721	sysfs_remove_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr);
722err_rev:
723	fw_cfg_io_cleanup();
724err_probe:
725	fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset);
726err_name:
727	fw_cfg_kobj_cleanup(fw_cfg_sel_ko);
728err_sel:
729	return err;
730}
731
732static 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
743static const struct of_device_id fw_cfg_sysfs_mmio_match[] = {
744	{ .compatible = "qemu,fw-cfg-mmio", },
745	{},
746};
747MODULE_DEVICE_TABLE(of, fw_cfg_sysfs_mmio_match);
748
749#ifdef CONFIG_ACPI
750static const struct acpi_device_id fw_cfg_sysfs_acpi_match[] = {
751	{ FW_CFG_ACPI_DEVICE_ID, },
752	{},
753};
754MODULE_DEVICE_TABLE(acpi, fw_cfg_sysfs_acpi_match);
755#endif
756
757static 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
769static 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
796static 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
862static 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
899static const struct kernel_param_ops fw_cfg_cmdline_param_ops = {
900	.set = fw_cfg_cmdline_set,
901	.get = fw_cfg_cmdline_get,
902};
903
904device_param_cb(ioport, &fw_cfg_cmdline_param_ops, NULL, S_IRUSR);
905device_param_cb(mmio, &fw_cfg_cmdline_param_ops, NULL, S_IRUSR);
906
907#endif /* CONFIG_FW_CFG_SYSFS_CMDLINE */
908
909static 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
925static 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
937module_init(fw_cfg_sysfs_init);
938module_exit(fw_cfg_sysfs_exit);
939