]>
Commit | Line | Data |
---|---|---|
75f3e8e4 GS |
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 | * | |
14d1824c | 13 | * [qemu_fw_cfg.]ioport=<size>@<base>[:<ctrl_off>:<data_off>[:<dma_off>]] |
75f3e8e4 | 14 | * or |
14d1824c | 15 | * [qemu_fw_cfg.]mmio=<size>@<base>[:<ctrl_off>:<data_off>[:<dma_off>]] |
75f3e8e4 GS |
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 | |
14d1824c | 22 | * <dma_off> := (optional) offset of dma register |
75f3e8e4 GS |
23 | * |
24 | * e.g.: | |
14d1824c | 25 | * qemu_fw_cfg.ioport=12@0x510:0:1:4 (the default on x86) |
75f3e8e4 | 26 | * or |
14d1824c | 27 | * qemu_fw_cfg.mmio=16@0x9020000:8:0:16 (the default on arm) |
75f3e8e4 GS |
28 | */ |
29 | ||
30 | #include <linux/module.h> | |
1b7369ac | 31 | #include <linux/mod_devicetable.h> |
75f3e8e4 GS |
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> | |
1f57bc12 | 37 | #include <uapi/linux/qemu_fw_cfg.h> |
2d6d60a3 MAL |
38 | #include <linux/delay.h> |
39 | #include <linux/crash_dump.h> | |
40 | #include <linux/crash_core.h> | |
75f3e8e4 GS |
41 | |
42 | MODULE_AUTHOR("Gabriel L. Somlo <somlo@cmu.edu>"); | |
43 | MODULE_DESCRIPTION("QEMU fw_cfg sysfs support"); | |
44 | MODULE_LICENSE("GPL"); | |
45 | ||
2d6d60a3 MAL |
46 | /* fw_cfg revision attribute, in /sys/firmware/qemu_fw_cfg top-level dir. */ |
47 | static u32 fw_cfg_rev; | |
48 | ||
75f3e8e4 GS |
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; | |
14d1824c | 56 | static void __iomem *fw_cfg_reg_dma; |
75f3e8e4 GS |
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 */ | |
8d59d5bd | 62 | static void fw_cfg_sel_endianness(u16 key) |
75f3e8e4 | 63 | { |
8d59d5bd MAL |
64 | if (fw_cfg_is_mmio) |
65 | iowrite16be(key, fw_cfg_reg_ctrl); | |
66 | else | |
67 | iowrite16(key, fw_cfg_reg_ctrl); | |
75f3e8e4 GS |
68 | } |
69 | ||
2d6d60a3 MAL |
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 | ||
75f3e8e4 | 130 | /* read chunk of given fw_cfg blob (caller responsible for sanity-check) */ |
b1cc4097 MAL |
131 | static ssize_t fw_cfg_read_blob(u16 key, |
132 | void *buf, loff_t pos, size_t count) | |
75f3e8e4 | 133 | { |
d4f6e272 | 134 | u32 glk = -1U; |
def7ac80 GS |
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); | |
b1cc4097 | 145 | return -EINVAL; |
def7ac80 GS |
146 | } |
147 | ||
75f3e8e4 | 148 | mutex_lock(&fw_cfg_dev_lock); |
8d59d5bd | 149 | fw_cfg_sel_endianness(key); |
75f3e8e4 GS |
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); | |
def7ac80 GS |
154 | |
155 | acpi_release_global_lock(glk); | |
b1cc4097 | 156 | return count; |
75f3e8e4 GS |
157 | } |
158 | ||
2d6d60a3 MAL |
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 | ||
75f3e8e4 GS |
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 */ | |
9b3ec23a | 213 | #if !(defined(FW_CFG_CTRL_OFF) && defined(FW_CFG_DATA_OFF)) |
75f3e8e4 GS |
214 | # if (defined(CONFIG_ARM) || defined(CONFIG_ARM64)) |
215 | # define FW_CFG_CTRL_OFF 0x08 | |
216 | # define FW_CFG_DATA_OFF 0x00 | |
14d1824c | 217 | # define FW_CFG_DMA_OFF 0x10 |
75f3e8e4 GS |
218 | # elif (defined(CONFIG_PPC_PMAC) || defined(CONFIG_SPARC32)) /* ppc/mac,sun4m */ |
219 | # define FW_CFG_CTRL_OFF 0x00 | |
220 | # define FW_CFG_DATA_OFF 0x02 | |
221 | # elif (defined(CONFIG_X86) || defined(CONFIG_SPARC64)) /* x86, sun4u */ | |
222 | # define FW_CFG_CTRL_OFF 0x00 | |
223 | # define FW_CFG_DATA_OFF 0x01 | |
14d1824c | 224 | # define FW_CFG_DMA_OFF 0x04 |
75f3e8e4 | 225 | # else |
00411b7b | 226 | # error "QEMU FW_CFG not available on this architecture!" |
75f3e8e4 GS |
227 | # endif |
228 | #endif | |
229 | ||
230 | /* initialize fw_cfg device i/o from platform data */ | |
231 | static int fw_cfg_do_platform_probe(struct platform_device *pdev) | |
232 | { | |
233 | char sig[FW_CFG_SIG_SIZE]; | |
14d1824c | 234 | struct resource *range, *ctrl, *data, *dma; |
75f3e8e4 GS |
235 | |
236 | /* acquire i/o range details */ | |
237 | fw_cfg_is_mmio = false; | |
238 | range = platform_get_resource(pdev, IORESOURCE_IO, 0); | |
239 | if (!range) { | |
240 | fw_cfg_is_mmio = true; | |
241 | range = platform_get_resource(pdev, IORESOURCE_MEM, 0); | |
242 | if (!range) | |
243 | return -EINVAL; | |
244 | } | |
245 | fw_cfg_p_base = range->start; | |
246 | fw_cfg_p_size = resource_size(range); | |
247 | ||
248 | if (fw_cfg_is_mmio) { | |
249 | if (!request_mem_region(fw_cfg_p_base, | |
250 | fw_cfg_p_size, "fw_cfg_mem")) | |
251 | return -EBUSY; | |
252 | fw_cfg_dev_base = ioremap(fw_cfg_p_base, fw_cfg_p_size); | |
253 | if (!fw_cfg_dev_base) { | |
254 | release_mem_region(fw_cfg_p_base, fw_cfg_p_size); | |
255 | return -EFAULT; | |
256 | } | |
257 | } else { | |
258 | if (!request_region(fw_cfg_p_base, | |
259 | fw_cfg_p_size, "fw_cfg_io")) | |
260 | return -EBUSY; | |
261 | fw_cfg_dev_base = ioport_map(fw_cfg_p_base, fw_cfg_p_size); | |
262 | if (!fw_cfg_dev_base) { | |
263 | release_region(fw_cfg_p_base, fw_cfg_p_size); | |
264 | return -EFAULT; | |
265 | } | |
266 | } | |
267 | ||
268 | /* were custom register offsets provided (e.g. on the command line)? */ | |
269 | ctrl = platform_get_resource_byname(pdev, IORESOURCE_REG, "ctrl"); | |
270 | data = platform_get_resource_byname(pdev, IORESOURCE_REG, "data"); | |
14d1824c | 271 | dma = platform_get_resource_byname(pdev, IORESOURCE_REG, "dma"); |
75f3e8e4 GS |
272 | if (ctrl && data) { |
273 | fw_cfg_reg_ctrl = fw_cfg_dev_base + ctrl->start; | |
274 | fw_cfg_reg_data = fw_cfg_dev_base + data->start; | |
275 | } else { | |
276 | /* use architecture-specific offsets */ | |
277 | fw_cfg_reg_ctrl = fw_cfg_dev_base + FW_CFG_CTRL_OFF; | |
278 | fw_cfg_reg_data = fw_cfg_dev_base + FW_CFG_DATA_OFF; | |
279 | } | |
280 | ||
14d1824c MAL |
281 | if (dma) |
282 | fw_cfg_reg_dma = fw_cfg_dev_base + dma->start; | |
283 | #ifdef FW_CFG_DMA_OFF | |
284 | else | |
285 | fw_cfg_reg_dma = fw_cfg_dev_base + FW_CFG_DMA_OFF; | |
286 | #endif | |
287 | ||
75f3e8e4 | 288 | /* verify fw_cfg device signature */ |
b1cc4097 MAL |
289 | if (fw_cfg_read_blob(FW_CFG_SIGNATURE, sig, |
290 | 0, FW_CFG_SIG_SIZE) < 0 || | |
291 | memcmp(sig, "QEMU", FW_CFG_SIG_SIZE) != 0) { | |
75f3e8e4 GS |
292 | fw_cfg_io_cleanup(); |
293 | return -ENODEV; | |
294 | } | |
295 | ||
296 | return 0; | |
297 | } | |
298 | ||
75f3e8e4 GS |
299 | static ssize_t fw_cfg_showrev(struct kobject *k, struct attribute *a, char *buf) |
300 | { | |
301 | return sprintf(buf, "%u\n", fw_cfg_rev); | |
302 | } | |
303 | ||
304 | static const struct { | |
305 | struct attribute attr; | |
306 | ssize_t (*show)(struct kobject *k, struct attribute *a, char *buf); | |
307 | } fw_cfg_rev_attr = { | |
308 | .attr = { .name = "rev", .mode = S_IRUSR }, | |
309 | .show = fw_cfg_showrev, | |
310 | }; | |
311 | ||
312 | /* fw_cfg_sysfs_entry type */ | |
313 | struct fw_cfg_sysfs_entry { | |
314 | struct kobject kobj; | |
d174ea7d MAL |
315 | u32 size; |
316 | u16 select; | |
317 | char name[FW_CFG_MAX_FILE_PATH]; | |
75f3e8e4 GS |
318 | struct list_head list; |
319 | }; | |
320 | ||
2d6d60a3 MAL |
321 | #ifdef CONFIG_CRASH_CORE |
322 | static ssize_t fw_cfg_write_vmcoreinfo(const struct fw_cfg_file *f) | |
323 | { | |
324 | static struct fw_cfg_vmcoreinfo *data; | |
325 | ssize_t ret; | |
326 | ||
327 | data = kmalloc(sizeof(struct fw_cfg_vmcoreinfo), GFP_KERNEL); | |
328 | if (!data) | |
329 | return -ENOMEM; | |
330 | ||
331 | *data = (struct fw_cfg_vmcoreinfo) { | |
332 | .guest_format = cpu_to_le16(FW_CFG_VMCOREINFO_FORMAT_ELF), | |
333 | .size = cpu_to_le32(VMCOREINFO_NOTE_SIZE), | |
334 | .paddr = cpu_to_le64(paddr_vmcoreinfo_note()) | |
335 | }; | |
336 | /* spare ourself reading host format support for now since we | |
337 | * don't know what else to format - host may ignore ours | |
338 | */ | |
339 | ret = fw_cfg_write_blob(be16_to_cpu(f->select), data, | |
340 | 0, sizeof(struct fw_cfg_vmcoreinfo)); | |
341 | ||
342 | kfree(data); | |
343 | return ret; | |
344 | } | |
345 | #endif /* CONFIG_CRASH_CORE */ | |
346 | ||
75f3e8e4 GS |
347 | /* get fw_cfg_sysfs_entry from kobject member */ |
348 | static inline struct fw_cfg_sysfs_entry *to_entry(struct kobject *kobj) | |
349 | { | |
350 | return container_of(kobj, struct fw_cfg_sysfs_entry, kobj); | |
351 | } | |
352 | ||
353 | /* fw_cfg_sysfs_attribute type */ | |
354 | struct fw_cfg_sysfs_attribute { | |
355 | struct attribute attr; | |
356 | ssize_t (*show)(struct fw_cfg_sysfs_entry *entry, char *buf); | |
357 | }; | |
358 | ||
359 | /* get fw_cfg_sysfs_attribute from attribute member */ | |
360 | static inline struct fw_cfg_sysfs_attribute *to_attr(struct attribute *attr) | |
361 | { | |
362 | return container_of(attr, struct fw_cfg_sysfs_attribute, attr); | |
363 | } | |
364 | ||
365 | /* global cache of fw_cfg_sysfs_entry objects */ | |
366 | static LIST_HEAD(fw_cfg_entry_cache); | |
367 | ||
368 | /* kobjects removed lazily by kernel, mutual exclusion needed */ | |
369 | static DEFINE_SPINLOCK(fw_cfg_cache_lock); | |
370 | ||
371 | static inline void fw_cfg_sysfs_cache_enlist(struct fw_cfg_sysfs_entry *entry) | |
372 | { | |
373 | spin_lock(&fw_cfg_cache_lock); | |
374 | list_add_tail(&entry->list, &fw_cfg_entry_cache); | |
375 | spin_unlock(&fw_cfg_cache_lock); | |
376 | } | |
377 | ||
378 | static inline void fw_cfg_sysfs_cache_delist(struct fw_cfg_sysfs_entry *entry) | |
379 | { | |
380 | spin_lock(&fw_cfg_cache_lock); | |
381 | list_del(&entry->list); | |
382 | spin_unlock(&fw_cfg_cache_lock); | |
383 | } | |
384 | ||
385 | static void fw_cfg_sysfs_cache_cleanup(void) | |
386 | { | |
387 | struct fw_cfg_sysfs_entry *entry, *next; | |
388 | ||
389 | list_for_each_entry_safe(entry, next, &fw_cfg_entry_cache, list) { | |
390 | /* will end up invoking fw_cfg_sysfs_cache_delist() | |
391 | * via each object's release() method (i.e. destructor) | |
392 | */ | |
393 | kobject_put(&entry->kobj); | |
394 | } | |
395 | } | |
396 | ||
397 | /* default_attrs: per-entry attributes and show methods */ | |
398 | ||
399 | #define FW_CFG_SYSFS_ATTR(_attr) \ | |
400 | struct fw_cfg_sysfs_attribute fw_cfg_sysfs_attr_##_attr = { \ | |
401 | .attr = { .name = __stringify(_attr), .mode = S_IRUSR }, \ | |
402 | .show = fw_cfg_sysfs_show_##_attr, \ | |
403 | } | |
404 | ||
405 | static ssize_t fw_cfg_sysfs_show_size(struct fw_cfg_sysfs_entry *e, char *buf) | |
406 | { | |
d174ea7d | 407 | return sprintf(buf, "%u\n", e->size); |
75f3e8e4 GS |
408 | } |
409 | ||
410 | static ssize_t fw_cfg_sysfs_show_key(struct fw_cfg_sysfs_entry *e, char *buf) | |
411 | { | |
d174ea7d | 412 | return sprintf(buf, "%u\n", e->select); |
75f3e8e4 GS |
413 | } |
414 | ||
415 | static ssize_t fw_cfg_sysfs_show_name(struct fw_cfg_sysfs_entry *e, char *buf) | |
416 | { | |
d174ea7d | 417 | return sprintf(buf, "%s\n", e->name); |
75f3e8e4 GS |
418 | } |
419 | ||
420 | static FW_CFG_SYSFS_ATTR(size); | |
421 | static FW_CFG_SYSFS_ATTR(key); | |
422 | static FW_CFG_SYSFS_ATTR(name); | |
423 | ||
424 | static struct attribute *fw_cfg_sysfs_entry_attrs[] = { | |
425 | &fw_cfg_sysfs_attr_size.attr, | |
426 | &fw_cfg_sysfs_attr_key.attr, | |
427 | &fw_cfg_sysfs_attr_name.attr, | |
428 | NULL, | |
429 | }; | |
430 | ||
431 | /* sysfs_ops: find fw_cfg_[entry, attribute] and call appropriate show method */ | |
432 | static ssize_t fw_cfg_sysfs_attr_show(struct kobject *kobj, struct attribute *a, | |
433 | char *buf) | |
434 | { | |
435 | struct fw_cfg_sysfs_entry *entry = to_entry(kobj); | |
436 | struct fw_cfg_sysfs_attribute *attr = to_attr(a); | |
437 | ||
438 | return attr->show(entry, buf); | |
439 | } | |
440 | ||
441 | static const struct sysfs_ops fw_cfg_sysfs_attr_ops = { | |
442 | .show = fw_cfg_sysfs_attr_show, | |
443 | }; | |
444 | ||
445 | /* release: destructor, to be called via kobject_put() */ | |
446 | static void fw_cfg_sysfs_release_entry(struct kobject *kobj) | |
447 | { | |
448 | struct fw_cfg_sysfs_entry *entry = to_entry(kobj); | |
449 | ||
450 | fw_cfg_sysfs_cache_delist(entry); | |
451 | kfree(entry); | |
452 | } | |
453 | ||
454 | /* kobj_type: ties together all properties required to register an entry */ | |
455 | static struct kobj_type fw_cfg_sysfs_entry_ktype = { | |
456 | .default_attrs = fw_cfg_sysfs_entry_attrs, | |
457 | .sysfs_ops = &fw_cfg_sysfs_attr_ops, | |
458 | .release = fw_cfg_sysfs_release_entry, | |
459 | }; | |
460 | ||
461 | /* raw-read method and attribute */ | |
462 | static ssize_t fw_cfg_sysfs_read_raw(struct file *filp, struct kobject *kobj, | |
463 | struct bin_attribute *bin_attr, | |
464 | char *buf, loff_t pos, size_t count) | |
465 | { | |
466 | struct fw_cfg_sysfs_entry *entry = to_entry(kobj); | |
467 | ||
d174ea7d | 468 | if (pos > entry->size) |
75f3e8e4 GS |
469 | return -EINVAL; |
470 | ||
d174ea7d MAL |
471 | if (count > entry->size - pos) |
472 | count = entry->size - pos; | |
75f3e8e4 | 473 | |
b1cc4097 | 474 | return fw_cfg_read_blob(entry->select, buf, pos, count); |
75f3e8e4 GS |
475 | } |
476 | ||
477 | static struct bin_attribute fw_cfg_sysfs_attr_raw = { | |
478 | .attr = { .name = "raw", .mode = S_IRUSR }, | |
479 | .read = fw_cfg_sysfs_read_raw, | |
480 | }; | |
481 | ||
246c46eb GS |
482 | /* |
483 | * Create a kset subdirectory matching each '/' delimited dirname token | |
484 | * in 'name', starting with sysfs kset/folder 'dir'; At the end, create | |
485 | * a symlink directed at the given 'target'. | |
486 | * NOTE: We do this on a best-effort basis, since 'name' is not guaranteed | |
487 | * to be a well-behaved path name. Whenever a symlink vs. kset directory | |
488 | * name collision occurs, the kernel will issue big scary warnings while | |
489 | * refusing to add the offending link or directory. We follow up with our | |
490 | * own, slightly less scary error messages explaining the situation :) | |
491 | */ | |
492 | static int fw_cfg_build_symlink(struct kset *dir, | |
493 | struct kobject *target, const char *name) | |
494 | { | |
495 | int ret; | |
496 | struct kset *subdir; | |
497 | struct kobject *ko; | |
498 | char *name_copy, *p, *tok; | |
499 | ||
500 | if (!dir || !target || !name || !*name) | |
501 | return -EINVAL; | |
502 | ||
503 | /* clone a copy of name for parsing */ | |
504 | name_copy = p = kstrdup(name, GFP_KERNEL); | |
505 | if (!name_copy) | |
506 | return -ENOMEM; | |
507 | ||
508 | /* create folders for each dirname token, then symlink for basename */ | |
509 | while ((tok = strsep(&p, "/")) && *tok) { | |
510 | ||
511 | /* last (basename) token? If so, add symlink here */ | |
512 | if (!p || !*p) { | |
513 | ret = sysfs_create_link(&dir->kobj, target, tok); | |
514 | break; | |
515 | } | |
516 | ||
517 | /* does the current dir contain an item named after tok ? */ | |
518 | ko = kset_find_obj(dir, tok); | |
519 | if (ko) { | |
520 | /* drop reference added by kset_find_obj */ | |
521 | kobject_put(ko); | |
522 | ||
523 | /* ko MUST be a kset - we're about to use it as one ! */ | |
524 | if (ko->ktype != dir->kobj.ktype) { | |
525 | ret = -EINVAL; | |
526 | break; | |
527 | } | |
528 | ||
529 | /* descend into already existing subdirectory */ | |
530 | dir = to_kset(ko); | |
531 | } else { | |
532 | /* create new subdirectory kset */ | |
533 | subdir = kzalloc(sizeof(struct kset), GFP_KERNEL); | |
534 | if (!subdir) { | |
535 | ret = -ENOMEM; | |
536 | break; | |
537 | } | |
538 | subdir->kobj.kset = dir; | |
539 | subdir->kobj.ktype = dir->kobj.ktype; | |
540 | ret = kobject_set_name(&subdir->kobj, "%s", tok); | |
541 | if (ret) { | |
542 | kfree(subdir); | |
543 | break; | |
544 | } | |
545 | ret = kset_register(subdir); | |
546 | if (ret) { | |
547 | kfree(subdir); | |
548 | break; | |
549 | } | |
550 | ||
551 | /* descend into newly created subdirectory */ | |
552 | dir = subdir; | |
553 | } | |
554 | } | |
555 | ||
556 | /* we're done with cloned copy of name */ | |
557 | kfree(name_copy); | |
558 | return ret; | |
559 | } | |
560 | ||
561 | /* recursively unregister fw_cfg/by_name/ kset directory tree */ | |
562 | static void fw_cfg_kset_unregister_recursive(struct kset *kset) | |
563 | { | |
564 | struct kobject *k, *next; | |
565 | ||
566 | list_for_each_entry_safe(k, next, &kset->list, entry) | |
567 | /* all set members are ksets too, but check just in case... */ | |
568 | if (k->ktype == kset->kobj.ktype) | |
569 | fw_cfg_kset_unregister_recursive(to_kset(k)); | |
570 | ||
571 | /* symlinks are cleanly and automatically removed with the directory */ | |
572 | kset_unregister(kset); | |
573 | } | |
574 | ||
575 | /* kobjects & kset representing top-level, by_key, and by_name folders */ | |
75f3e8e4 GS |
576 | static struct kobject *fw_cfg_top_ko; |
577 | static struct kobject *fw_cfg_sel_ko; | |
246c46eb | 578 | static struct kset *fw_cfg_fname_kset; |
75f3e8e4 GS |
579 | |
580 | /* register an individual fw_cfg file */ | |
581 | static int fw_cfg_register_file(const struct fw_cfg_file *f) | |
582 | { | |
583 | int err; | |
584 | struct fw_cfg_sysfs_entry *entry; | |
585 | ||
2d6d60a3 MAL |
586 | #ifdef CONFIG_CRASH_CORE |
587 | if (fw_cfg_dma_enabled() && | |
588 | strcmp(f->name, FW_CFG_VMCOREINFO_FILENAME) == 0 && | |
589 | !is_kdump_kernel()) { | |
590 | if (fw_cfg_write_vmcoreinfo(f) < 0) | |
591 | pr_warn("fw_cfg: failed to write vmcoreinfo"); | |
592 | } | |
593 | #endif | |
594 | ||
75f3e8e4 GS |
595 | /* allocate new entry */ |
596 | entry = kzalloc(sizeof(*entry), GFP_KERNEL); | |
597 | if (!entry) | |
598 | return -ENOMEM; | |
599 | ||
600 | /* set file entry information */ | |
d174ea7d MAL |
601 | entry->size = be32_to_cpu(f->size); |
602 | entry->select = be16_to_cpu(f->select); | |
603 | memcpy(entry->name, f->name, FW_CFG_MAX_FILE_PATH); | |
75f3e8e4 GS |
604 | |
605 | /* register entry under "/sys/firmware/qemu_fw_cfg/by_key/" */ | |
606 | err = kobject_init_and_add(&entry->kobj, &fw_cfg_sysfs_entry_ktype, | |
d174ea7d | 607 | fw_cfg_sel_ko, "%d", entry->select); |
fe3c6068 QW |
608 | if (err) { |
609 | kobject_put(&entry->kobj); | |
610 | return err; | |
611 | } | |
75f3e8e4 GS |
612 | |
613 | /* add raw binary content access */ | |
614 | err = sysfs_create_bin_file(&entry->kobj, &fw_cfg_sysfs_attr_raw); | |
615 | if (err) | |
616 | goto err_add_raw; | |
617 | ||
246c46eb | 618 | /* try adding "/sys/firmware/qemu_fw_cfg/by_name/" symlink */ |
d174ea7d | 619 | fw_cfg_build_symlink(fw_cfg_fname_kset, &entry->kobj, entry->name); |
246c46eb | 620 | |
75f3e8e4 GS |
621 | /* success, add entry to global cache */ |
622 | fw_cfg_sysfs_cache_enlist(entry); | |
623 | return 0; | |
624 | ||
625 | err_add_raw: | |
626 | kobject_del(&entry->kobj); | |
75f3e8e4 GS |
627 | kfree(entry); |
628 | return err; | |
629 | } | |
630 | ||
631 | /* iterate over all fw_cfg directory entries, registering each one */ | |
632 | static int fw_cfg_register_dir_entries(void) | |
633 | { | |
634 | int ret = 0; | |
3d47a34b | 635 | __be32 files_count; |
75f3e8e4 GS |
636 | u32 count, i; |
637 | struct fw_cfg_file *dir; | |
638 | size_t dir_size; | |
639 | ||
b1cc4097 MAL |
640 | ret = fw_cfg_read_blob(FW_CFG_FILE_DIR, &files_count, |
641 | 0, sizeof(files_count)); | |
642 | if (ret < 0) | |
643 | return ret; | |
644 | ||
3d47a34b | 645 | count = be32_to_cpu(files_count); |
75f3e8e4 GS |
646 | dir_size = count * sizeof(struct fw_cfg_file); |
647 | ||
648 | dir = kmalloc(dir_size, GFP_KERNEL); | |
649 | if (!dir) | |
650 | return -ENOMEM; | |
651 | ||
b1cc4097 MAL |
652 | ret = fw_cfg_read_blob(FW_CFG_FILE_DIR, dir, |
653 | sizeof(files_count), dir_size); | |
654 | if (ret < 0) | |
655 | goto end; | |
75f3e8e4 GS |
656 | |
657 | for (i = 0; i < count; i++) { | |
75f3e8e4 GS |
658 | ret = fw_cfg_register_file(&dir[i]); |
659 | if (ret) | |
660 | break; | |
661 | } | |
662 | ||
b1cc4097 | 663 | end: |
75f3e8e4 GS |
664 | kfree(dir); |
665 | return ret; | |
666 | } | |
667 | ||
668 | /* unregister top-level or by_key folder */ | |
669 | static inline void fw_cfg_kobj_cleanup(struct kobject *kobj) | |
670 | { | |
671 | kobject_del(kobj); | |
672 | kobject_put(kobj); | |
673 | } | |
674 | ||
675 | static int fw_cfg_sysfs_probe(struct platform_device *pdev) | |
676 | { | |
677 | int err; | |
f295c8db | 678 | __le32 rev; |
75f3e8e4 GS |
679 | |
680 | /* NOTE: If we supported multiple fw_cfg devices, we'd first create | |
681 | * a subdirectory named after e.g. pdev->id, then hang per-device | |
246c46eb | 682 | * by_key (and by_name) subdirectories underneath it. However, only |
75f3e8e4 GS |
683 | * one fw_cfg device exist system-wide, so if one was already found |
684 | * earlier, we might as well stop here. | |
685 | */ | |
686 | if (fw_cfg_sel_ko) | |
687 | return -EBUSY; | |
688 | ||
246c46eb | 689 | /* create by_key and by_name subdirs of /sys/firmware/qemu_fw_cfg/ */ |
75f3e8e4 GS |
690 | err = -ENOMEM; |
691 | fw_cfg_sel_ko = kobject_create_and_add("by_key", fw_cfg_top_ko); | |
692 | if (!fw_cfg_sel_ko) | |
693 | goto err_sel; | |
246c46eb GS |
694 | fw_cfg_fname_kset = kset_create_and_add("by_name", NULL, fw_cfg_top_ko); |
695 | if (!fw_cfg_fname_kset) | |
696 | goto err_name; | |
75f3e8e4 GS |
697 | |
698 | /* initialize fw_cfg device i/o from platform data */ | |
699 | err = fw_cfg_do_platform_probe(pdev); | |
700 | if (err) | |
701 | goto err_probe; | |
702 | ||
703 | /* get revision number, add matching top-level attribute */ | |
b1cc4097 MAL |
704 | err = fw_cfg_read_blob(FW_CFG_ID, &rev, 0, sizeof(rev)); |
705 | if (err < 0) | |
706 | goto err_probe; | |
707 | ||
f295c8db | 708 | fw_cfg_rev = le32_to_cpu(rev); |
75f3e8e4 GS |
709 | err = sysfs_create_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr); |
710 | if (err) | |
711 | goto err_rev; | |
712 | ||
713 | /* process fw_cfg file directory entry, registering each file */ | |
714 | err = fw_cfg_register_dir_entries(); | |
715 | if (err) | |
716 | goto err_dir; | |
717 | ||
718 | /* success */ | |
719 | pr_debug("fw_cfg: loaded.\n"); | |
720 | return 0; | |
721 | ||
722 | err_dir: | |
723 | fw_cfg_sysfs_cache_cleanup(); | |
724 | sysfs_remove_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr); | |
725 | err_rev: | |
726 | fw_cfg_io_cleanup(); | |
727 | err_probe: | |
246c46eb GS |
728 | fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset); |
729 | err_name: | |
75f3e8e4 GS |
730 | fw_cfg_kobj_cleanup(fw_cfg_sel_ko); |
731 | err_sel: | |
732 | return err; | |
733 | } | |
734 | ||
735 | static int fw_cfg_sysfs_remove(struct platform_device *pdev) | |
736 | { | |
737 | pr_debug("fw_cfg: unloading.\n"); | |
738 | fw_cfg_sysfs_cache_cleanup(); | |
23f1b8d9 MAL |
739 | sysfs_remove_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr); |
740 | fw_cfg_io_cleanup(); | |
246c46eb | 741 | fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset); |
75f3e8e4 | 742 | fw_cfg_kobj_cleanup(fw_cfg_sel_ko); |
75f3e8e4 GS |
743 | return 0; |
744 | } | |
745 | ||
746 | static const struct of_device_id fw_cfg_sysfs_mmio_match[] = { | |
747 | { .compatible = "qemu,fw-cfg-mmio", }, | |
748 | {}, | |
749 | }; | |
750 | MODULE_DEVICE_TABLE(of, fw_cfg_sysfs_mmio_match); | |
751 | ||
752 | #ifdef CONFIG_ACPI | |
753 | static const struct acpi_device_id fw_cfg_sysfs_acpi_match[] = { | |
1f57bc12 | 754 | { FW_CFG_ACPI_DEVICE_ID, }, |
75f3e8e4 GS |
755 | {}, |
756 | }; | |
757 | MODULE_DEVICE_TABLE(acpi, fw_cfg_sysfs_acpi_match); | |
758 | #endif | |
759 | ||
760 | static struct platform_driver fw_cfg_sysfs_driver = { | |
761 | .probe = fw_cfg_sysfs_probe, | |
762 | .remove = fw_cfg_sysfs_remove, | |
763 | .driver = { | |
764 | .name = "fw_cfg", | |
765 | .of_match_table = fw_cfg_sysfs_mmio_match, | |
766 | .acpi_match_table = ACPI_PTR(fw_cfg_sysfs_acpi_match), | |
767 | }, | |
768 | }; | |
769 | ||
770 | #ifdef CONFIG_FW_CFG_SYSFS_CMDLINE | |
771 | ||
772 | static struct platform_device *fw_cfg_cmdline_dev; | |
773 | ||
774 | /* this probably belongs in e.g. include/linux/types.h, | |
775 | * but right now we are the only ones doing it... | |
776 | */ | |
777 | #ifdef CONFIG_PHYS_ADDR_T_64BIT | |
778 | #define __PHYS_ADDR_PREFIX "ll" | |
779 | #else | |
780 | #define __PHYS_ADDR_PREFIX "" | |
781 | #endif | |
782 | ||
783 | /* use special scanf/printf modifier for phys_addr_t, resource_size_t */ | |
784 | #define PH_ADDR_SCAN_FMT "@%" __PHYS_ADDR_PREFIX "i%n" \ | |
785 | ":%" __PHYS_ADDR_PREFIX "i" \ | |
14d1824c | 786 | ":%" __PHYS_ADDR_PREFIX "i%n" \ |
75f3e8e4 GS |
787 | ":%" __PHYS_ADDR_PREFIX "i%n" |
788 | ||
789 | #define PH_ADDR_PR_1_FMT "0x%" __PHYS_ADDR_PREFIX "x@" \ | |
790 | "0x%" __PHYS_ADDR_PREFIX "x" | |
791 | ||
792 | #define PH_ADDR_PR_3_FMT PH_ADDR_PR_1_FMT \ | |
793 | ":%" __PHYS_ADDR_PREFIX "u" \ | |
794 | ":%" __PHYS_ADDR_PREFIX "u" | |
795 | ||
14d1824c MAL |
796 | #define PH_ADDR_PR_4_FMT PH_ADDR_PR_3_FMT \ |
797 | ":%" __PHYS_ADDR_PREFIX "u" | |
798 | ||
75f3e8e4 GS |
799 | static int fw_cfg_cmdline_set(const char *arg, const struct kernel_param *kp) |
800 | { | |
14d1824c | 801 | struct resource res[4] = {}; |
75f3e8e4 GS |
802 | char *str; |
803 | phys_addr_t base; | |
14d1824c | 804 | resource_size_t size, ctrl_off, data_off, dma_off; |
75f3e8e4 GS |
805 | int processed, consumed = 0; |
806 | ||
807 | /* only one fw_cfg device can exist system-wide, so if one | |
808 | * was processed on the command line already, we might as | |
809 | * well stop here. | |
810 | */ | |
811 | if (fw_cfg_cmdline_dev) { | |
812 | /* avoid leaking previously registered device */ | |
813 | platform_device_unregister(fw_cfg_cmdline_dev); | |
814 | return -EINVAL; | |
815 | } | |
816 | ||
817 | /* consume "<size>" portion of command line argument */ | |
818 | size = memparse(arg, &str); | |
819 | ||
14d1824c | 820 | /* get "@<base>[:<ctrl_off>:<data_off>[:<dma_off>]]" chunks */ |
75f3e8e4 GS |
821 | processed = sscanf(str, PH_ADDR_SCAN_FMT, |
822 | &base, &consumed, | |
14d1824c MAL |
823 | &ctrl_off, &data_off, &consumed, |
824 | &dma_off, &consumed); | |
75f3e8e4 | 825 | |
14d1824c | 826 | /* sscanf() must process precisely 1, 3 or 4 chunks: |
75f3e8e4 | 827 | * <base> is mandatory, optionally followed by <ctrl_off> |
14d1824c | 828 | * and <data_off>, and <dma_off>; |
75f3e8e4 GS |
829 | * there must be no extra characters after the last chunk, |
830 | * so str[consumed] must be '\0'. | |
831 | */ | |
832 | if (str[consumed] || | |
14d1824c | 833 | (processed != 1 && processed != 3 && processed != 4)) |
75f3e8e4 GS |
834 | return -EINVAL; |
835 | ||
836 | res[0].start = base; | |
837 | res[0].end = base + size - 1; | |
838 | res[0].flags = !strcmp(kp->name, "mmio") ? IORESOURCE_MEM : | |
839 | IORESOURCE_IO; | |
840 | ||
841 | /* insert register offsets, if provided */ | |
842 | if (processed > 1) { | |
843 | res[1].name = "ctrl"; | |
844 | res[1].start = ctrl_off; | |
845 | res[1].flags = IORESOURCE_REG; | |
846 | res[2].name = "data"; | |
847 | res[2].start = data_off; | |
848 | res[2].flags = IORESOURCE_REG; | |
849 | } | |
14d1824c MAL |
850 | if (processed > 3) { |
851 | res[3].name = "dma"; | |
852 | res[3].start = dma_off; | |
853 | res[3].flags = IORESOURCE_REG; | |
854 | } | |
75f3e8e4 GS |
855 | |
856 | /* "processed" happens to nicely match the number of resources | |
857 | * we need to pass in to this platform device. | |
858 | */ | |
859 | fw_cfg_cmdline_dev = platform_device_register_simple("fw_cfg", | |
860 | PLATFORM_DEVID_NONE, res, processed); | |
75f3e8e4 | 861 | |
0a9e63aa | 862 | return PTR_ERR_OR_ZERO(fw_cfg_cmdline_dev); |
75f3e8e4 GS |
863 | } |
864 | ||
865 | static int fw_cfg_cmdline_get(char *buf, const struct kernel_param *kp) | |
866 | { | |
867 | /* stay silent if device was not configured via the command | |
868 | * line, or if the parameter name (ioport/mmio) doesn't match | |
869 | * the device setting | |
870 | */ | |
871 | if (!fw_cfg_cmdline_dev || | |
872 | (!strcmp(kp->name, "mmio") ^ | |
873 | (fw_cfg_cmdline_dev->resource[0].flags == IORESOURCE_MEM))) | |
874 | return 0; | |
875 | ||
876 | switch (fw_cfg_cmdline_dev->num_resources) { | |
877 | case 1: | |
878 | return snprintf(buf, PAGE_SIZE, PH_ADDR_PR_1_FMT, | |
879 | resource_size(&fw_cfg_cmdline_dev->resource[0]), | |
880 | fw_cfg_cmdline_dev->resource[0].start); | |
881 | case 3: | |
882 | return snprintf(buf, PAGE_SIZE, PH_ADDR_PR_3_FMT, | |
883 | resource_size(&fw_cfg_cmdline_dev->resource[0]), | |
884 | fw_cfg_cmdline_dev->resource[0].start, | |
885 | fw_cfg_cmdline_dev->resource[1].start, | |
886 | fw_cfg_cmdline_dev->resource[2].start); | |
14d1824c MAL |
887 | case 4: |
888 | return snprintf(buf, PAGE_SIZE, PH_ADDR_PR_4_FMT, | |
889 | resource_size(&fw_cfg_cmdline_dev->resource[0]), | |
890 | fw_cfg_cmdline_dev->resource[0].start, | |
891 | fw_cfg_cmdline_dev->resource[1].start, | |
892 | fw_cfg_cmdline_dev->resource[2].start, | |
893 | fw_cfg_cmdline_dev->resource[3].start); | |
75f3e8e4 GS |
894 | } |
895 | ||
896 | /* Should never get here */ | |
897 | WARN(1, "Unexpected number of resources: %d\n", | |
898 | fw_cfg_cmdline_dev->num_resources); | |
899 | return 0; | |
900 | } | |
901 | ||
902 | static const struct kernel_param_ops fw_cfg_cmdline_param_ops = { | |
903 | .set = fw_cfg_cmdline_set, | |
904 | .get = fw_cfg_cmdline_get, | |
905 | }; | |
906 | ||
907 | device_param_cb(ioport, &fw_cfg_cmdline_param_ops, NULL, S_IRUSR); | |
908 | device_param_cb(mmio, &fw_cfg_cmdline_param_ops, NULL, S_IRUSR); | |
909 | ||
910 | #endif /* CONFIG_FW_CFG_SYSFS_CMDLINE */ | |
911 | ||
912 | static int __init fw_cfg_sysfs_init(void) | |
913 | { | |
e8aabc64 MT |
914 | int ret; |
915 | ||
75f3e8e4 GS |
916 | /* create /sys/firmware/qemu_fw_cfg/ top level directory */ |
917 | fw_cfg_top_ko = kobject_create_and_add("qemu_fw_cfg", firmware_kobj); | |
918 | if (!fw_cfg_top_ko) | |
919 | return -ENOMEM; | |
920 | ||
e8aabc64 MT |
921 | ret = platform_driver_register(&fw_cfg_sysfs_driver); |
922 | if (ret) | |
923 | fw_cfg_kobj_cleanup(fw_cfg_top_ko); | |
924 | ||
925 | return ret; | |
75f3e8e4 GS |
926 | } |
927 | ||
928 | static void __exit fw_cfg_sysfs_exit(void) | |
929 | { | |
930 | platform_driver_unregister(&fw_cfg_sysfs_driver); | |
931 | ||
932 | #ifdef CONFIG_FW_CFG_SYSFS_CMDLINE | |
933 | platform_device_unregister(fw_cfg_cmdline_dev); | |
934 | #endif | |
935 | ||
936 | /* clean up /sys/firmware/qemu_fw_cfg/ */ | |
937 | fw_cfg_kobj_cleanup(fw_cfg_top_ko); | |
938 | } | |
939 | ||
940 | module_init(fw_cfg_sysfs_init); | |
941 | module_exit(fw_cfg_sysfs_exit); |