1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Virtual PTP 1588 clock for use with LM-safe VMclock device.
4 *
5 * Copyright © 2024 Amazon.com, Inc. or its affiliates.
6 */
7
8 #include "linux/poll.h"
9 #include "linux/types.h"
10 #include "linux/wait.h"
11 #include <linux/acpi.h>
12 #include <linux/device.h>
13 #include <linux/err.h>
14 #include <linux/file.h>
15 #include <linux/fs.h>
16 #include <linux/init.h>
17 #include <linux/io.h>
18 #include <linux/interrupt.h>
19 #include <linux/kernel.h>
20 #include <linux/miscdevice.h>
21 #include <linux/mm.h>
22 #include <linux/module.h>
23 #include <linux/of.h>
24 #include <linux/platform_device.h>
25 #include <linux/slab.h>
26
27 #include <uapi/linux/vmclock-abi.h>
28
29 #include <linux/ptp_clock_kernel.h>
30
31 #ifdef CONFIG_X86
32 #include <asm/pvclock.h>
33 #include <asm/kvmclock.h>
34 #endif
35
36 #ifdef CONFIG_KVM_GUEST
37 #define SUPPORT_KVMCLOCK
38 #endif
39
40 static DEFINE_IDA(vmclock_ida);
41
42 ACPI_MODULE_NAME("vmclock");
43
44 struct vmclock_state {
45 struct resource res;
46 struct vmclock_abi *clk;
47 struct miscdevice miscdev;
48 wait_queue_head_t disrupt_wait;
49 struct ptp_clock_info ptp_clock_info;
50 struct ptp_clock *ptp_clock;
51 enum clocksource_ids cs_id, sys_cs_id;
52 int index;
53 char *name;
54 };
55
56 #define VMCLOCK_MAX_WAIT ms_to_ktime(100)
57
58 /* Require at least the flags field to be present. All else can be optional. */
59 #define VMCLOCK_MIN_SIZE offsetof(struct vmclock_abi, pad)
60
61 #define VMCLOCK_FIELD_PRESENT(_c, _f) \
62 (le32_to_cpu((_c)->size) >= (offsetof(struct vmclock_abi, _f) + \
63 sizeof((_c)->_f)))
64
65 /*
66 * Multiply a 64-bit count by a 64-bit tick 'period' in units of seconds >> 64
67 * and add the fractional second part of the reference time.
68 *
69 * The result is a 128-bit value, the top 64 bits of which are seconds, and
70 * the low 64 bits are (seconds >> 64).
71 */
mul_u64_u64_shr_add_u64(uint64_t * res_hi,uint64_t delta,uint64_t period,uint8_t shift,uint64_t frac_sec)72 static uint64_t mul_u64_u64_shr_add_u64(uint64_t *res_hi, uint64_t delta,
73 uint64_t period, uint8_t shift,
74 uint64_t frac_sec)
75 {
76 unsigned __int128 res = (unsigned __int128)delta * period;
77
78 res >>= shift;
79 res += frac_sec;
80 *res_hi = res >> 64;
81 return (uint64_t)res;
82 }
83
tai_adjust(struct vmclock_abi * clk,uint64_t * sec)84 static bool tai_adjust(struct vmclock_abi *clk, uint64_t *sec)
85 {
86 if (clk->time_type == VMCLOCK_TIME_TAI)
87 return true;
88
89 if (clk->time_type == VMCLOCK_TIME_UTC &&
90 (le64_to_cpu(clk->flags) & VMCLOCK_FLAG_TAI_OFFSET_VALID)) {
91 if (sec)
92 *sec -= (int16_t)le16_to_cpu(clk->tai_offset_sec);
93 return true;
94 }
95 return false;
96 }
97
vmclock_get_crosststamp(struct vmclock_state * st,struct ptp_system_timestamp * sts,struct system_counterval_t * system_counter,struct timespec64 * tspec)98 static int vmclock_get_crosststamp(struct vmclock_state *st,
99 struct ptp_system_timestamp *sts,
100 struct system_counterval_t *system_counter,
101 struct timespec64 *tspec)
102 {
103 ktime_t deadline = ktime_add(ktime_get(), VMCLOCK_MAX_WAIT);
104 uint64_t cycle, delta, seq, frac_sec;
105
106 #ifdef CONFIG_X86
107 /*
108 * We'd expect the hypervisor to know this and to report the clock
109 * status as VMCLOCK_STATUS_UNRELIABLE. But be paranoid.
110 */
111 if (check_tsc_unstable())
112 return -EINVAL;
113 #endif
114
115 while (1) {
116 seq = le32_to_cpu(st->clk->seq_count) & ~1ULL;
117
118 /*
119 * This pairs with a write barrier in the hypervisor
120 * which populates this structure.
121 */
122 virt_rmb();
123
124 if (st->clk->clock_status == VMCLOCK_STATUS_UNRELIABLE)
125 return -EINVAL;
126
127 /*
128 * When invoked for gettimex64(), fill in the pre/post system
129 * times. The simple case is when system time is based on the
130 * same counter as st->cs_id, in which case all three times
131 * will be derived from the *same* counter value.
132 *
133 * If the system isn't using the same counter, then the value
134 * from ptp_read_system_prets() will still be used as pre_ts,
135 * and ptp_read_system_postts() is called to populate postts
136 * after calling get_cycles().
137 */
138 if (sts) {
139 ptp_read_system_prets(sts);
140 if (sts->pre_sts.cs_id == st->cs_id) {
141 cycle = sts->pre_sts.cycles;
142 sts->post_sts = sts->pre_sts;
143 } else if (sts->pre_sts.hw_csid == st->cs_id &&
144 sts->pre_sts.hw_cycles) {
145 cycle = sts->pre_sts.hw_cycles;
146 sts->post_sts = sts->pre_sts;
147 } else {
148 cycle = get_cycles();
149 ptp_read_system_postts(sts);
150 }
151 } else {
152 cycle = get_cycles();
153 }
154
155 delta = cycle - le64_to_cpu(st->clk->counter_value);
156
157 frac_sec = mul_u64_u64_shr_add_u64(&tspec->tv_sec, delta,
158 le64_to_cpu(st->clk->counter_period_frac_sec),
159 st->clk->counter_period_shift,
160 le64_to_cpu(st->clk->time_frac_sec));
161 tspec->tv_nsec = mul_u64_u64_shr(frac_sec, NSEC_PER_SEC, 64);
162 tspec->tv_sec += le64_to_cpu(st->clk->time_sec);
163
164 if (!tai_adjust(st->clk, &tspec->tv_sec))
165 return -EINVAL;
166
167 /*
168 * This pairs with a write barrier in the hypervisor
169 * which populates this structure.
170 */
171 virt_rmb();
172 if (seq == le32_to_cpu(st->clk->seq_count))
173 break;
174
175 if (ktime_after(ktime_get(), deadline))
176 return -ETIMEDOUT;
177 }
178
179 if (system_counter) {
180 system_counter->cycles = cycle;
181 system_counter->cs_id = st->cs_id;
182 }
183
184 return 0;
185 }
186
187 #ifdef SUPPORT_KVMCLOCK
188 /*
189 * In the case where the system is using the KVM clock for timekeeping, convert
190 * the TSC value into a KVM clock time in order to return a paired reading that
191 * get_device_system_crosststamp() can cope with.
192 */
vmclock_get_crosststamp_kvmclock(struct vmclock_state * st,struct ptp_system_timestamp * sts,struct system_counterval_t * system_counter,struct timespec64 * tspec)193 static int vmclock_get_crosststamp_kvmclock(struct vmclock_state *st,
194 struct ptp_system_timestamp *sts,
195 struct system_counterval_t *system_counter,
196 struct timespec64 *tspec)
197 {
198 struct pvclock_vcpu_time_info *pvti = this_cpu_pvti();
199 unsigned int pvti_ver;
200 int ret;
201
202 preempt_disable_notrace();
203
204 do {
205 pvti_ver = pvclock_read_begin(pvti);
206
207 ret = vmclock_get_crosststamp(st, sts, system_counter, tspec);
208 if (ret)
209 break;
210
211 system_counter->cycles = __pvclock_read_cycles(pvti,
212 system_counter->cycles);
213 system_counter->cs_id = CSID_X86_KVM_CLK;
214
215 /*
216 * This retry should never really happen; if the TSC is
217 * stable and reliable enough across vCPUS that it is sane
218 * for the hypervisor to expose a VMCLOCK device which uses
219 * it as the reference counter, then the KVM clock sohuld be
220 * in 'master clock mode' and basically never changed. But
221 * the KVM clock is a fickle and often broken thing, so do
222 * it "properly" just in case.
223 */
224 } while (pvclock_read_retry(pvti, pvti_ver));
225
226 preempt_enable_notrace();
227
228 return ret;
229 }
230 #endif
231
ptp_vmclock_get_time_fn(ktime_t * device_time,struct system_counterval_t * system_counter,void * ctx)232 static int ptp_vmclock_get_time_fn(ktime_t *device_time,
233 struct system_counterval_t *system_counter,
234 void *ctx)
235 {
236 struct vmclock_state *st = ctx;
237 struct timespec64 tspec;
238 int ret;
239
240 #ifdef SUPPORT_KVMCLOCK
241 if (READ_ONCE(st->sys_cs_id) == CSID_X86_KVM_CLK)
242 ret = vmclock_get_crosststamp_kvmclock(st, NULL, system_counter,
243 &tspec);
244 else
245 #endif
246 ret = vmclock_get_crosststamp(st, NULL, system_counter, &tspec);
247
248 if (!ret)
249 *device_time = timespec64_to_ktime(tspec);
250
251 return ret;
252 }
253
ptp_vmclock_getcrosststamp(struct ptp_clock_info * ptp,struct system_device_crosststamp * xtstamp)254 static int ptp_vmclock_getcrosststamp(struct ptp_clock_info *ptp,
255 struct system_device_crosststamp *xtstamp)
256 {
257 struct vmclock_state *st = container_of(ptp, struct vmclock_state,
258 ptp_clock_info);
259 int ret = get_device_system_crosststamp(ptp_vmclock_get_time_fn, st,
260 NULL, xtstamp);
261 #ifdef SUPPORT_KVMCLOCK
262 /*
263 * On x86, the KVM clock may be used for the system time. We can
264 * actually convert a TSC reading to that, and return a paired
265 * timestamp that get_device_system_crosststamp() *can* handle.
266 */
267 if (ret == -ENODEV) {
268 struct system_time_snapshot systime_snapshot;
269
270 ktime_get_snapshot_id(CLOCK_REALTIME, &systime_snapshot);
271
272 if (systime_snapshot.cs_id == CSID_X86_TSC ||
273 systime_snapshot.cs_id == CSID_X86_KVM_CLK) {
274 WRITE_ONCE(st->sys_cs_id, systime_snapshot.cs_id);
275 ret = get_device_system_crosststamp(ptp_vmclock_get_time_fn,
276 st, NULL, xtstamp);
277 }
278 }
279 #endif
280 return ret;
281 }
282
283 /*
284 * PTP clock operations
285 */
286
ptp_vmclock_adjfine(struct ptp_clock_info * ptp,long delta)287 static int ptp_vmclock_adjfine(struct ptp_clock_info *ptp, long delta)
288 {
289 return -EOPNOTSUPP;
290 }
291
ptp_vmclock_adjtime(struct ptp_clock_info * ptp,s64 delta)292 static int ptp_vmclock_adjtime(struct ptp_clock_info *ptp, s64 delta)
293 {
294 return -EOPNOTSUPP;
295 }
296
ptp_vmclock_settime(struct ptp_clock_info * ptp,const struct timespec64 * ts)297 static int ptp_vmclock_settime(struct ptp_clock_info *ptp,
298 const struct timespec64 *ts)
299 {
300 return -EOPNOTSUPP;
301 }
302
ptp_vmclock_gettimex(struct ptp_clock_info * ptp,struct timespec64 * ts,struct ptp_system_timestamp * sts)303 static int ptp_vmclock_gettimex(struct ptp_clock_info *ptp, struct timespec64 *ts,
304 struct ptp_system_timestamp *sts)
305 {
306 struct vmclock_state *st = container_of(ptp, struct vmclock_state,
307 ptp_clock_info);
308
309 return vmclock_get_crosststamp(st, sts, NULL, ts);
310 }
311
ptp_vmclock_enable(struct ptp_clock_info * ptp,struct ptp_clock_request * rq,int on)312 static int ptp_vmclock_enable(struct ptp_clock_info *ptp,
313 struct ptp_clock_request *rq, int on)
314 {
315 return -EOPNOTSUPP;
316 }
317
318 static const struct ptp_clock_info ptp_vmclock_info = {
319 .owner = THIS_MODULE,
320 .max_adj = 0,
321 .n_ext_ts = 0,
322 .n_pins = 0,
323 .pps = 0,
324 .adjfine = ptp_vmclock_adjfine,
325 .adjtime = ptp_vmclock_adjtime,
326 .gettimex64 = ptp_vmclock_gettimex,
327 .settime64 = ptp_vmclock_settime,
328 .enable = ptp_vmclock_enable,
329 .getcrosststamp = ptp_vmclock_getcrosststamp,
330 };
331
vmclock_ptp_register(struct device * dev,struct vmclock_state * st)332 static struct ptp_clock *vmclock_ptp_register(struct device *dev,
333 struct vmclock_state *st)
334 {
335 enum clocksource_ids cs_id;
336
337 if (IS_ENABLED(CONFIG_ARM64) &&
338 st->clk->counter_id == VMCLOCK_COUNTER_ARM_VCNT) {
339 /* Can we check it's the virtual counter? */
340 cs_id = CSID_ARM_ARCH_COUNTER;
341 } else if (IS_ENABLED(CONFIG_X86) &&
342 st->clk->counter_id == VMCLOCK_COUNTER_X86_TSC) {
343 cs_id = CSID_X86_TSC;
344 } else {
345 return NULL;
346 }
347
348 /* Accept TAI directly, or UTC with valid offset for conversion to TAI */
349 if (!tai_adjust(st->clk, NULL)) {
350 dev_info(dev, "vmclock does not provide unambiguous time\n");
351 return NULL;
352 }
353
354 st->sys_cs_id = cs_id;
355 st->cs_id = cs_id;
356 st->ptp_clock_info = ptp_vmclock_info;
357 strscpy(st->ptp_clock_info.name, st->name);
358
359 return ptp_clock_register(&st->ptp_clock_info, dev);
360 }
361
362 struct vmclock_file_state {
363 struct vmclock_state *st;
364 atomic_t seq;
365 };
366
vmclock_miscdev_mmap(struct file * fp,struct vm_area_struct * vma)367 static int vmclock_miscdev_mmap(struct file *fp, struct vm_area_struct *vma)
368 {
369 struct vmclock_file_state *fst = fp->private_data;
370 struct vmclock_state *st = fst->st;
371
372 if ((vma->vm_flags & (VM_READ|VM_WRITE)) != VM_READ)
373 return -EROFS;
374
375 /*
376 * Restrict the read-only mapping so it cannot be upgraded to
377 * writable later with mprotect().
378 */
379 vm_flags_clear(vma, VM_MAYWRITE);
380
381 if (vma->vm_end - vma->vm_start != PAGE_SIZE || vma->vm_pgoff)
382 return -EINVAL;
383
384 if (io_remap_pfn_range(vma, vma->vm_start,
385 st->res.start >> PAGE_SHIFT, PAGE_SIZE,
386 vma->vm_page_prot))
387 return -EAGAIN;
388
389 return 0;
390 }
391
vmclock_miscdev_read(struct file * fp,char __user * buf,size_t count,loff_t * ppos)392 static ssize_t vmclock_miscdev_read(struct file *fp, char __user *buf,
393 size_t count, loff_t *ppos)
394 {
395 ktime_t deadline = ktime_add(ktime_get(), VMCLOCK_MAX_WAIT);
396 struct vmclock_file_state *fst = fp->private_data;
397 struct vmclock_state *st = fst->st;
398 uint32_t seq, old_seq;
399 size_t max_count;
400
401 if (*ppos >= PAGE_SIZE)
402 return 0;
403
404 max_count = PAGE_SIZE - *ppos;
405 if (count > max_count)
406 count = max_count;
407
408 old_seq = atomic_read(&fst->seq);
409 while (1) {
410 seq = le32_to_cpu(st->clk->seq_count) & ~1U;
411 /* Pairs with hypervisor wmb */
412 virt_rmb();
413
414 if (copy_to_user(buf, ((char *)st->clk) + *ppos, count))
415 return -EFAULT;
416
417 /* Pairs with hypervisor wmb */
418 virt_rmb();
419 if (seq == le32_to_cpu(st->clk->seq_count)) {
420 /*
421 * Either we updated fst->seq to seq (the latest version we observed)
422 * or someone else did (old_seq == seq), so we can break.
423 */
424 if (atomic_try_cmpxchg(&fst->seq, &old_seq, seq) ||
425 old_seq == seq) {
426 break;
427 }
428 }
429
430 if (ktime_after(ktime_get(), deadline))
431 return -ETIMEDOUT;
432 }
433
434 *ppos += count;
435 return count;
436 }
437
vmclock_miscdev_poll(struct file * fp,poll_table * wait)438 static __poll_t vmclock_miscdev_poll(struct file *fp, poll_table *wait)
439 {
440 struct vmclock_file_state *fst = fp->private_data;
441 struct vmclock_state *st = fst->st;
442 uint32_t seq;
443
444 /*
445 * Hypervisor will not send us any notifications, so fail immediately
446 * to avoid having caller sleeping for ever.
447 */
448 if (!(le64_to_cpu(st->clk->flags) & VMCLOCK_FLAG_NOTIFICATION_PRESENT))
449 return POLLHUP;
450
451 poll_wait(fp, &st->disrupt_wait, wait);
452
453 seq = le32_to_cpu(st->clk->seq_count);
454 if (atomic_read(&fst->seq) != seq)
455 return POLLIN | POLLRDNORM;
456
457 return 0;
458 }
459
vmclock_miscdev_open(struct inode * inode,struct file * fp)460 static int vmclock_miscdev_open(struct inode *inode, struct file *fp)
461 {
462 struct vmclock_state *st = container_of(fp->private_data,
463 struct vmclock_state, miscdev);
464 struct vmclock_file_state *fst = kzalloc_obj(*fst);
465
466 if (!fst)
467 return -ENOMEM;
468
469 fst->st = st;
470 atomic_set(&fst->seq, 0);
471
472 fp->private_data = fst;
473
474 return 0;
475 }
476
vmclock_miscdev_release(struct inode * inode,struct file * fp)477 static int vmclock_miscdev_release(struct inode *inode, struct file *fp)
478 {
479 kfree(fp->private_data);
480 return 0;
481 }
482
483 static const struct file_operations vmclock_miscdev_fops = {
484 .owner = THIS_MODULE,
485 .open = vmclock_miscdev_open,
486 .release = vmclock_miscdev_release,
487 .mmap = vmclock_miscdev_mmap,
488 .read = vmclock_miscdev_read,
489 .poll = vmclock_miscdev_poll,
490 };
491
492 /* module operations */
493
494 #if IS_ENABLED(CONFIG_ACPI)
vmclock_acpi_resources(struct acpi_resource * ares,void * data)495 static acpi_status vmclock_acpi_resources(struct acpi_resource *ares, void *data)
496 {
497 struct vmclock_state *st = data;
498 struct resource_win win;
499 struct resource *res = &win.res;
500
501 if (ares->type == ACPI_RESOURCE_TYPE_END_TAG)
502 return AE_OK;
503
504 /* There can be only one */
505 if (resource_type(&st->res) == IORESOURCE_MEM)
506 return AE_ERROR;
507
508 if (acpi_dev_resource_memory(ares, res) ||
509 acpi_dev_resource_address_space(ares, &win)) {
510
511 if (resource_type(res) != IORESOURCE_MEM ||
512 resource_size(res) < sizeof(st->clk))
513 return AE_ERROR;
514
515 st->res = *res;
516 return AE_OK;
517 }
518
519 return AE_ERROR;
520 }
521
522 static void
vmclock_acpi_notification_handler(acpi_handle __always_unused handle,u32 __always_unused event,void * dev)523 vmclock_acpi_notification_handler(acpi_handle __always_unused handle,
524 u32 __always_unused event, void *dev)
525 {
526 struct device *device = dev;
527 struct vmclock_state *st = device->driver_data;
528
529 wake_up_interruptible(&st->disrupt_wait);
530 }
531
vmclock_setup_acpi_notification(struct device * dev)532 static int vmclock_setup_acpi_notification(struct device *dev)
533 {
534 struct acpi_device *adev = ACPI_COMPANION(dev);
535 acpi_status status;
536
537 /*
538 * This should never happen as this function is only called when
539 * has_acpi_companion(dev) is true, but the logic is sufficiently
540 * complex that Coverity can't see the tautology.
541 */
542 if (!adev)
543 return -ENODEV;
544
545 status = acpi_install_notify_handler(adev->handle, ACPI_DEVICE_NOTIFY,
546 vmclock_acpi_notification_handler,
547 dev);
548 if (ACPI_FAILURE(status)) {
549 dev_err(dev, "failed to install notification handler");
550 return -ENODEV;
551 }
552
553 return 0;
554 }
555
vmclock_probe_acpi(struct device * dev,struct vmclock_state * st)556 static int vmclock_probe_acpi(struct device *dev, struct vmclock_state *st)
557 {
558 struct acpi_device *adev = ACPI_COMPANION(dev);
559 acpi_status status;
560
561 /*
562 * This should never happen as this function is only called when
563 * has_acpi_companion(dev) is true, but the logic is sufficiently
564 * complex that Coverity can't see the tautology.
565 */
566 if (!adev)
567 return -ENODEV;
568
569 status = acpi_walk_resources(adev->handle, METHOD_NAME__CRS,
570 vmclock_acpi_resources, st);
571 if (ACPI_FAILURE(status) || resource_type(&st->res) != IORESOURCE_MEM) {
572 dev_err(dev, "failed to get resources\n");
573 return -ENODEV;
574 }
575
576 return 0;
577 }
578 #endif /* CONFIG_ACPI */
579
vmclock_of_irq_handler(int __always_unused irq,void * _st)580 static irqreturn_t vmclock_of_irq_handler(int __always_unused irq, void *_st)
581 {
582 struct vmclock_state *st = _st;
583
584 wake_up_interruptible(&st->disrupt_wait);
585 return IRQ_HANDLED;
586 }
587
vmclock_probe_dt(struct device * dev,struct vmclock_state * st)588 static int vmclock_probe_dt(struct device *dev, struct vmclock_state *st)
589 {
590 struct platform_device *pdev = to_platform_device(dev);
591 struct resource *res;
592
593 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
594 if (!res)
595 return -ENODEV;
596
597 st->res = *res;
598
599 return 0;
600 }
601
vmclock_setup_of_notification(struct device * dev)602 static int vmclock_setup_of_notification(struct device *dev)
603 {
604 struct platform_device *pdev = to_platform_device(dev);
605 int irq;
606
607 irq = platform_get_irq(pdev, 0);
608 if (irq < 0)
609 return irq;
610
611 return devm_request_irq(dev, irq, vmclock_of_irq_handler, IRQF_SHARED,
612 "vmclock", dev->driver_data);
613 }
614
vmclock_setup_notification(struct device * dev,struct vmclock_state * st)615 static int vmclock_setup_notification(struct device *dev,
616 struct vmclock_state *st)
617 {
618 /* The device does not support notifications. Nothing else to do */
619 if (!(le64_to_cpu(st->clk->flags) & VMCLOCK_FLAG_NOTIFICATION_PRESENT))
620 return 0;
621
622 #if IS_ENABLED(CONFIG_ACPI)
623 if (has_acpi_companion(dev))
624 return vmclock_setup_acpi_notification(dev);
625 #endif
626 return vmclock_setup_of_notification(dev);
627 }
628
vmclock_remove(void * data)629 static void vmclock_remove(void *data)
630 {
631 struct device *dev = data;
632 struct vmclock_state *st = dev->driver_data;
633
634 if (!st) {
635 dev_err(dev, "%s called with NULL driver_data", __func__);
636 return;
637 }
638
639 #if IS_ENABLED(CONFIG_ACPI)
640 if (has_acpi_companion(dev))
641 acpi_remove_notify_handler(ACPI_COMPANION(dev)->handle,
642 ACPI_DEVICE_NOTIFY,
643 vmclock_acpi_notification_handler);
644 #endif
645
646 if (st->ptp_clock)
647 ptp_clock_unregister(st->ptp_clock);
648
649 if (st->miscdev.minor != MISC_DYNAMIC_MINOR)
650 misc_deregister(&st->miscdev);
651
652 dev->driver_data = NULL;
653 }
654
vmclock_put_idx(void * data)655 static void vmclock_put_idx(void *data)
656 {
657 struct vmclock_state *st = data;
658
659 ida_free(&vmclock_ida, st->index);
660 }
661
vmclock_probe(struct platform_device * pdev)662 static int vmclock_probe(struct platform_device *pdev)
663 {
664 struct device *dev = &pdev->dev;
665 struct vmclock_state *st;
666 int ret;
667
668 st = devm_kzalloc(dev, sizeof(*st), GFP_KERNEL);
669 if (!st)
670 return -ENOMEM;
671
672 #if IS_ENABLED(CONFIG_ACPI)
673 if (has_acpi_companion(dev))
674 ret = vmclock_probe_acpi(dev, st);
675 else
676 #endif
677 ret = vmclock_probe_dt(dev, st);
678
679 if (ret) {
680 dev_info(dev, "Failed to obtain physical address: %d\n", ret);
681 return ret;
682 }
683
684 if (resource_size(&st->res) < VMCLOCK_MIN_SIZE) {
685 dev_info(dev, "Region too small (0x%llx)\n",
686 resource_size(&st->res));
687 return -EINVAL;
688 }
689 st->clk = devm_memremap(dev, st->res.start, resource_size(&st->res),
690 MEMREMAP_WB | MEMREMAP_DEC);
691 if (IS_ERR(st->clk)) {
692 ret = PTR_ERR(st->clk);
693 dev_info(dev, "failed to map shared memory\n");
694 st->clk = NULL;
695 return ret;
696 }
697
698 if (le32_to_cpu(st->clk->magic) != VMCLOCK_MAGIC ||
699 le32_to_cpu(st->clk->size) > resource_size(&st->res) ||
700 le16_to_cpu(st->clk->version) != 1) {
701 dev_info(dev, "vmclock magic fields invalid\n");
702 return -EINVAL;
703 }
704
705 ret = ida_alloc(&vmclock_ida, GFP_KERNEL);
706 if (ret < 0)
707 return ret;
708
709 st->index = ret;
710 ret = devm_add_action_or_reset(&pdev->dev, vmclock_put_idx, st);
711 if (ret)
712 return ret;
713
714 st->name = devm_kasprintf(&pdev->dev, GFP_KERNEL, "vmclock%d", st->index);
715 if (!st->name)
716 return -ENOMEM;
717
718 st->miscdev.minor = MISC_DYNAMIC_MINOR;
719
720 init_waitqueue_head(&st->disrupt_wait);
721 dev->driver_data = st;
722
723 ret = devm_add_action_or_reset(&pdev->dev, vmclock_remove, dev);
724 if (ret)
725 return ret;
726
727 ret = vmclock_setup_notification(dev, st);
728 if (ret)
729 return ret;
730
731 /*
732 * If the structure is big enough, it can be mapped to userspace.
733 * Theoretically a guest OS even using larger pages could still
734 * use 4KiB PTEs to map smaller MMIO regions like this, but let's
735 * cross that bridge if/when we come to it.
736 */
737 if (le32_to_cpu(st->clk->size) >= PAGE_SIZE) {
738 st->miscdev.fops = &vmclock_miscdev_fops;
739 st->miscdev.name = st->name;
740
741 ret = misc_register(&st->miscdev);
742 if (ret)
743 return ret;
744 }
745
746 /* If there is valid clock information, register a PTP clock */
747 if (VMCLOCK_FIELD_PRESENT(st->clk, time_frac_sec)) {
748 /* Can return a silent NULL, or an error. */
749 st->ptp_clock = vmclock_ptp_register(dev, st);
750 if (IS_ERR(st->ptp_clock)) {
751 ret = PTR_ERR(st->ptp_clock);
752 st->ptp_clock = NULL;
753 return ret;
754 }
755 }
756
757 if (!st->miscdev.minor && !st->ptp_clock) {
758 /* Neither miscdev nor PTP registered */
759 dev_info(dev, "vmclock: Neither miscdev nor PTP available; not registering\n");
760 return -ENODEV;
761 }
762
763 dev_info(dev, "%s: registered %s%s%s\n", st->name,
764 st->miscdev.minor ? "miscdev" : "",
765 (st->miscdev.minor && st->ptp_clock) ? ", " : "",
766 st->ptp_clock ? "PTP" : "");
767
768 return 0;
769 }
770
771 static const struct acpi_device_id vmclock_acpi_ids[] = {
772 { "AMZNC10C", 0 },
773 { "VMCLOCK", 0 },
774 {}
775 };
776 MODULE_DEVICE_TABLE(acpi, vmclock_acpi_ids);
777
778 static const struct of_device_id vmclock_of_ids[] = {
779 { .compatible = "amazon,vmclock", },
780 { },
781 };
782 MODULE_DEVICE_TABLE(of, vmclock_of_ids);
783
784 static struct platform_driver vmclock_platform_driver = {
785 .probe = vmclock_probe,
786 .driver = {
787 .name = "vmclock",
788 .acpi_match_table = vmclock_acpi_ids,
789 .of_match_table = vmclock_of_ids,
790 },
791 };
792
793 module_platform_driver(vmclock_platform_driver)
794
795 MODULE_AUTHOR("David Woodhouse <dwmw2@infradead.org>");
796 MODULE_DESCRIPTION("PTP clock using VMCLOCK");
797 MODULE_LICENSE("GPL");
798