1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Fast user context implementation of clock_gettime, gettimeofday, and time.
4 *
5 * Copyright (C) 2019 ARM Limited.
6 * Copyright 2006 Andi Kleen, SUSE Labs.
7 * 32 Bit compat layer by Stefani Seibold <stefani@seibold.net>
8 *  sponsored by Rohde & Schwarz GmbH & Co. KG Munich/Germany
9 */
10#ifndef __ASM_VDSO_GETTIMEOFDAY_H
11#define __ASM_VDSO_GETTIMEOFDAY_H
12
13#ifndef __ASSEMBLY__
14
15#include <uapi/linux/time.h>
16#include <asm/vgtod.h>
17#include <asm/vvar.h>
18#include <asm/unistd.h>
19#include <asm/msr.h>
20#include <asm/pvclock.h>
21#include <clocksource/hyperv_timer.h>
22
23#define __vdso_data (VVAR(_vdso_data))
24#define __timens_vdso_data (TIMENS(_vdso_data))
25
26#define VDSO_HAS_TIME 1
27
28#define VDSO_HAS_CLOCK_GETRES 1
29
30/*
31 * Declare the memory-mapped vclock data pages.  These come from hypervisors.
32 * If we ever reintroduce something like direct access to an MMIO clock like
33 * the HPET again, it will go here as well.
34 *
35 * A load from any of these pages will segfault if the clock in question is
36 * disabled, so appropriate compiler barriers and checks need to be used
37 * to prevent stray loads.
38 *
39 * These declarations MUST NOT be const.  The compiler will assume that
40 * an extern const variable has genuinely constant contents, and the
41 * resulting code won't work, since the whole point is that these pages
42 * change over time, possibly while we're accessing them.
43 */
44
45#ifdef CONFIG_PARAVIRT_CLOCK
46/*
47 * This is the vCPU 0 pvclock page.  We only use pvclock from the vDSO
48 * if the hypervisor tells us that all vCPUs can get valid data from the
49 * vCPU 0 page.
50 */
51extern struct pvclock_vsyscall_time_info pvclock_page
52	__attribute__((visibility("hidden")));
53#endif
54
55#ifdef CONFIG_HYPERV_TIMER
56extern struct ms_hyperv_tsc_page hvclock_page
57	__attribute__((visibility("hidden")));
58#endif
59
60#ifdef CONFIG_TIME_NS
61static __always_inline const struct vdso_data *__arch_get_timens_vdso_data(void)
62{
63	return __timens_vdso_data;
64}
65#endif
66
67#ifndef BUILD_VDSO32
68
69static __always_inline
70long clock_gettime_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)
71{
72	long ret;
73
74	asm ("syscall" : "=a" (ret), "=m" (*_ts) :
75	     "0" (__NR_clock_gettime), "D" (_clkid), "S" (_ts) :
76	     "rcx", "r11");
77
78	return ret;
79}
80
81static __always_inline
82long gettimeofday_fallback(struct __kernel_old_timeval *_tv,
83			   struct timezone *_tz)
84{
85	long ret;
86
87	asm("syscall" : "=a" (ret) :
88	    "0" (__NR_gettimeofday), "D" (_tv), "S" (_tz) : "memory");
89
90	return ret;
91}
92
93static __always_inline
94long clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)
95{
96	long ret;
97
98	asm ("syscall" : "=a" (ret), "=m" (*_ts) :
99	     "0" (__NR_clock_getres), "D" (_clkid), "S" (_ts) :
100	     "rcx", "r11");
101
102	return ret;
103}
104
105#else
106
107static __always_inline
108long clock_gettime_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)
109{
110	long ret;
111
112	asm (
113		"mov %%ebx, %%edx \n"
114		"mov %[clock], %%ebx \n"
115		"call __kernel_vsyscall \n"
116		"mov %%edx, %%ebx \n"
117		: "=a" (ret), "=m" (*_ts)
118		: "0" (__NR_clock_gettime64), [clock] "g" (_clkid), "c" (_ts)
119		: "edx");
120
121	return ret;
122}
123
124static __always_inline
125long clock_gettime32_fallback(clockid_t _clkid, struct old_timespec32 *_ts)
126{
127	long ret;
128
129	asm (
130		"mov %%ebx, %%edx \n"
131		"mov %[clock], %%ebx \n"
132		"call __kernel_vsyscall \n"
133		"mov %%edx, %%ebx \n"
134		: "=a" (ret), "=m" (*_ts)
135		: "0" (__NR_clock_gettime), [clock] "g" (_clkid), "c" (_ts)
136		: "edx");
137
138	return ret;
139}
140
141static __always_inline
142long gettimeofday_fallback(struct __kernel_old_timeval *_tv,
143			   struct timezone *_tz)
144{
145	long ret;
146
147	asm(
148		"mov %%ebx, %%edx \n"
149		"mov %2, %%ebx \n"
150		"call __kernel_vsyscall \n"
151		"mov %%edx, %%ebx \n"
152		: "=a" (ret)
153		: "0" (__NR_gettimeofday), "g" (_tv), "c" (_tz)
154		: "memory", "edx");
155
156	return ret;
157}
158
159static __always_inline long
160clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)
161{
162	long ret;
163
164	asm (
165		"mov %%ebx, %%edx \n"
166		"mov %[clock], %%ebx \n"
167		"call __kernel_vsyscall \n"
168		"mov %%edx, %%ebx \n"
169		: "=a" (ret), "=m" (*_ts)
170		: "0" (__NR_clock_getres_time64), [clock] "g" (_clkid), "c" (_ts)
171		: "edx");
172
173	return ret;
174}
175
176static __always_inline
177long clock_getres32_fallback(clockid_t _clkid, struct old_timespec32 *_ts)
178{
179	long ret;
180
181	asm (
182		"mov %%ebx, %%edx \n"
183		"mov %[clock], %%ebx \n"
184		"call __kernel_vsyscall \n"
185		"mov %%edx, %%ebx \n"
186		: "=a" (ret), "=m" (*_ts)
187		: "0" (__NR_clock_getres), [clock] "g" (_clkid), "c" (_ts)
188		: "edx");
189
190	return ret;
191}
192
193#endif
194
195#ifdef CONFIG_PARAVIRT_CLOCK
196static u64 vread_pvclock(void)
197{
198	const struct pvclock_vcpu_time_info *pvti = &pvclock_page.pvti;
199	u32 version;
200	u64 ret;
201
202	/*
203	 * Note: The kernel and hypervisor must guarantee that cpu ID
204	 * number maps 1:1 to per-CPU pvclock time info.
205	 *
206	 * Because the hypervisor is entirely unaware of guest userspace
207	 * preemption, it cannot guarantee that per-CPU pvclock time
208	 * info is updated if the underlying CPU changes or that that
209	 * version is increased whenever underlying CPU changes.
210	 *
211	 * On KVM, we are guaranteed that pvti updates for any vCPU are
212	 * atomic as seen by *all* vCPUs.  This is an even stronger
213	 * guarantee than we get with a normal seqlock.
214	 *
215	 * On Xen, we don't appear to have that guarantee, but Xen still
216	 * supplies a valid seqlock using the version field.
217	 *
218	 * We only do pvclock vdso timing at all if
219	 * PVCLOCK_TSC_STABLE_BIT is set, and we interpret that bit to
220	 * mean that all vCPUs have matching pvti and that the TSC is
221	 * synced, so we can just look at vCPU 0's pvti.
222	 */
223
224	do {
225		version = pvclock_read_begin(pvti);
226
227		if (unlikely(!(pvti->flags & PVCLOCK_TSC_STABLE_BIT)))
228			return U64_MAX;
229
230		ret = __pvclock_read_cycles(pvti, rdtsc_ordered());
231	} while (pvclock_read_retry(pvti, version));
232
233	return ret;
234}
235#endif
236
237#ifdef CONFIG_HYPERV_TIMER
238static u64 vread_hvclock(void)
239{
240	return hv_read_tsc_page(&hvclock_page);
241}
242#endif
243
244static inline u64 __arch_get_hw_counter(s32 clock_mode,
245					const struct vdso_data *vd)
246{
247	if (likely(clock_mode == VDSO_CLOCKMODE_TSC))
248		return (u64)rdtsc_ordered();
249	/*
250	 * For any memory-mapped vclock type, we need to make sure that gcc
251	 * doesn't cleverly hoist a load before the mode check.  Otherwise we
252	 * might end up touching the memory-mapped page even if the vclock in
253	 * question isn't enabled, which will segfault.  Hence the barriers.
254	 */
255#ifdef CONFIG_PARAVIRT_CLOCK
256	if (clock_mode == VDSO_CLOCKMODE_PVCLOCK) {
257		barrier();
258		return vread_pvclock();
259	}
260#endif
261#ifdef CONFIG_HYPERV_TIMER
262	if (clock_mode == VDSO_CLOCKMODE_HVCLOCK) {
263		barrier();
264		return vread_hvclock();
265	}
266#endif
267	return U64_MAX;
268}
269
270static __always_inline const struct vdso_data *__arch_get_vdso_data(void)
271{
272	return __vdso_data;
273}
274
275static inline bool arch_vdso_clocksource_ok(const struct vdso_data *vd)
276{
277	return true;
278}
279#define vdso_clocksource_ok arch_vdso_clocksource_ok
280
281/*
282 * Clocksource read value validation to handle PV and HyperV clocksources
283 * which can be invalidated asynchronously and indicate invalidation by
284 * returning U64_MAX, which can be effectively tested by checking for a
285 * negative value after casting it to s64.
286 */
287static inline bool arch_vdso_cycles_ok(u64 cycles)
288{
289	return (s64)cycles >= 0;
290}
291#define vdso_cycles_ok arch_vdso_cycles_ok
292
293/*
294 * x86 specific delta calculation.
295 *
296 * The regular implementation assumes that clocksource reads are globally
297 * monotonic. The TSC can be slightly off across sockets which can cause
298 * the regular delta calculation (@cycles - @last) to return a huge time
299 * jump.
300 *
301 * Therefore it needs to be verified that @cycles are greater than
302 * @last. If not then use @last, which is the base time of the current
303 * conversion period.
304 *
305 * This variant also removes the masking of the subtraction because the
306 * clocksource mask of all VDSO capable clocksources on x86 is U64_MAX
307 * which would result in a pointless operation. The compiler cannot
308 * optimize it away as the mask comes from the vdso data and is not compile
309 * time constant.
310 */
311static __always_inline
312u64 vdso_calc_delta(u64 cycles, u64 last, u64 mask, u32 mult)
313{
314	if (cycles > last)
315		return (cycles - last) * mult;
316	return 0;
317}
318#define vdso_calc_delta vdso_calc_delta
319
320#endif /* !__ASSEMBLY__ */
321
322#endif /* __ASM_VDSO_GETTIMEOFDAY_H */
323