vdso/vsyscall: Avoid slow division loop in auxiliary clock update

The call to __iter_div_u64_rem() in vdso_time_update_aux() is a wrapper
around subtraction. It cannot be used to divide large numbers, as that
introduces long, computationally expensive delays.  A regular u64 division
is also not possible in the timekeeper update path as it can be too slow.

Instead of splitting the ktime_t offset into into second and subsecond
components during the timekeeper update fast-path, do it together with the
adjustment of tk->offs_aux in the slow-path. Equivalent to the handling of
offs_boot and monotonic_to_boot.

Reuse the storage of monotonic_to_boot for the new field, as it is not used
by auxiliary timekeepers.

Fixes: 380b84e168e5 ("vdso/vsyscall: Update auxiliary clock data in the datapage")
Reported-by: Miroslav Lichvar <mlichvar@redhat.com>
Signed-off-by: Thomas Weißschuh <thomas.weissschuh@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20250825-vdso-auxclock-division-v1-1-a1d32a16a313@linutronix.de
Closes: https://lore.kernel.org/lkml/aKwsNNWsHJg8IKzj@localhost/

diff --git a/include/linux/timekeeper_internal.h b/include/linux/timekeeper_internal.h
index c27aac67cb3f15f0892c667a990d4a6e16a06d76..b8ae89ea28abfca96f424b6a2250de7ebfe6364e 100644 (file)
--- a/include/linux/timekeeper_internal.h
+++ b/include/linux/timekeeper_internal.h
@@ -76,6 +76,7 @@ struct tk_read_base {
  * @cs_was_changed_seq:                The sequence number of clocksource change events
  * @clock_valid:               Indicator for valid clock
  * @monotonic_to_boot:         CLOCK_MONOTONIC to CLOCK_BOOTTIME offset
+ * @monotonic_to_aux:          CLOCK_MONOTONIC to CLOCK_AUX offset
  * @cycle_interval:            Number of clock cycles in one NTP interval
  * @xtime_interval:            Number of clock shifted nano seconds in one NTP
  *                             interval.
@@ -117,6 +118,9 @@ struct tk_read_base {
  * @offs_aux is used by the auxiliary timekeepers which do not utilize any
  * of the regular timekeeper offset fields.
  *
+ * @monotonic_to_aux is a timespec64 representation of @offs_aux to
+ * accelerate the VDSO update for CLOCK_AUX.
+ *
  * The cacheline ordering of the structure is optimized for in kernel usage of
  * the ktime_get() and ktime_get_ts64() family of time accessors. Struct
  * timekeeper is prepended in the core timekeeping code with a sequence count,
@@ -159,7 +163,10 @@ struct timekeeper {
        u8                      cs_was_changed_seq;
        u8                      clock_valid;
 
-       struct timespec64       monotonic_to_boot;
+       union {
+               struct timespec64       monotonic_to_boot;
+               struct timespec64       monotonic_to_aux;
+       };
 
        u64                     cycle_interval;
        u64                     xtime_interval;

diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index 059fa8b79be60dee30a788b3e7c43a72dbb465cd..b6974fce800cd8b8a9d66b3663f3c91fb338a751 100644 (file)
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -83,6 +83,12 @@ static inline bool tk_is_aux(const struct timekeeper *tk)
 }
 #endif
 
+static inline void tk_update_aux_offs(struct timekeeper *tk, ktime_t offs)
+{
+       tk->offs_aux = offs;
+       tk->monotonic_to_aux = ktime_to_timespec64(offs);
+}
+
 /* flag for if timekeeping is suspended */
 int __read_mostly timekeeping_suspended;
 
@@ -1506,7 +1512,7 @@ static int __timekeeping_inject_offset(struct tk_data *tkd, const struct timespe
                        timekeeping_restore_shadow(tkd);
                        return -EINVAL;
                }
-               tks->offs_aux = offs;
+               tk_update_aux_offs(tks, offs);
        }
 
        timekeeping_update_from_shadow(tkd, TK_UPDATE_ALL);
@@ -2937,7 +2943,7 @@ static int aux_clock_set(const clockid_t id, const struct timespec64 *tnew)
         * xtime ("realtime") is not applicable for auxiliary clocks and
         * kept in sync with "monotonic".
         */
-       aux_tks->offs_aux = ktime_sub(timespec64_to_ktime(*tnew), tnow);
+       tk_update_aux_offs(aux_tks, ktime_sub(timespec64_to_ktime(*tnew), tnow));
 
        timekeeping_update_from_shadow(aux_tkd, TK_UPDATE_ALL);
        return 0;

diff --git a/kernel/time/vsyscall.c b/kernel/time/vsyscall.c
index 8ba8b0d8a3873123da1ed0b1c6dbe7ef172d49f2..aa59919b8f2c23e9a502f676a8359b4d533f270c 100644 (file)
--- a/kernel/time/vsyscall.c
+++ b/kernel/time/vsyscall.c
@@ -159,10 +159,10 @@ void vdso_time_update_aux(struct timekeeper *tk)
        if (clock_mode != VDSO_CLOCKMODE_NONE) {
                fill_clock_configuration(vc, &tk->tkr_mono);
 
-               vdso_ts->sec    = tk->xtime_sec;
+               vdso_ts->sec = tk->xtime_sec + tk->monotonic_to_aux.tv_sec;
 
                nsec = tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift;
-               nsec += tk->offs_aux;
+               nsec += tk->monotonic_to_aux.tv_nsec;
                vdso_ts->sec += __iter_div_u64_rem(nsec, NSEC_PER_SEC, &nsec);
                nsec = nsec << tk->tkr_mono.shift;
                vdso_ts->nsec = nsec;