posix-cpu-timers: Get rid of 64bit divisions
authorThomas Gleixner <tglx@linutronix.de>
Wed, 21 Aug 2019 19:09:21 +0000 (21:09 +0200)
committerThomas Gleixner <tglx@linutronix.de>
Wed, 28 Aug 2019 09:50:41 +0000 (11:50 +0200)
Instead of dividing A to match the units of B it's more efficient to
multiply B to match the units of A.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lkml.kernel.org/r/20190821192922.458286860@linutronix.de
kernel/time/posix-cpu-timers.c

index caafdfdd6f0fa09ea6c9354f95a5ebf14433b2d9..dcdf9c8241b10f5cced42bb6d29f806f3839fb2e 100644 (file)
@@ -798,10 +798,11 @@ static void check_thread_timers(struct task_struct *tsk,
         */
        soft = task_rlimit(tsk, RLIMIT_RTTIME);
        if (soft != RLIM_INFINITY) {
+               /* Task RT timeout is accounted in jiffies. RTTIME is usec */
+               unsigned long rtim = tsk->rt.timeout * (USEC_PER_SEC / HZ);
                unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
 
-               if (hard != RLIM_INFINITY &&
-                   tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
+               if (hard != RLIM_INFINITY && rtim >= hard) {
                        /*
                         * At the hard limit, we just die.
                         * No need to calculate anything else now.
@@ -813,7 +814,7 @@ static void check_thread_timers(struct task_struct *tsk,
                        __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
                        return;
                }
-               if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
+               if (rtim >= soft) {
                        /*
                         * At the soft limit, send a SIGXCPU every second.
                         */
@@ -910,11 +911,13 @@ static void check_process_timers(struct task_struct *tsk,
 
        soft = task_rlimit(tsk, RLIMIT_CPU);
        if (soft != RLIM_INFINITY) {
-               u64 softns, ptime = samples[CPUCLOCK_PROF];
+               /* RLIMIT_CPU is in seconds. Samples are nanoseconds */
                unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
-               unsigned long psecs = div_u64(ptime, NSEC_PER_SEC);
+               u64 ptime = samples[CPUCLOCK_PROF];
+               u64 softns = (u64)soft * NSEC_PER_SEC;
+               u64 hardns = (u64)hard * NSEC_PER_SEC;
 
-               if (hard != RLIM_INFINITY && psecs >= hard) {
+               if (hard != RLIM_INFINITY && ptime >= hardns) {
                        /*
                         * At the hard limit, we just die.
                         * No need to calculate anything else now.
@@ -926,7 +929,7 @@ static void check_process_timers(struct task_struct *tsk,
                        __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
                        return;
                }
-               if (psecs >= soft) {
+               if (ptime >= softns) {
                        /*
                         * At the soft limit, send a SIGXCPU every second.
                         */
@@ -936,11 +939,12 @@ static void check_process_timers(struct task_struct *tsk,
                        }
                        __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
                        if (soft < hard) {
-                               soft++;
-                               sig->rlim[RLIMIT_CPU].rlim_cur = soft;
+                               sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
+                               softns += NSEC_PER_SEC;
                        }
                }
-               softns = soft * NSEC_PER_SEC;
+
+               /* Update the expiry cache */
                if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
                        pct->bases[CPUCLOCK_PROF].nextevt = softns;
        }