git: 93ca6ff2958c - main - umtx: allow to configure minimal timeout (in nanoseconds)

From: Konstantin Belousov <kib_at_FreeBSD.org>
Date: Tue, 18 Apr 2023 23:22:56 UTC
The branch main has been updated by kib:

URL: https://cgit.FreeBSD.org/src/commit/?id=93ca6ff2958cace3b7b883ed797ea6539881a6a2

commit 93ca6ff2958cace3b7b883ed797ea6539881a6a2
Author:     Konstantin Belousov <kib@FreeBSD.org>
AuthorDate: 2023-04-15 17:46:12 +0000
Commit:     Konstantin Belousov <kib@FreeBSD.org>
CommitDate: 2023-04-18 23:22:28 +0000

    umtx: allow to configure minimal timeout (in nanoseconds)
    
    PR:     270785
    Reviewed by:    markj, mav
    Sponsored by:   The FreeBSD Foundation
    MFC after:      1 week
    Differential revision:  https://reviews.freebsd.org/D39584
---
 lib/libc/sys/_umtx_op.2      | 11 +++++++++++
 lib/libthr/libthr.3          |  6 ++++++
 lib/libthr/thread/thr_init.c | 11 +++++++++++
 sys/kern/kern_thread.c       |  4 ++--
 sys/kern/kern_umtx.c         | 45 ++++++++++++++++++++++++++++++++++++++++++++
 sys/sys/proc.h               |  1 +
 sys/sys/umtx.h               |  2 ++
 7 files changed, 78 insertions(+), 2 deletions(-)

diff --git a/lib/libc/sys/_umtx_op.2 b/lib/libc/sys/_umtx_op.2
index f4a908b85017..93078a765495 100644
--- a/lib/libc/sys/_umtx_op.2
+++ b/lib/libc/sys/_umtx_op.2
@@ -1270,6 +1270,17 @@ in which case list processing is continued.
 See
 .Sx ROBUST UMUTEXES
 subsection for details.
+.It Dv UMTX_OP_GET_MIN_TIMEOUT
+Writes out the current value of minimal umtx operations timeout,
+in nanoseconds, into the long integer variable pointed to by
+.Fa uaddr1 .
+.It Dv UMTX_OP_SET_MIN_TIMEOUT
+Set the minimal amount of time, in nanoseconds, the thread is required
+to sleep for umtx operations specifying a timeout using absolute clocks.
+The value is taken from the
+.Fa val
+argument of the call.
+Zero means no minimum.
 .El
 .Pp
 The
diff --git a/lib/libthr/libthr.3 b/lib/libthr/libthr.3
index 3018a6f20b86..0477ac0647a8 100644
--- a/lib/libthr/libthr.3
+++ b/lib/libthr/libthr.3
@@ -196,6 +196,12 @@ The integer value of the variable specifies how often blocked
 threads are inserted at the head of the sleep queue, instead of its tail.
 Bigger values reduce the frequency of the FIFO discipline.
 The value must be between 0 and 255.
+.It Dv LIBPTHREAD_UMTX_MIN_TIMEOUT
+The minimal amount of time, in nanoseconds, the thread is required to sleep
+for pthread operations specifying a timeout.
+If the operation requests a timeout less than the value provided,
+it is silently increased to the value.
+The value of zero means no minimum (default).
 .Pp
 .El
 The following
diff --git a/lib/libthr/thread/thr_init.c b/lib/libthr/thread/thr_init.c
index 1c4867cb8185..6010dd4c0451 100644
--- a/lib/libthr/thread/thr_init.c
+++ b/lib/libthr/thread/thr_init.c
@@ -522,6 +522,17 @@ init_private(void)
 		if (env)
 			_thr_queuefifo = atoi(env);
 		TAILQ_INIT(&_thr_atfork_list);
+		env = getenv("LIBPTHREAD_UMTX_MIN_TIMEOUT");
+		if (env) {
+			char *endptr;
+			long mint;
+
+			mint = strtol(env, &endptr, 0);
+			if (*endptr == '\0' && mint >= 0) {
+				_umtx_op(NULL, UMTX_OP_SET_MIN_TIMEOUT,
+				    mint, NULL, NULL);
+			}
+		}
 	}
 	init_once = 1;
 }
diff --git a/sys/kern/kern_thread.c b/sys/kern/kern_thread.c
index ad54783dd4a9..0e432d6978e6 100644
--- a/sys/kern/kern_thread.c
+++ b/sys/kern/kern_thread.c
@@ -101,7 +101,7 @@ _Static_assert(offsetof(struct proc, p_filemon) == 0x3c8,
     "struct proc KBI p_filemon");
 _Static_assert(offsetof(struct proc, p_comm) == 0x3e0,
     "struct proc KBI p_comm");
-_Static_assert(offsetof(struct proc, p_emuldata) == 0x4c8,
+_Static_assert(offsetof(struct proc, p_emuldata) == 0x4d0,
     "struct proc KBI p_emuldata");
 #endif
 #ifdef __i386__
@@ -121,7 +121,7 @@ _Static_assert(offsetof(struct proc, p_filemon) == 0x270,
     "struct proc KBI p_filemon");
 _Static_assert(offsetof(struct proc, p_comm) == 0x284,
     "struct proc KBI p_comm");
-_Static_assert(offsetof(struct proc, p_emuldata) == 0x310,
+_Static_assert(offsetof(struct proc, p_emuldata) == 0x318,
     "struct proc KBI p_emuldata");
 #endif
 
diff --git a/sys/kern/kern_umtx.c b/sys/kern/kern_umtx.c
index 86d644c835c7..92c7de6f2a2c 100644
--- a/sys/kern/kern_umtx.c
+++ b/sys/kern/kern_umtx.c
@@ -700,6 +700,19 @@ umtx_abs_timeout_init2(struct umtx_abs_timeout *timo,
 	    (umtxtime->_flags & UMTX_ABSTIME) != 0, &umtxtime->_timeout);
 }
 
+static void
+umtx_abs_timeout_enforce_min(sbintime_t *sbt)
+{
+	sbintime_t when, mint;
+
+	mint = curproc->p_umtx_min_timeout;
+	if (__predict_false(mint != 0)) {
+		when = sbinuptime() + mint;
+		if (*sbt < when)
+			*sbt = when;
+	}
+}
+
 static int
 umtx_abs_timeout_getsbt(struct umtx_abs_timeout *timo, sbintime_t *sbt,
     int *flags)
@@ -739,6 +752,7 @@ umtx_abs_timeout_getsbt(struct umtx_abs_timeout *timo, sbintime_t *sbt,
 			return (0);
 		}
 		*sbt = bttosbt(bt);
+		umtx_abs_timeout_enforce_min(sbt);
 
 		/*
 		 * Check if the absolute time should be aligned to
@@ -4594,6 +4608,33 @@ __umtx_op_robust_lists(struct thread *td, struct _umtx_op_args *uap,
 	return (0);
 }
 
+static int
+__umtx_op_get_min_timeout(struct thread *td, struct _umtx_op_args *uap,
+    const struct umtx_copyops *ops)
+{
+	long val;
+	int error, val1;
+
+	val = sbttons(td->td_proc->p_umtx_min_timeout);
+	if (ops->compat32) {
+		val1 = (int)val;
+		error = copyout(&val1, uap->uaddr1, sizeof(val1));
+	} else {
+		error = copyout(&val, uap->uaddr1, sizeof(val));
+	}
+	return (error);
+}
+
+static int
+__umtx_op_set_min_timeout(struct thread *td, struct _umtx_op_args *uap,
+    const struct umtx_copyops *ops)
+{
+	if (uap->val < 0)
+		return (EINVAL);
+	td->td_proc->p_umtx_min_timeout = nstosbt(uap->val);
+	return (0);
+}
+
 #if defined(__i386__) || defined(__amd64__)
 /*
  * Provide the standard 32-bit definitions for x86, since native/compat32 use a
@@ -4816,6 +4857,8 @@ static const _umtx_op_func op_table[] = {
 	[UMTX_OP_SEM2_WAKE]	= __umtx_op_sem2_wake,
 	[UMTX_OP_SHM]		= __umtx_op_shm,
 	[UMTX_OP_ROBUST_LISTS]	= __umtx_op_robust_lists,
+	[UMTX_OP_GET_MIN_TIMEOUT] = __umtx_op_get_min_timeout,
+	[UMTX_OP_SET_MIN_TIMEOUT] = __umtx_op_set_min_timeout,
 };
 
 static const struct umtx_copyops umtx_native_ops = {
@@ -4990,6 +5033,8 @@ umtx_exec(struct proc *p)
 		umtx_thread_cleanup(td);
 		td->td_rb_list = td->td_rbp_list = td->td_rb_inact = 0;
 	}
+
+	p->p_umtx_min_timeout = 0;
 }
 
 /*
diff --git a/sys/sys/proc.h b/sys/sys/proc.h
index 172e716d8785..1292315990b3 100644
--- a/sys/sys/proc.h
+++ b/sys/sys/proc.h
@@ -741,6 +741,7 @@ struct proc {
 	uint64_t	p_elf_flags;	/* (x) ELF flags */
 	void		*p_elf_brandinfo; /* (x) Elf_Brandinfo, NULL for
 						 non ELF binaries. */
+	sbintime_t	p_umtx_min_timeout;
 /* End area that is copied on creation. */
 #define	p_endcopy	p_xexit
 
diff --git a/sys/sys/umtx.h b/sys/sys/umtx.h
index 82a8e0e4b0f0..ddd3bd387099 100644
--- a/sys/sys/umtx.h
+++ b/sys/sys/umtx.h
@@ -103,6 +103,8 @@
 #define	UMTX_OP_SEM2_WAKE	24
 #define	UMTX_OP_SHM		25
 #define	UMTX_OP_ROBUST_LISTS	26
+#define	UMTX_OP_GET_MIN_TIMEOUT	27
+#define	UMTX_OP_SET_MIN_TIMEOUT	28
 
 /*
  * Flags for ops; the double-underbar convention must be maintained for future