Skip site navigation (1)Skip section navigation (2) 
Date:      Tue, 17 Apr 2018 16:36:53 +0000 (UTC)
From:      Brooks Davis <brooks@FreeBSD.org>
To:        src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org
Subject:   svn commit: r332650 - head/sys/kern
Message-ID:  <201804171636.w3HGarpw035761@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help 
Author: brooks
Date: Tue Apr 17 16:36:53 2018
New Revision: 332650
URL: https://svnweb.freebsd.org/changeset/base/332650

Log:
  Stop using fuswintr() and suswintr() in the profiler.
  
  Always take the AST path rather than calling MD functions which are
  often implemented as always failing. The is the case on amd64, arm,
  i386, and powerpc. This optimization (inherited from 4.4 Lite) is a
  pessimization on those architectures and is the sole use of these
  functions. They will be removed in a seperate commit.
  
  Reviewed by:	kib
  Sponsored by:	DARPA, AFRL
  Differential Revision:	https://reviews.freebsd.org/D15101

Modified:
  head/sys/kern/subr_prof.c

Modified: head/sys/kern/subr_prof.c
==============================================================================
--- head/sys/kern/subr_prof.c	Tue Apr 17 16:21:23 2018	(r332649)
+++ head/sys/kern/subr_prof.c	Tue Apr 17 16:36:53 2018	(r332650)
@@ -447,10 +447,9 @@ sys_profil(struct thread *td, struct profil_args *uap)
 /*
  * Collect user-level profiling statistics; called on a profiling tick,
  * when a process is running in user-mode.  This routine may be called
- * from an interrupt context.  We try to update the user profiling buffers
- * cheaply with fuswintr() and suswintr().  If that fails, we revert to
- * an AST that will vector us to trap() with a context in which copyin
- * and copyout will work.  Trap will then call addupc_task().
+ * from an interrupt context.  We perform the update with an AST
+ * that will vector us to trap() with a context in which copyin and
+ * copyout will work.  Trap will then call addupc_task().
  *
  * Note that we may (rarely) not get around to the AST soon enough, and
  * lose profile ticks when the next tick overwrites this one, but in this
@@ -461,35 +460,28 @@ void
 addupc_intr(struct thread *td, uintfptr_t pc, u_int ticks)
 {
 	struct uprof *prof;
-	caddr_t addr;
-	u_int i;
-	int v;
 
 	if (ticks == 0)
 		return;
 	prof = &td->td_proc->p_stats->p_prof;
 	PROC_PROFLOCK(td->td_proc);
-	if (pc < prof->pr_off ||
-	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) {
+	if (pc < prof->pr_off || PC_TO_INDEX(pc, prof) >= prof->pr_size) {
 		PROC_PROFUNLOCK(td->td_proc);
 		return;			/* out of range; ignore */
 	}
 
-	addr = prof->pr_base + i;
 	PROC_PROFUNLOCK(td->td_proc);
-	if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + ticks) == -1) {
-		td->td_profil_addr = pc;
-		td->td_profil_ticks = ticks;
-		td->td_pflags |= TDP_OWEUPC;
-		thread_lock(td);
-		td->td_flags |= TDF_ASTPENDING;
-		thread_unlock(td);
-	}
+	td->td_profil_addr = pc;
+	td->td_profil_ticks = ticks;
+	td->td_pflags |= TDP_OWEUPC;
+	thread_lock(td);
+	td->td_flags |= TDF_ASTPENDING;
+	thread_unlock(td);
 }
 
 /*
- * Much like before, but we can afford to take faults here.  If the
- * update fails, we simply turn off profiling.
+ * Actually update the profiling statistics.  If the update fails, we
+ * simply turn off profiling.
  */
 void
 addupc_task(struct thread *td, uintfptr_t pc, u_int ticks)

Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201804171636.w3HGarpw035761>