OK - another go. Perhaps everyone slept through this the last time, cos
I can't see it being very contentious...
Anyway - if it sucks, please tell me...
Neil
ps: this needs to go into 2.2 and 2.3 I guess.
--------------E7E0228CC5B597DB152B972E
Content-Type: message/rfc822
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
Message-ID: <373AB53E.C662FFD7@ukaea.org.uk>
Date: Thu, 13 May 1999 12:19:26 +0100
From: Neil Conway <nconway.list@ukaea.org.uk>
Organization: Fusion
X-Mailer: Mozilla 4.06 [en] (X11; I; Linux 2.0.35 i686)
MIME-Version: 1.0
To: "linux-kernel@vger.rutgers.edu" <linux-kernel@vger.rutgers.edu>,
Linus Torvalds <torvalds@transmeta.com>
Subject: [PATCH] Assorted counter/sched stuff] (resend)
Content-Type: multipart/mixed; boundary="------------B308A2BA919884E255927F5D"
This is a multi-part message in MIME format.
--------------B308A2BA919884E255927F5D
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
This is a resend of a patch I made back in January. It was a bit
premature then but since 2.3 has just forked I guess the world should be
ready for it now ;-))
*** NB: I'm off-list right now, so please CC me if you want me to notice
a reply. ***
I just checked it, and it applies cleanly to 2.2.8, even though it's a
little grey and dusty by now...
cheers
Neil
PS: As I see it, this will mean that nice +20 will actually cut things
to 5% of CPU instead of 10%ish. I forgot to mention that first time
around. If I'm being brain-dead please feel free to tell me though.
--------------B308A2BA919884E255927F5D
Content-Type: message/rfc822
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
Message-ID: <36AE4B22.486D842B@ukaea.org.uk>
Date: Tue, 26 Jan 1999 23:09:22 +0000
From: Neil Conway <nconway.list@ukaea.org.uk>
Organization: Fusion
X-Mailer: Mozilla 4.06 [en] (X11; I; Linux 2.0.35 i686)
MIME-Version: 1.0
To: Linus Torvalds <torvalds@transmeta.com>,
"linux-kernel@vger.rutgers.edu" <linux-kernel@vger.rutgers.edu>
Subject: [PATCH] Assorted counter/sched stuff
Content-Type: multipart/mixed; boundary="------------06F43BBD6BF0811BBB6B78A0"
This is a multi-part message in MIME format.
--------------06F43BBD6BF0811BBB6B78A0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
(I didn't get this done before pre9 so I waited till now!)
This is a collection of two types of fix: (a) drivers doing
"current->counter =0;schedule()" with TASK_INTERRUPTIBLE set; (b) lots
of architectures doing "if (current->counter <0)" when they should
really have tested for "<= 0". (I don't know if it's safe to test "==
0" since I'm not certain that setting need_resched will ALWAYS prevent
another timer-tick happening in the same context.)
This also reverts the timeslice length comments to "200 ms" (calling it
"210 ms" was incorrect on Alphas anyway due to HZ != 100).
There are still loads of places (mostly in the fs code) where we do
"current->counter =0;schedule()", but I don't know of any better way to
do the waits in those cases. Perhaps they are conceptually optimal
anyway.
Neil
--------------06F43BBD6BF0811BBB6B78A0
Content-Type: text/plain; charset=us-ascii; name="njccounterpatch"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline; filename="njccounterpatch"
diff -ur linux-2.2.0/arch/alpha/kernel/smp.c linux-2.2.0.njc/arch/alpha/kernel/smp.c
--- linux-2.2.0/arch/alpha/kernel/smp.c Sun Jan 17 01:02:50 1999
+++ linux-2.2.0.njc/arch/alpha/kernel/smp.c Tue Jan 26 22:42:14 1999
@@ -390,7 +390,7 @@
if (current->pid) {
update_one_process(current, 1, user, !user, cpu);
- if (--current->counter < 0) {
+ if (--current->counter <= 0) {
current->counter = 0;
current->need_resched = 1;
}
diff -ur linux-2.2.0/arch/i386/kernel/smp.c linux-2.2.0.njc/arch/i386/kernel/smp.c
--- linux-2.2.0/arch/i386/kernel/smp.c Tue Jan 26 18:09:10 1999
+++ linux-2.2.0.njc/arch/i386/kernel/smp.c Tue Jan 26 22:49:43 1999
@@ -1696,7 +1696,7 @@
update_one_process(p, 1, user, system, cpu);
p->counter -= 1;
- if (p->counter < 0) {
+ if (p->counter <= 0) {
p->counter = 0;
p->need_resched = 1;
}
diff -ur linux-2.2.0/arch/ppc/kernel/smp.c linux-2.2.0.njc/arch/ppc/kernel/smp.c
--- linux-2.2.0/arch/ppc/kernel/smp.c Wed Dec 30 18:55:07 1998
+++ linux-2.2.0.njc/arch/ppc/kernel/smp.c Tue Jan 26 22:43:06 1999
@@ -77,7 +77,7 @@
update_one_process(p, 1, user, system, cpu);
p->counter -= 1;
- if (p->counter < 0) {
+ if (p->counter <= 0) {
p->counter = 0;
current->need_resched = 1;
}
diff -ur linux-2.2.0/arch/sparc/kernel/sun4d_smp.c linux-2.2.0.njc/arch/sparc/kernel/sun4d_smp.c
--- linux-2.2.0/arch/sparc/kernel/sun4d_smp.c Mon Nov 16 18:37:28 1998
+++ linux-2.2.0.njc/arch/sparc/kernel/sun4d_smp.c Tue Jan 26 22:42:37 1999
@@ -471,7 +471,7 @@
if(current->pid) {
update_one_process(current, 1, user, !user, cpu);
- if(--current->counter < 0) {
+ if(--current->counter <= 0) {
current->counter = 0;
current->need_resched = 1;
}
diff -ur linux-2.2.0/arch/sparc/kernel/sun4m_smp.c linux-2.2.0.njc/arch/sparc/kernel/sun4m_smp.c
--- linux-2.2.0/arch/sparc/kernel/sun4m_smp.c Mon Nov 16 18:37:28 1998
+++ linux-2.2.0.njc/arch/sparc/kernel/sun4m_smp.c Tue Jan 26 22:42:54 1999
@@ -451,7 +451,7 @@
if(current->pid) {
update_one_process(current, 1, user, !user, cpu);
- if(--current->counter < 0) {
+ if(--current->counter <= 0) {
current->counter = 0;
current->need_resched = 1;
}
diff -ur linux-2.2.0/arch/sparc64/kernel/smp.c linux-2.2.0.njc/arch/sparc64/kernel/smp.c
--- linux-2.2.0/arch/sparc64/kernel/smp.c Tue Oct 27 17:52:20 1998
+++ linux-2.2.0.njc/arch/sparc64/kernel/smp.c Tue Jan 26 22:43:37 1999
@@ -578,7 +578,7 @@
unsigned int *inc, *inc2;
update_one_process(current, 1, user, !user, cpu);
- if(--current->counter < 0) {
+ if(--current->counter <= 0) {
current->counter = 0;
current->need_resched = 1;
}
diff -ur linux-2.2.0/drivers/char/cyclades.c linux-2.2.0.njc/drivers/char/cyclades.c
--- linux-2.2.0/drivers/char/cyclades.c Wed Jan 20 21:26:20 1999
+++ linux-2.2.0.njc/drivers/char/cyclades.c Tue Jan 26 22:37:33 1999
@@ -2597,7 +2597,6 @@
printk("Not clean (jiff=%lu)...", jiffies);
#endif
current->state = TASK_INTERRUPTIBLE;
- current->counter = 0; /* make us low-priority */
schedule_timeout(char_time);
if (signal_pending(current))
break;
@@ -2610,7 +2609,6 @@
}
/* Run one more char cycle */
current->state = TASK_INTERRUPTIBLE;
- current->counter = 0; /* make us low-priority */
schedule_timeout(char_time * 5);
current->state = TASK_RUNNING;
#ifdef CY_DEBUG_WAIT_UNTIL_SENT
diff -ur linux-2.2.0/drivers/char/esp.c linux-2.2.0.njc/drivers/char/esp.c
--- linux-2.2.0/drivers/char/esp.c Sat Oct 31 19:06:21 1998
+++ linux-2.2.0.njc/drivers/char/esp.c Tue Jan 26 22:38:03 1999
@@ -2180,7 +2180,6 @@
while ((serial_in(info, UART_ESI_STAT1) != 0x03) ||
(serial_in(info, UART_ESI_STAT2) != 0xff)) {
current->state = TASK_INTERRUPTIBLE;
- current->counter = 0;
schedule_timeout(char_time);
if (signal_pending(current))
diff -ur linux-2.2.0/drivers/char/rocket.c linux-2.2.0.njc/drivers/char/rocket.c
--- linux-2.2.0/drivers/char/rocket.c Thu Nov 5 17:58:43 1998
+++ linux-2.2.0.njc/drivers/char/rocket.c Tue Jan 26 22:38:25 1999
@@ -1663,7 +1663,6 @@
jiffies, check_time);
#endif
current->state = TASK_INTERRUPTIBLE;
- current->counter = 0; /* make us low-priority */
schedule_timeout(check_time);
if (signal_pending(current))
break;
diff -ur linux-2.2.0/drivers/char/serial.c linux-2.2.0.njc/drivers/char/serial.c
--- linux-2.2.0/drivers/char/serial.c Thu Jan 14 18:27:53 1999
+++ linux-2.2.0.njc/drivers/char/serial.c Tue Jan 26 22:38:48 1999
@@ -2376,7 +2376,6 @@
printk("lsr = %d (jiff=%lu)...", lsr, jiffies);
#endif
current->state = TASK_INTERRUPTIBLE;
- current->counter = 0; /* make us low-priority */
schedule_timeout(char_time);
if (signal_pending(current))
break;
diff -ur linux-2.2.0/drivers/macintosh/macserial.c linux-2.2.0.njc/drivers/macintosh/macserial.c
--- linux-2.2.0/drivers/macintosh/macserial.c Wed Dec 23 17:44:41 1998
+++ linux-2.2.0.njc/drivers/macintosh/macserial.c Tue Jan 26 22:39:57 1999
@@ -1324,7 +1324,6 @@
char_time = MIN(char_time, timeout);
while ((read_zsreg(info->zs_channel, 1) & ALL_SNT) == 0) {
current->state = TASK_INTERRUPTIBLE;
- current->counter = 0; /* make us low-priority */
schedule_timeout(char_time);
if (signal_pending(current))
break;
diff -ur linux-2.2.0/drivers/sbus/char/sab82532.c linux-2.2.0.njc/drivers/sbus/char/sab82532.c
--- linux-2.2.0/drivers/sbus/char/sab82532.c Wed Dec 23 17:44:41 1998
+++ linux-2.2.0.njc/drivers/sbus/char/sab82532.c Tue Jan 26 22:39:15 1999
@@ -1672,7 +1672,6 @@
#endif
while (info->xmit_cnt || !info->all_sent) {
current->state = TASK_INTERRUPTIBLE;
- current->counter = 0;
schedule_timeout(char_time);
if (signal_pending(current))
break;
diff -ur linux-2.2.0/drivers/sbus/char/su.c linux-2.2.0.njc/drivers/sbus/char/su.c
--- linux-2.2.0/drivers/sbus/char/su.c Wed Dec 23 17:44:41 1998
+++ linux-2.2.0.njc/drivers/sbus/char/su.c Tue Jan 26 22:39:35 1999
@@ -1814,7 +1814,6 @@
printk("lsr = %d (jiff=%lu)...", lsr, jiffies);
#endif
current->state = TASK_INTERRUPTIBLE;
- current->counter = 0; /* make us low-priority */
schedule_timeout(char_time);
if (signal_pending(current))
break;
diff -ur linux-2.2.0/include/linux/sched.h linux-2.2.0.njc/include/linux/sched.h
--- linux-2.2.0/include/linux/sched.h Tue Jan 26 00:06:22 1999
+++ linux-2.2.0.njc/include/linux/sched.h Tue Jan 26 22:44:43 1999
@@ -338,7 +338,7 @@
*/
#define _STK_LIM (8*1024*1024)
-#define DEF_PRIORITY (20*HZ/100) /* 210 ms time slices */
+#define DEF_PRIORITY (20*HZ/100) /* 200 ms time slices */
/*
* INIT_TASK is used to set up the first task table, touch at
diff -ur linux-2.2.0/kernel/sched.c linux-2.2.0.njc/kernel/sched.c
--- linux-2.2.0/kernel/sched.c Tue Jan 19 18:19:58 1999
+++ linux-2.2.0.njc/kernel/sched.c Tue Jan 26 22:52:31 1999
@@ -1350,7 +1350,7 @@
unsigned long user = ticks - system;
if (p->pid) {
p->counter -= ticks;
- if (p->counter < 0) {
+ if (p->counter <= 0) {
p->counter = 0;
p->need_resched = 1;
}
@@ -1553,7 +1553,7 @@
* do a "normalization" of the priority (traditionally
* Unix nice values are -20 to 20; Linux doesn't really
* use that kind of thing, but uses the length of the
- * timeslice instead (default 210 ms). The rounding is
+ * timeslice instead (default 200 ms). The rounding is
* why we want to avoid negative values.
*/
newprio = (newprio * DEF_PRIORITY + 10) / 20;
--------------06F43BBD6BF0811BBB6B78A0--
--------------B308A2BA919884E255927F5D--
--------------E7E0228CC5B597DB152B972E--
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/