LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* [RFC][PATCH] cpuidle: avoid singing capacitors
@ 2008-02-29 18:38 Pierre Ossman
  2008-02-29 21:44 ` Lennart Sorensen
                   ` (3 more replies)
  0 siblings, 4 replies; 11+ messages in thread
From: Pierre Ossman @ 2008-02-29 18:38 UTC (permalink / raw)
  To: Venkatesh Pallipadi, Adam Belay; +Cc: linux-pm, LKML

Many devices today are of a less than stellar quality, and singing
transistors are a common problem. A high-pitch noise is created, caused
by power fluctuations as the processor enters and leaves deep sleep at
a high frequency.

Instead of just disabling the deep sleep (which wastes power). This
patch merely reduces the number of times it is entered so that the
frequency doesn't exceed 500 Hz. That should make sure the problem is
inaudible.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
--

The basic idea is above, but the implementation doesn't quite work. It seems jiffies is not a good time source in this scenario. The time spent in C3 takes a much bigger hit than I expect. The fact that powertop says that it has an average residency of 200 ms in C2 tells me that those should have been spent in C3.

So, pointers on what else to do?

(Patch also needs an on/off switch, but that's a later problem)

diff --git a/drivers/cpuidle/governors/menu.c b/drivers/cpuidle/governors/menu.c
index 78d77c5..171d838 100644
--- a/drivers/cpuidle/governors/menu.c
+++ b/drivers/cpuidle/governors/menu.c
@@ -16,6 +16,12 @@
 
 #define BREAK_FUZZ     4       /* 4 us */
 
+/*
+ * The minimum number of ticks needed to not oscillate faster than
+ * 500 Hz.
+ */
+#define MIN_DEEP_INTERVAL (HZ / 500)
+
 struct menu_device {
        int             last_state_idx;
 
@@ -23,6 +29,8 @@ struct menu_device {
        unsigned int    predicted_us;
        unsigned int    last_measured_us;
        unsigned int    elapsed_us;
+
+       unsigned long   last_deep_jif;
 };
 
 static DEFINE_PER_CPU(struct menu_device, menu_devices);
@@ -50,9 +58,16 @@ static int menu_select(struct cpuidle_device *dev)
                        break;
                if (s->exit_latency > pm_qos_requirement(PM_QOS_CPU_DMA_LATENCY))
                        break;
+               if ((dev->states[i].flags & CPUIDLE_FLAG_DEEP) &&
+                       time_before_eq(jiffies, data->last_deep_jif + MIN_DEEP_INTERVAL))
+                       break;
        }
 
        data->last_state_idx = i - 1;
+
+       if (dev->states[i - 1].flags & CPUIDLE_FLAG_DEEP)
+               data->last_deep_jif = jiffies;
+
        return i - 1;
 }
 


-- 
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  PulseAudio, core developer          http://pulseaudio.org
  rdesktop, core developer          http://www.rdesktop.org

^ permalink raw reply	[flat|nested] 11+ messages in thread

end of thread, other threads:[~2008-03-05 18:03 UTC | newest]

Thread overview: 11+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <a2itf-5jB-27@gated-at.bofh.it>
2008-02-29 21:09 ` [RFC][PATCH] cpuidle: avoid singing capacitors Alan Jenkins
2008-03-01 12:29   ` Pierre Ossman
2008-03-05 17:32     ` Jeremy Fitzhardinge
2008-03-05 18:01       ` Pierre Ossman
2008-02-29 18:38 Pierre Ossman
2008-02-29 21:44 ` Lennart Sorensen
2008-03-01 12:31   ` Pierre Ossman
2008-03-01 13:40 ` Pierre Ossman
2008-03-02  2:27 ` Lee Revell
2008-03-02 14:17   ` Pierre Ossman
2008-03-03 12:36 ` Andi Kleen

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).