[linux-audio-user] Re: [a-users] Re: [a-devel] File system and Voluntary Kernel Preemption
16 Jul
2004
16 Jul
'04
2:05 a.m.
On Thu, 2004-07-15 at 04:53, Damien Cirotteau wrote:
On Thu, Jul 15, 2004 at 10:35:47AM +0200, Free
Ekanayaka wrote :
The intrepid user in need of low latency should try 2.6.8-rc1-mm1. I
sent a bunch of XRUN traces to Andrew Morton, and this kernel contains
numerous latency fixes as a result. Some of these fixes were to
extremely important areas (get_user_pages, a fix for
framebuffer-scrolling issue, and more). The improvement should be
major. He is going to be unavailable until July 26th, so the next few
weeks are a great opportunity to test and collect data.
You are correct in that reiserfs as shipped with the kernel should
probably not be used for now if you need low latency. However, I
strongly suspect that SuSe (which users reiser by default) contains
fixes for this which are not in the mainstream kernel. By enabling
CONFIG_SND_DEBUG *and* turning on XRUN debugging at runtime (echo 2 >
proc/asound/cardX/pcmX/xrun_debug) you will be able to see whether this
is affecting you.
Here is the announcement:
http://www.ussg.iu.edu/hypermail/linux/kernel/0407.1/1453.html
Lee
|--==> "DC" == Damien Cirotteau
<damien.cirotteau(a)agnula.org> writes:
DC> Hi all,
DC> some very interesting links submited by simmo on IRC
DC> File system:
DC> http://kerneltrap.org/node/view/3466
DC> Voluntary Kernel Preemption:
DC> http://kerneltrap.org/node/view/3440
Yes these are definitively interesting!
I'll read them carefully and see if we can adopt some new trick :)
Two first thing that i can retain:
- Don't use reseirfs for now (especially with 2.6)
- Enable CONFIG_SND_DEBUG in the kernel so it will be easier to trace
the xruns. It looks that it is a very useful feature to understand where
the xruns comes from and is very good feedback to send to kernel folks 
16 Jul
16 Jul
6:43 a.m.
Some of these fixes were to
extremely important areas (get_user_pages, a fix for
framebuffer-scrolling issue, and more).
It goes against common sense (to anyone with kernel knowledge) to use the
console framebuffer anyway.
The duration of kernel locks and the CPU expense of blowing bitmaps around
like that are shocking.
Seriously. Use text-mode consoles, or a native AGP-driver a la nvidia, ati,
or whatever. But using a "stupid" framebuffer driver is asking for hardcore
latency blow-outs.
=MB=
--
A focus on Quality.
5:34 p.m.
On Fri, 2004-07-16 at 04:33, Malcolm Baldridge wrote:
I am using text-mode consoles, and switching back and forth between
those and X using ctrl-alt-F[1-7]. I sent Andrew Morton some XRUN
traces and he said one of them was caused by framebuffer scolling inside
lock_kernel(), and the fix was to remove (un)lock_kernel() from
do_tty_write(). This particular fix may not be in -mm1, because he
seemed to think it would cause problems (display corruption, etc). It
has not caused any problems for me.
Maybe I am confused about the 'framebuffer' vs. a text-mode console, I
thought they were the same. Anyway, the point is, switching from X to a
text-mode console and back caused XRUNS before, and now it doesn't.
Lee
Some of these
fixes were to
extremely important areas (get_user_pages, a fix for
framebuffer-scrolling issue, and more).
It goes against common sense (to anyone with kernel knowledge) to use the
console framebuffer anyway.
The duration of kernel locks and the CPU expense of blowing bitmaps around
like that are shocking.
Seriously. Use text-mode consoles, or a native AGP-driver a la nvidia, ati,
or whatever. But using a "stupid" framebuffer driver is asking for hardcore
latency blow-outs.
7481
days inactive
7481
days old
linux-audio-user@lists.linuxaudio.org
2 comments
2 participants
participants (2)
-
Lee Revell -
Malcolm Baldridge