1 Dec
2006
1 Dec
'06
4:42 a.m.
Hi
Reading the thread(s) on realtime-kernels, I decided to have another go
at putting my sound card (IBM T41 laptop) on it's own IRQ. Through luck,
some black BIOS-magic and help from linux thinkpad mailing list I
managed to move the sound card, PC-card, modem and SMBus
controller(whatever that is) to their own IRQ. I'm still trying to get
*only* the sound card's IRQ moved.
My question is: Since I don't use the pc-card or the modem, and provided
that SMBus is something I don't use, will having these extra stuff on
the same IRQ as the sound card be a problem? Will it make any difference
if I don't load the kernel modules for the pc card/modem thereby
disabling them?
--
peace, love & harmony
Atte
http://www.atte.dk | quintet: http://www.anagrammer.dk
| compositions: http://www.atte.dk/compositions
1 Dec
1 Dec
3:45 p.m.
I found this. It looks like it might be
of interest to you.
http://cae.best.vwh.net/irqtune/
Atte André Jensen wrote:
Hi
Reading the thread(s) on realtime-kernels, I decided to have another go
at putting my sound card (IBM T41 laptop) on it's own IRQ. Through luck,
some black BIOS-magic and help from linux thinkpad mailing list I
managed to move the sound card, PC-card, modem and SMBus
controller(whatever that is) to their own IRQ. I'm still trying to get
*only* the sound card's IRQ moved.
My question is: Since I don't use the pc-card or the modem, and provided
that SMBus is something I don't use, will having these extra stuff on
the same IRQ as the sound card be a problem? Will it make any difference
if I don't load the kernel modules for the pc card/modem thereby
disabling them?
4:05 p.m.
On Fri, 2006-12-01 at 11:39 -0800, st wrote:
I found this. It looks like it might be
of interest to you.
http://cae.best.vwh.net/irqtune/
Useless with -rt kernel which already lets you prioritize interrupts by
setting the RT priority of the handler.
IRQ sharing is hard wired in a laptop - it's not going to be possible
for you to get the soundcard on its own IRQ. It will help if you don't
load drivers for other devices sharing that IRQ though.
Anyway having your soundcard on a non-shared IRQ is overrated - as long
as you aren't sharing it with a "heavy" device like a disk controller or
ethernet adapter. If you find it impossible to eliminate xruns on a
laptop there's a good chance that SMI is the problem and you must use a
higher latency.
Lee
Atte André Jensen wrote:
Hi
Reading the thread(s) on realtime-kernels, I decided to have another go
at putting my sound card (IBM T41 laptop) on it's own IRQ. Through luck,
some black BIOS-magic and help from linux thinkpad mailing list I
managed to move the sound card, PC-card, modem and SMBus
controller(whatever that is) to their own IRQ. I'm still trying to get
*only* the sound card's IRQ moved.
My question is: Since I don't use the pc-card or the modem, and provided
that SMBus is something I don't use, will having these extra stuff on
the same IRQ as the sound card be a problem? Will it make any difference
if I don't load the kernel modules for the pc card/modem thereby
disabling them?
4:53 p.m.
Lee Revell wrote:
IRQ sharing is hard wired in a laptop - it's not
going to be possible
for you to get the soundcard on its own IRQ. It will help if you
don't load drivers for other devices sharing that IRQ though.
Well, at least I found out how move interrupts of "groups" of hardware
in the BIOS. Before the sound card was using the same interrupts as the
usb controller, the graphics card and wlan.
After doing this + setting the priority of the audiocards (+ unused pc
cards + unused modem + SMBus controller (still not sure what that is),
these share "group") I found that the source of all my trouble was
powernowd. I removed that (actually I don't feel any difference, maybe
battery will last longer) and things work *extremely much* better.
For instance I can run jack forever (while going wild on opening
programs, switching desktops and anything I can think of) without any
xruns at 11ms latency. The latest song I'm working on has 9 vocal tracks
with fx in ardour + seq24 driving specimen and 9 (!) parts in
zynaddsubfx with quite a busy arrangement. It runs smooth as butter and
I can do what-ever else I do on the laptop!
Actually I can go even lower in latency without any xruns, but this song
(that I've been using as a test) will glitch from time to time, I
suspect its zynaddsubfx that can't keep up. No xruns, just a single
click every minute or so from (I suspect) zynaddsubfx.
Anyway having your soundcard on a non-shared IRQ is
overrated - as
long as you aren't sharing it with a "heavy" device like a disk
controller or ethernet adapter. If you find it impossible to
eliminate xruns on a laptop there's a good chance that SMI is the
problem and you must use a higher latency.
I was quite surprised that the performance is (as far as I can tell) the
same whether or not I put the audio card on an interrupt together with
the afore mentioned unused devices + set it's priority. Actually I'd be
most interested in hearing your thoughts on the following:
1) Am I right in doing this:
chrt -f -p 90 `pidof "IRQ 11"`
chrt -f -p 85 `pidof "IRQ 8"`
provided my soundcard (+ idle stuff) is on IRQ 11 and rtc is on IRQ 8?
2) If the above is correct (or at least better than nothing) shouldn't I
be getting at least *some* performance difference?
BTW: it's on an IBM T41 1.6Ghz Pentium M w 1G ram, with debian/unstable,
mingo-patched 2.6.18-rt7 kernel and openbox.
--
peace, love & harmony
Atte
http://www.atte.dk | quintet: http://www.anagrammer.dk
| compositions: http://www.atte.dk/compositions
7:34 p.m.
On Friday 01 December 2006 21:53, Atte André Jensen wrote:
1) Am I right in doing this:
chrt -f -p 90 `pidof "IRQ 11"`
chrt -f -p 85 `pidof "IRQ 8"`
provided my soundcard (+ idle stuff) is on IRQ 11 and rtc is on IRQ 8?
2) If the above is correct (or at least better than nothing) shouldn't I
be getting at least *some* performance difference?
You should probably also set jackd to use a prio of around 70 or so [assuming
the other irq handlers are left to their defaults in the 50s range]..
Otherwise these other irq handlers might steal the cpu from jackd.
Flo
--
Palimm Palimm!
http://tapas.affenbande.org
7:52 p.m.
Florian Schmidt wrote:
You should probably also set jackd to use a prio of
around 70 or so [assuming
the other irq handlers are left to their defaults in the 50s range]..
I'm running jackd at priority 71, seq24 at 86, specimen, zynaddsubfx and
ardour at 70...
--
peace, love & harmony
Atte
http://www.atte.dk | quintet: http://www.anagrammer.dk
| compositions: http://www.atte.dk/compositions
12 Dec
12 Dec
7:12 a.m.
2006/12/1, Atte André Jensen <atte.jensen(a)gmail.com>om>:
After doing this + setting the priority of the
audiocards (+ unused pc
cards + unused modem + SMBus controller (still not sure what that is),
these share "group") I found that the source of all my trouble was
powernowd. I removed that (actually I don't feel any difference, maybe
battery will last longer) and things work *extremely much* better.
Well, powernowd changes the cpu-speed according to system-usage. These
changes cause jack to bail out _every_ time this happens at my
machine. So before doing any audio-work I always select the
performance-governour to control cpu-speed...
Arnold
--
visit http://www.arnoldarts.de/
---
Wenn man mit Raubkopien Bands wie Brosis oder Britney Spears wirklich
verhindern könnte, würde ich mir noch heute einen Stapel Brenner und
einen Sack Rohlinge kaufen.
6558
days inactive
6569
days old
linux-audio-user@lists.linuxaudio.org
6 comments
5 participants
participants (5)
-
Arnold Krille -
Atte André Jensen -
Florian Schmidt -
Lee Revell -
st