Bruno 2010-12-16 02:41:51 PST
Fix segfault in ALSA_OpenDevice()
When alsa is linked to libsdl instead of being loaded on demand at runtime the
alsa macro snd_pcm_hw_params_alloca(ptr); misbehaves.
That macro calls snd_pcm_hw_params_sizeof() in order to determine the amount of
memory to allocate but due to mis-placed define that function is replaced with
ALSA_snd_pcm_hw_params_sizeof by preprocessor (and in case of link to alsa that
function pointer is initialized to itself).
Attached patch fixes the issue.
Description From Michael Stone 2008-09-25 19:27:29 (-) [reply]
To determine whether a pid is occupied with the kill(pid, 0) idiom, you have to
test
#include <signal.h>
#include <errno.h>
kill(pid, 0) < 0 && errno == ESRCH
not just
#include <signal.h>
kill(pid, 0) < 0
otherwise you get incorrect results when pid is running as a different user
(causing kill(pid, 0) to return -1 + EPERM).
src/audio/alsa/SDL_alsa_audio.c is certainly affected by this bug in both
1.2.13 and 1.3-trunk. It probably occurs in other places as well.
If a driver can definitely see available devices, it is chosen. Otherwise,
we'll take the first driver that initializes but saw no devices...this might
be because it can't enumerate them, or there really aren't any available.
This prevents the dsp driver from hogging control when there are no /dev/dsp*
nodes (for example, on a Linux box with ALSA and no OSS emulation).
Notable changes:
- Converters between types are autogenerated. Instead of making multiple
passes over the data with seperate filters for endianess, size, signedness,
etc, converting between data types is always one specialized filter. This
simplifies SDL_BuildAudioCVT(), which otherwise had a million edge cases
with the new types, and makes the actually conversions more CPU cache
friendly. Left a stub for adding specific optimized versions of these
routines (SSE/MMX/Altivec, assembler, etc)
- Autogenerated converters are built by SDL/src/audio/sdlgenaudiocvt.pl. This
does not need to be run unless tweaking the code, and thus doesn't need
integration into the build system.
- Went through all the drivers and tried to weed out all the "Uint16"
references that are better specified with the new SDL_AudioFormat typedef.
- Cleaned out a bunch of hardcoded bitwise magic numbers and replaced them
with new SDL_AUDIO_* macros.
- Added initial float32 and int32 support code. Theoretically, existing
drivers will push these through converters to get the data they want to
feed to the hardware.
Still TODO:
- Optimize and debug new converters.
- Update the CoreAudio backend to accept float32 data directly.
- Other backends, too?
- SDL_LoadWAV() needs to be updated to support int32 and float32 .wav files
(both of which exist and can be generated by 'sox' for testing purposes).
- Update the mixer to handle new datatypes.
- Optionally update SDL_sound and SDL_mixer, etc.
Implemented snd_pcm_sw_params_set_start_threshold() and snd_pcm_sw_params_set_avail_min() in the ALSA 0.9 driver.
This doesn't actually change any latency for me, but it's the right thing to do...
FIXME:
Change #include <stdlib.h> to #include "SDL_stdlib.h"
Change #include <string.h> to #include "SDL_string.h"
Make sure nothing else broke because of this...
surround sound on Linux using the Alsa driver. To use them, naturally
you need a sound card that will do 4 or 6 channels and probably also a
recent version of the Alsa drivers and library. Since the only SDL
output driver that knows about surround sound is the Alsa driver,
you���ll want to choose it, using:
export SDL_AUDIODRIVER=alsa
There are no syntactic changes to the programming API. No new
library calls, no differences in arguments.
There are two semantic changes:
(1) For library calls with number of channels as an argument, formerly
you could use only 1 or 2 for the number of channels. Now you
can also use 4 or 6.
(2) The two "left" and "right" arguments to Mix_SetPanning, for the
case of 4 or 6 channels, no longer simply control the volumes of
the left and right channels. Now the "left" argument is converted
to an angle and Mix_SetPosition is called, and the "right" argu-
ment is ignored.
With two exceptions, so far as I know, the modified SDL12 and
SDL_mixer work the same way as the original versions, when opened for
1 or 2 channel output. The two exceptions are bugs which I fixed.
Well, the first, anyway, is a bug for sure. When rate conversions up
or down by a factor of two are applied (in src/audio/SDL_audiocvt.c),
streams with different numbers of channels (that is, mono and stereo)
are treated the same way: either each sample is copied or every other
sample is omitted. This is ok for mono, but for stereo, it is frames
that should be copied or omitted, where by "frame" I mean a portion of
the stream containing one sample for each channel. (In the SDL source,
confusingly, sometimes frames are called "samples".) So for these
rate conversions, stereo streams have to be treated differently, and
they are, in my modified version.
The other problem that might be characterized as a bug arises
when SDL_mixer is passed a multichannel chunk which does not have an
integral number of frames. Due to the way the effect_position code
loops over frames, when the chunk ends with a partial frame, memory
outside the chunk buffer will be accessed. In the case of stereo,
it���s possible that because malloc may give more memory than requested,
this potential problem never actually causes a segment fault. I don���t
know. For 6 channel chunks, I do know, and it does cause segment
faults.
If SDL_mixer is passed defective chunks and this causes a segment
fault, arguably, that���s not a bug in SDL_mixer. Still, whether or not
it counts as a bug, it���s easy to protect against, so why not? I added
code in mixer.c to discard any partial frame at the end of a chunk.
Then what about when SDL or SDL_mixer is opened for 4 or 6 chan-
nel output? What happens with the parts of the current library
designed for stereo? I don���t know whether I���ve covered all the bases,
but I���ve tried:
(1) For playing 2 channel waves, or other cases where SDL knows it has
to match up a 2 channel source with a 4 or 6 channel output, I���ve
added code in SDL_audiocvt.c to make the necessary conversions.
(2) For playing midis using timidity, I���ve converted timidity to do 4
or 6 channel output, upon request.
(3) For playing mods using mikmod, I put ad hoc code in music.c to
convert the stereo output that mikmod produces to 4 or 6 chan-
nels. Obviously it would be better to change the mikmod code to
mix down into 4 or 6 channels, but I have a hard time following
the code in mikmod, so I didn���t do that.
(4) For playing mp3s, I put ad hoc code in smpeg to copy channels in
the case when 4 or 6 channel output is needed.
(5) There seems to be no problem with .ogg files - stereo .oggs can be
up converted as .wavs are.
(6) The effect_position code in SDL_mixer is now generalized to in-
clude the cases of 4 and 6 channel streams.
I���ve done a very limited amount of compatibility testing for some
of the games using SDL I happen to have. For details, see the file
TESTS.
I���ve put into a separate archive, Surround-SDL-testfiles.tgz, a
couple of 6 channel wave files for testing and a 6 channel ogg file.
If you have the right hardware and version of Alsa, you should be able
to play the wave files with the Alsa utility aplay (and hear all
channels, except maybe lfe, for chan-id.wav, since it���s rather faint).
Don���t expect aplay to give good sound, though. There���s something
wrong with the current version of aplay.
The canyon.ogg file is to test loading of 6 channel oggs. After
patching and compiling, you can play it with playmus. (My version of
ogg123 will not play it, and I had to patch mplayer to get it to play
6 channel oggs.)
Greg Lee <greg@ling.lll.hawaii.edu>
Thus, July 1, 2004
From: Martin_Storsj
Subject: Update for dynamic loading of ALSA
I sent you a patch a few months ago which enables SDL to load ALSA
dynamically. Now I've finally got time to tweak this yet some more. I've
added code from alsa.m4 (from alsa's dev package) to acinclude.m4, and
made the detection of the alsa library name a bit better. I've also
fixed up the loading versioned symbols with dlvsym, so that it falls
back to dlsym.
I wouldn't say the configure script is complete yet, but this is how far
I've come this time, and I'm no expert at those things.
From: Martin_Storsj
Subject: Dynamic loading of ALSA
I recently discovered that SDL can dynamically load ESD and aRts, and
made a patch which adds this same functionality to ALSA.
The update for configure.in isn't too good (it should e.g. look for
libasound.so in other directories than /usr/lib), because I'm not too
good at shellscripting and autoconf.
The reason for using dlfcn.h and dlopen instead of SDL_LoadLibrary and
SDL_LoadFunction is that libasound uses versioned symbols, and it is
necessary to load the correct version using dlvsym. This isn't probably
any real portability issue, because ALSA is linux-only.
nanosleep because nanosleep may not be portable to all systems
using SDL with the ALSA backend. This may be a moot point with
the switch to blocking writes anyway...
Date: Sat, 27 Dec 2003 21:47:36 +0100
From: Michel Daenzer
To: Debian Bug Tracking System
Subject: [SDL] Bug#225252: [PATCH] ALSA fixes
Package: libsdl1.2debian-all
Version: 1.2.6-2
Severity: normal
Tags: patch
For SDL 1.2.6, the ALSA backend was changed to call snd_pcm_open() with
SND_PCM_NONBLOCK. That's a good idea per se, however, it causes high CPU
usage, interrupted sound and stuttering in some games here. Taking a nanosleep
whenever snd_pcm_writei() returns -EAGAIN fixes this, but I think it's more
efficient to use blocking mode for the actual sound playback. Feedback from the
SDL and ALSA lists appreciated.
The patch also fixes the default ALSA device to be used.
From: Stephane Marchesin
Subject: Re: [SDL] compile problem with sdl <-> alsa
The attathced patch does the trick (although there will have to be code
changes when the alsa 0.9 api get deprecated).
On a side-note, the good news is that using alsa 1.0.0rc1 solves the
previous issues I had with my ac97 via sound chipset.
