MORE NOTES ON HD-AUDIO DRIVER ============================= Takashi Iwai GENERAL ------- HD-audio is the new standard on-board audio component on modern PCs after AC97. Although Linux has been supporting HD-audio since long time ago, there are often problems with new machines. A part of the problem is broken BIOS, and the rest is the driver implementation. This document explains the brief trouble-shooting and debugging methods for the HD-audio hardware. The HD-audio component consists of two parts: the controller chip and the codec chips on the HD-audio bus. Linux provides a single driver for all controllers, snd-hda-intel. Although the driver name contains a word of a well-known harware vendor, it's not specific to it but for all controller chips by other companies. Since the HD-audio controllers are supposed to be compatible, the single snd-hda-driver should work in most cases. But, not surprisingly, there are known bugs and issues specific to each controller type. The snd-hda-intel driver has a bunch of workarounds for these as described below. A controller may have multiple codecs. Usually you have one audio codec and optionally one modem codec. In theory, there might be multiple audio codecs, e.g. for analog and digital outputs, and the driver might not work properly because of conflict of mixer elements. This should be fixed in future if such hardware really exists. The snd-hda-intel driver has several different codec parsers depending on the codec. It has a generic parser as a fallback, but this functionality is fairly limited until now. Instead of the generic parser, usually the codec-specific parser (coded in patch_*.c) is used for the codec-specific implementations. The details about the codec-specific problems are explained in the later sections. If you are interested in the deep debugging of HD-audio, read the HD-audio specification at first. The specification is found on Intel's web page, for example: - http://www.intel.com/standards/hdaudio/ HD-AUDIO CONTROLLER ------------------- DMA-Position Problem ~~~~~~~~~~~~~~~~~~~~ The most common problem of the controller is the inaccurate DMA pointer reporting. The DMA pointer for playback and capture can be read in two ways, either via a LPIB register or via a position-buffer map. As default the driver tries to read from the io-mapped position-buffer, and falls back to LPIB if the position-buffer appears dead. However, this detection isn't perfect on some devices. In such a case, you can change the default method via `position_fix` option. `position_fix=1` means to use LPIB method explicitly. `position_fix=2` means to use the position-buffer. 0 is the default value, the automatic check and fallback to LPIB as described in the above. If you get a problem of repeated sounds, this option might help. In addition to that, every controller is known to be broken regarding the wake-up timing. It wakes up a few samples before actually processing the data on the buffer. This caused a lot of problems, for example, with ALSA dmix or JACK. Since 2.6.27 kernel, the driver puts an artificial delay to the wake up timing. This delay is controlled via `bdl_pos_adj` option. When `bdl_pos_adj` is a negative value (as default), it's assigned to an appropriate value depending on the controller chip. For Intel chips, it'd be 1 while it'd be 32 for others. Usually this works. Only in case it doesn't work and you get warning messages, you should change this parameter to other values. Codec-Probing Problem ~~~~~~~~~~~~~~~~~~~~~ A less often but a more severe problem is the codec probing. When BIOS reports the available codec slots wrongly, the driver gets confused and tries to access the non-existing codec slot. This often results in the total screw-up, and destructs the further communication with the codec chips. The symptom appears usually as error messages like: ------------------------------------------------------------------------ hda_intel: azx_get_response timeout, switching to polling mode: last cmd=0x12345678 hda_intel: azx_get_response timeout, switching to single_cmd mode: last cmd=0x12345678 ------------------------------------------------------------------------ The first line is a warning, and this is usually relatively harmless. It means that the codec response isn't notified via an IRQ. The driver uses explicit polling method to read the response. It gives very slight CPU overhead, but you'd unlikely notice it. The second line is, however, a fatal error. If this happens, usually it means that something is really wrong. Most likely you are accessing a non-existing codec slot. Thus, if the second error message appears, try to narrow the probed codec slots via `probe_mask` option. It's a bitmask, and each bit corresponds to the codec slot. For example, to probe only the first slot, pass `probe_mask=1`. For the first and the third slots, pass `probe_mask=5` (where 5 = 1 | 4), and so on. Since 2.6.29 kernel, the driver has a more robust probing method, so this error might happen rarely, though. Interrupt Handling ~~~~~~~~~~~~~~~~~~ In rare but some cases, the interrupt isn't properly handled as default. You would notice this by the DMA transfer error reported by ALSA PCM core, for example. Using MSI might help in such a case. Pass `enable_msi=1` option for enabling MSI. HD-AUDIO CODEC -------------- Model Option ~~~~~~~~~~~~ The most common problem regarding the HD-audio driver is the unsupported codec features or the mismatched device configuration. Most of codec-specific code has several preset models, either to override the BIOS setup or to provide more comprehensive features. The driver checks PCI SSID and looks through the static configuration table until any matching entry is found. If you have a new machine, you may see a message like below: ------------------------------------------------------------------------ hda_codec: Unknown model for ALC880, trying auto-probe from BIOS... ------------------------------------------------------------------------ Even if you see such a message, DON'T PANIC. Take a deep breath and keep your towel. First of all, it's an informational message, no warning, no error. This means that the PCI SSID of your device isn't listed in the known preset model (white-)list. But, this doesn't mean that the driver is broken. Many codec-drivers provide the automatic configuration mechanism based on the BIOS setup. The HD-audio codec has usually "pin" widgets, and BIOS sets the default configuration of each pin, which indicates the location, the connection type, the jack color, etc. The HD-audio driver can guess the right connection judging from these default configuration values. However -- some codec-support codes, such as patch_analog.c, don't support the automatic probing (yet as of 2.6.28). And, BIOS is often, yes, pretty often broken. It sets up wrong values and screws up the driver. The preset model is provided basically to overcome such a situation. When the matching preset model is found in the white-list, the driver assumes the static configuration of that preset and builds the mixer elements and PCM streams based on the static information. Thus, if you have a newer machine with a slightly different PCI SSID from the existing one, you may have a good chance to re-use the same model. You can pass the `model` option to specify the preset model instead of PCI SSID look-up. What `model` option values are available depends on the codec chip. Check your codec chip from the codec proc file (see "Codec Proc-File" section below). It will show the vendor/product name of your codec chip. Then, see Documentation/sound/alsa/HD-Audio-Modelstxt file, the section of HD-audio driver. You can find a list of codecs and `model` options belonging to each codec. For example, for Realtek ALC262 codec chip, pass `model=ultra` for devices that are compatible with Samsung Q1 Ultra. Thus, the first thing you can do for any brand-new, unsupported and non-working HD-audio hardware is to check HD-audio codec and several different `model` option values. If you have a luck, some of them might suit with your device well. Some codecs such as ALC880 have a special model option `model=test`. This configures the driver to provide as many mixer controls as possible for every single pin feature except for the unsolicited events (and maybe some other specials). Adjust each mixer element and try the I/O in the way of trial-and-error until figuring out the whole I/O pin mappings. Note that `model=generic` has a special meaning. It means to use the generic parser regardless of the codec. Usually the codec-specific parser is much better than the generic parser (as now). Thus this option is more about the debugging purpose. Speaker and Headphone Output ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ One of the most frequent (and obvious) bugs with HD-audio is the silent output from either or both of a built-in speaker and a headphone jack. In general, you should try a headphone output at first. A speaker output often requires more additional controls like the external amplifier bits. Thus a headphone output has a slightly better chance. Before making a bug report, double-check whether the mixer is set up correctly. The recent version of snd-hda-intel driver provides mostly "Master" volume control as well as "Front" volume (where Front indicates the front-channels). In addition, there can be individual "Headphone" and "Speaker" controls. Ditto for the speaker output. There can be "External Amplifier" switch on some codecs. Turn on this if present. Another related problem is the automatic mute of speaker output by headphone plugging. This feature is implemented in most cases, but not on every preset model or codec-support code. In anyway, try a different model option if you have such a problem. Some other models may match better and give you more matching functionality. If none of the available models works, send a bug report. See the bug report section for details. If you are masochistic enough to debug the driver problem, note the following: - The speaker (and the headphone, too) output often requires the external amplifier. This can be set usually via EAPD verb or a certain GPIO. If the codec pin supports EAPD, you have a better chance via SET_EAPD_BTL verb (0x70c). On others, GPIO pin (mostly it's either GPIO0 or GPIO1) may turn on/off EAPD. - Some Realtek codecs require special vendor-specific coefficients to turn on the amplifier. See patch_realtek.c. - IDT codecs may have extra power-enable/disable controls on each analog pin. See patch_sigmatel.c. - Very rare but some devices don't accept the pin-detection verb until triggered. Issuing GET_PIN_SENSE verb (0xf09) may result in the codec-communication stall. Some examples are found in patch_realtek.c. Capture Problems ~~~~~~~~~~~~~~~~ The capture problems are often because of missing setups of mixers. Thus, before submitting a bug report, make sure that you set up the mixer correctly. For example, both "Capture Volume" and "Capture Switch" have to be set properly in addition to the right "Capture Source" or "Input Source" selection. Some devices have "Mic Boost" volume or switch. When the PCM device is opened via "default" PCM (without pulse-audio plugin), you'll likely have "Digital Capture Volume" control as well. This is provided for the extra gain/attenuation of the signal in software, especially for the inputs without the hardware volume control such as digital microphones. Unless really needed, this should be set to exactly 50%, corresponding to 0dB -- neither extra gain nor attenuation. When you use "hw" PCM, i.e., a raw access PCM, this control will have no influence, though. It's known that some codecs / devices have fairly bad analog circuits, and the recorded sound contains a certain DC-offset. This is no bug of the driver. Most of modern laptops have no analog CD-input connection. Thus, the recording from CD input won't work in many cases although the driver provides it as the capture source. Use CDDA instead. The automatic switching of the built-in and external mic per plugging is implemented on some codec models but not on every model. Partly because of my laziness but mostly lack of testers. Feel free to submit the improvement patch to the author. Direct Debugging ~~~~~~~~~~~~~~~~ If no model option gives you a better result, and you are a tough guy to fight against evil, try debugging via hitting the raw HD-audio codec verbs to the device. Some tools are available: hda-emu and hda-analyzer. The detailed description is found in the sections below. You'd need to enable hwdep for using these tools. See "Kernel Configuration" section. OTHER ISSUES ------------ Kernel Configuration ~~~~~~~~~~~~~~~~~~~~ In general, I recommend you to enable the sound debug option, `CONFIG_SND_DEBUG=y`, no matter whether you are debugging or not. This enables snd_printd() macro and others, and you'll get additional kernel messages at probing. In addition, you can enable `CONFIG_SND_DEBUG_VERBOSE=y`. But this will give you far more messages. Thus turn this on only when you are sure to want it. Don't forget to turn on the appropriate `CONFIG_SND_HDA_CODEC_*` options. Note that each of them corresponds to the codec chip, not the controller chip. Thus, even if lspci shows the Nvidia controller, you may need to choose the option for other vendors. If you are unsure, just select all yes. `CONFIG_SND_HDA_HWDEP` is a useful option for debugging the driver. When this is enabled, the driver creates hardware-dependent devices (one per each codec), and you have a raw access to the device via these device files. For example, `hwC0D2` will be created for the codec slot #2 of the first card (#0). For debug-tools such as hda-verb and hda-analyzer, the hwdep device has to be enabled. Thus, it'd be better to turn this on always. `CONFIG_SND_HDA_RECONFIG` is a new option, and this depends on the hwdep option above. When enabled, you'll have some sysfs files under the corresponding hwdep directory. See "HD-audio reconfiguration" section below. `CONFIG_SND_HDA_POWER_SAVE` option enables the power-saving feature. See "Power-saving" section below. Codec Proc-File ~~~~~~~~~~~~~~~ The codec proc-file is a treasure-chest for debugging HD-audio. It shows most of useful information of each codec widget. The proc file is located in /proc/asound/card*/codec#*, one file per each codec slot. You can know the codec vendor, product id and names, the type of each widget, capabilities and so on. This file, however, doesn't show the jack sensing state, so far. This is because the jack-sensing might be depending on the trigger state. This file will be picked up by the debug tools, and also it can be fed to the emulator as the primary codec information. See the debug tools section below. This proc file can be also used to check whether the generic parser is used. When the generic parser is used, the vendor/product ID name will appear as "Realtek ID 0262", instead of "Realtek ALC262". HD-Audio Reconfiguration ~~~~~~~~~~~~~~~~~~~~~~~~ This is an experimental feature to allow you re-configure the HD-audio codec dynamically without reloading the driver. The following sysfs files are available under each codec-hwdep device directory (e.g. /sys/class/sound/hwC0D0): vendor_id:: Shows the 32bit codec vendor-id hex number. You can change the vendor-id value by writing to this file. subsystem_id:: Shows the 32bit codec subsystem-id hex number. You can change the subsystem-id value by writing to this file. revision_id:: Shows the 32bit codec revision-id hex number. You can change the revision-id value by writing to this file. afg:: Shows the AFG ID. This is read-only. mfg:: Shows the MFG ID. This is read-only. name:: Shows the codec name string. Can be changed by writing to this file. modelname:: Shows the currently set `model` option. Can be changed by writing to this file. init_verbs:: The extra verbs to execute at initialization. You can add a verb by writing to this file. Pass tree numbers, nid, verb and parameter. hints:: Shows hint strings for codec parsers for any use. Right now it's not used. reconfig:: Triggers the codec re-configuration. When any value is written to this file, the driver re-initialize and parses the codec tree again. All the changes done by the sysfs entries above are taken into account. clear:: Resets the codec, removes the mixer elements and PCM stuff of the specified codec, and clear all init verbs and hints. Power-Saving ~~~~~~~~~~~~ The power-saving is a kind of auto-suspend of the device. When the device is inactive for a certain time, the device is automatically turned off to save the power. The time to go down is specified via `power_save` module option, and this option can be changed dynamically via sysfs. The power-saving won't work when the analog loopback is enabled on some codecs. Make sure that you mute all unneeded signal routes when you want the power-saving. The power-saving feature might cause audible click noises at each power-down/up depending on the device. Some of them might be solvable, but some are hard, I'm afraid. Some distros such as openSUSE enables the power-saving feature automatically when the power cable is unplugged. Thus, if you hear noises, suspect first the power-saving. See /sys/module/snd_hda_intel/parameters/power_save to check the current value. If it's non-zero, the feature is turned on. Development Tree ~~~~~~~~~~~~~~~~ The latest development codes for HD-audio are found on sound git tree: - git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6.git The master branch or for-next branches can be used as the main development branches in general while the HD-audio specific patches are committed in topic/hda branch. If you are using the latest Linus tree, it'd be better to pull the above GIT tree onto it. If you are using the older kernels, an easy way to try the latest ALSA code is to build from the snapshot tarball. There are daily tarballs and the latest snapshot tarball. All can be built just like normal alsa-driver release packages, that is, installed via the usual spells: configure, make and make install(-modules). See INSTALL in the package. The snapshot tarballs are found at: - ftp://ftp.kernel.org/pub/linux/kernel/people/tiwai/snapshot/ Sending a Bug Report ~~~~~~~~~~~~~~~~~~~~ If any model or module options don't work for your device, it's time to send a bug report to the developers. Give the following in your bug report: - Hardware vendor, product and model names - Kernel version (and ALSA-driver version if you built externally) - `alsa-info.sh` output; run with `--no-upload` option. See the section below about alsa-info If it's a regression, at best, send alsa-info outputs of both working and non-working kernels. This is really helpful because we can compare the codec registers directly. Send a bug report either the followings: kernel-bugzilla:: http://bugme.linux-foundation.org/ alsa-devel ML:: alsa-devel@alsa-project.org DEBUG TOOLS ----------- This section describes some tools available for debugging HD-audio problems. alsa-info ~~~~~~~~~ The script `alsa-info.sh` is a very useful tool to gather the audio device information. You can fetch the latest version from: - http://www.alsa-project.org/alsa-info.sh Run this script as root, and it will gather the important information such as the module lists, module parameters, proc file contents including the codec proc files, mixer outputs and the control elements. As default, it will store the information onto a web server on alsa-project.org. But, if you send a bug report, it'd be better to run with `--no-upload` option, and attach the generated file. There are some other useful options. See `--help` option output for details. hda-verb ~~~~~~~~ hda-verb is a tiny program that allows you to access the HD-audio codec directly. You can execute a raw HD-audio codec verb with this. This program accesses the hwdep device, thus you need to enable the kernel config `CONFIG_SND_HDA_HWDEP=y` beforehand. The hda-verb program takes four arguments: the hwdep device file, the widget NID, the verb and the parameter. When you access to the codec on the slot 2 of the card 0, pass /dev/snd/hwC0D2 to the first argument, typically. (However, the real path name depends on the system.) The second parameter is the widget number-id to access. The third parameter can be either a hex/digit number or a string corresponding to a verb. Similarly, the last parameter is the value to write, or can be a string for the parameter type. ------------------------------------------------------------------------ % hda-verb /dev/snd/hwC0D0 0x12 0x701 2 nid = 0x12, verb = 0x701, param = 0x2 value = 0x0 % hda-verb /dev/snd/hwC0D0 0x0 PARAMETERS VENDOR_ID nid = 0x0, verb = 0xf00, param = 0x0 value = 0x10ec0262 % hda-verb /dev/snd/hwC0D0 2 set_a 0xb080 nid = 0x2, verb = 0x300, param = 0xb080 value = 0x0 ------------------------------------------------------------------------ Although you can issue any verbs with this program, the driver state won't be always updated. For example, the volume values are usually cached in the driver, and thus changing the widget amp value directly via hda-verb won't change the mixer value. The hda-verb program is found in the ftp directory: - ftp://ftp.kernel.org/pub/linux/kernel/people/tiwai/misc/ Also a git repository is available: - git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/hda-verb.git See README file in the tarball for more details about hda-verb program. hda-analyzer ~~~~~~~~~~~~ hda-analyzer provides a graphical interface to access the raw HD-audio control, based on pyGTK2 binding. It's a more powerful version of hda-verb. The program gives you an easy-to-use GUI stuff for showing the widget information and adjusting the amp values, as well as the proc-compatible output. The hda-analyzer is a part of alsa.git repository in alsa-project.org: - http://git.alsa-project.org/?p=alsa.git;a=tree;f=hda-analyzer Codecgraph ~~~~~~~~~~ Codecgraph is a utility program to generate a graph and visualizes the codec-node connection of a codec chip. It's especially useful when you analyze or debug a codec without a proper datasheet. The program parses the given codec proc file and converts to SVG via graphiz program. The tarball and GIT trees are found in the web page at: - http://helllabs.org/codecgraph/ hda-emu ~~~~~~~ hda-emu is an HD-audio emulator. The main purpose of this program is to debug an HD-audio codec without the real hardware. Thus, it doesn't emulate the behavior with the real audio I/O, but it just dumps the codec register changes and the ALSA-driver internal changes at probing and operating the HD-audio driver. The program requires a codec proc-file to simulate. Get a proc file for the target codec beforehand, or pick up an example codec from the codec proc collections in the tarball. Then, run the program with the proc file, and the hda-emu program will start parsing the codec file and simulates the HD-audio driver: ------------------------------------------------------------------------ % hda-emu codecs/stac9200-dell-d820-laptop # Parsing.. hda_codec: Unknown model for STAC9200, using BIOS defaults hda_codec: pin nid 08 bios pin config 40c003fa .... ------------------------------------------------------------------------ The program gives you only a very dumb command-line interface. You can get a proc-file dump at the current state, get a list of control (mixer) elements, set/get the control element value, simulate the PCM operation, the jack plugging simulation, etc. The package is found in: - ftp://ftp.kernel.org/pub/linux/kernel/people/tiwai/misc/ A git repository is available: - git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/hda-emu.git See README file in the tarball for more details about hda-emu program.