Writing an ALSA Driver

The ALSA drivers are provided in two ways.

One is the trees provided as a tarball or via cvs from the ALSA's ftp site, and another is the 2.6 (or later) Linux kernel tree. To synchronize both, the ALSA driver tree is split into two different trees: alsa-kernel and alsa-driver. The former contains purely the source code for the Linux 2.6 (or later) tree. This tree is designed only for compilation on 2.6 or later environment. The latter, alsa-driver, contains many subtle files for compiling ALSA drivers outside of the Linux kernel tree, wrapper functions for older 2.2 and 2.4 kernels, to adapt the latest kernel API, and additional drivers which are still in development or in tests. The drivers in alsa-driver tree will be moved to alsa-kernel (and eventually to the 2.6 kernel tree) when they are finished and confirmed to work fine.

The file tree structure of ALSA driver is depicted below. Both alsa-kernel and alsa-driver have almost the same file structure, except for “core” directory. It's named as “acore” in alsa-driver tree.

This directory contains the middle layer which is the heart of ALSA drivers. In this directory, the native ALSA modules are stored. The sub-directories contain different modules and are dependent upon the kernel config.

This is the place for the public header files of ALSA drivers, which are to be exported to user-space, or included by several files at different directories. Basically, the private header files should not be placed in this directory, but you may still find files there, due to historical reasons :)

This directory contains code shared among different drivers on different architectures. They are hence supposed not to be architecture-specific. For example, the dummy pcm driver and the serial MIDI driver are found in this directory. In the sub-directories, there is code for components which are independent from bus and cpu architectures.

This contains the ALSA i2c components.

Although there is a standard i2c layer on Linux, ALSA has its own i2c code for some cards, because the soundcard needs only a simple operation and the standard i2c API is too complicated for such a purpose.

This contains the synth middle-level modules.

So far, there is only Emu8000/Emu10k1 synth driver under the synth/emux sub-directory.

This directory and its sub-directories hold the top-level card modules for PCI soundcards and the code specific to the PCI BUS.

The drivers compiled from a single file are stored directly in the pci directory, while the drivers with several source files are stored on their own sub-directory (e.g. emu10k1, ice1712).

This directory and its sub-directories hold the top-level card modules for ISA soundcards.

They are used for top-level card modules which are specific to one of these architectures.

This directory contains the USB-audio driver. In the latest version, the USB MIDI driver is integrated in the usb-audio driver.

The PCMCIA, especially PCCard drivers will go here. CardBus drivers will be in the pci directory, because their API is identical to that of standard PCI cards.

The OSS/Lite source files are stored here in Linux 2.6 (or later) tree. In the ALSA driver tarball, this directory is empty, of course :)

The minimum flow for PCI soundcards is as follows:

The code example is shown below. Some parts are kept unimplemented at this moment but will be filled in the next sections. The numbers in the comment lines of the snd_mychip_probe() function refer to details explained in the following section.

  #include <linux/init.h>
  #include <linux/pci.h>
  #include <linux/slab.h>
  #include <sound/core.h>
  #include <sound/initval.h>

  /* module parameters (see "Module Parameters") */
  /* SNDRV_CARDS: maximum number of cards supported by this module */
  static int index[SNDRV_CARDS] = SNDRV_DEFAULT_IDX;
  static char *id[SNDRV_CARDS] = SNDRV_DEFAULT_STR;
  static int enable[SNDRV_CARDS] = SNDRV_DEFAULT_ENABLE_PNP;

  /* definition of the chip-specific record */
  struct mychip {
          struct snd_card *card;
          /* the rest of the implementation will be in section
           * "PCI Resource Management"
           */
  };

  /* chip-specific destructor
   * (see "PCI Resource Management")
   */
  static int snd_mychip_free(struct mychip *chip)
  {
          .... /* will be implemented later... */
  }

  /* component-destructor
   * (see "Management of Cards and Components")
   */
  static int snd_mychip_dev_free(struct snd_device *device)
  {
          return snd_mychip_free(device->device_data);
  }

  /* chip-specific constructor
   * (see "Management of Cards and Components")
   */
  static int __devinit snd_mychip_create(struct snd_card *card,
                                         struct pci_dev *pci,
                                         struct mychip **rchip)
  {
          struct mychip *chip;
          int err;
          static struct snd_device_ops ops = {
                 .dev_free = snd_mychip_dev_free,
          };

          *rchip = NULL;

          /* check PCI availability here
           * (see "PCI Resource Management")
           */
          ....

          /* allocate a chip-specific data with zero filled */
          chip = kzalloc(sizeof(*chip), GFP_KERNEL);
          if (chip == NULL)
                  return -ENOMEM;

          chip->card = card;

          /* rest of initialization here; will be implemented
           * later, see "PCI Resource Management"
           */
          ....

          err = snd_device_new(card, SNDRV_DEV_LOWLEVEL, chip, &ops);
          if (err < 0) {
                  snd_mychip_free(chip);
                  return err;
          }

          snd_card_set_dev(card, &pci->dev);

          *rchip = chip;
          return 0;
  }

  /* constructor -- see "Constructor" sub-section */
  static int __devinit snd_mychip_probe(struct pci_dev *pci,
                               const struct pci_device_id *pci_id)
  {
          static int dev;
          struct snd_card *card;
          struct mychip *chip;
          int err;

          /* (1) */
          if (dev >= SNDRV_CARDS)
                  return -ENODEV;
          if (!enable[dev]) {
                  dev++;
                  return -ENOENT;
          }

          /* (2) */
          err = snd_card_create(index[dev], id[dev], THIS_MODULE, 0, &card);
          if (err < 0)
                  return err;

          /* (3) */
          err = snd_mychip_create(card, pci, &chip);
          if (err < 0) {
                  snd_card_free(card);
                  return err;
          }

          /* (4) */
          strcpy(card->driver, "My Chip");
          strcpy(card->shortname, "My Own Chip 123");
          sprintf(card->longname, "%s at 0x%lx irq %i",
                  card->shortname, chip->ioport, chip->irq);

          /* (5) */
          .... /* implemented later */

          /* (6) */
          err = snd_card_register(card);
          if (err < 0) {
                  snd_card_free(card);
                  return err;
          }

          /* (7) */
          pci_set_drvdata(pci, card);
          dev++;
          return 0;
  }

  /* destructor -- see the "Destructor" sub-section */
  static void __devexit snd_mychip_remove(struct pci_dev *pci)
  {
          snd_card_free(pci_get_drvdata(pci));
          pci_set_drvdata(pci, NULL);
  }

          

The real constructor of PCI drivers is the probe callback. The probe callback and other component-constructors which are called from the probe callback should be defined with the __devinit prefix. You cannot use the __init prefix for them, because any PCI device could be a hotplug device.

In the probe callback, the following scheme is often used.

  static int dev;
  ....
  if (dev >= SNDRV_CARDS)
          return -ENODEV;
  if (!enable[dev]) {
          dev++;
          return -ENOENT;
  }

            

  struct snd_card *card;
  int err;
  ....
  err = snd_card_create(index[dev], id[dev], THIS_MODULE, 0, &card);

            

  struct mychip *chip;
  ....
  err = snd_mychip_create(card, pci, &chip);
  if (err < 0) {
          snd_card_free(card);
          return err;
  }

            

  strcpy(card->driver, "My Chip");
  strcpy(card->shortname, "My Own Chip 123");
  sprintf(card->longname, "%s at 0x%lx irq %i",
          card->shortname, chip->ioport, chip->irq);

            

  err = snd_card_register(card);
  if (err < 0) {
          snd_card_free(card);
          return err;
  }

            

        pci_set_drvdata(pci, card);
        dev++;
        return 0;

            

The destructor, remove callback, simply releases the card instance. Then the ALSA middle layer will release all the attached components automatically.

It would be typically like the following:

  static void __devexit snd_mychip_remove(struct pci_dev *pci)
  {
          snd_card_free(pci_get_drvdata(pci));
          pci_set_drvdata(pci, NULL);
  }

          

The above code assumes that the card pointer is set to the PCI driver data.

For the above example, at least the following include files are necessary.

  #include <linux/init.h>
  #include <linux/pci.h>
  #include <linux/slab.h>
  #include <sound/core.h>
  #include <sound/initval.h>

          

where the last one is necessary only when module options are defined in the source file. If the code is split into several files, the files without module options don't need them.

In addition to these headers, you'll need <linux/interrupt.h> for interrupt handling, and <asm/io.h> for I/O access. If you use the mdelay() or udelay() functions, you'll need to include <linux/delay.h> too.

The ALSA interfaces like the PCM and control APIs are defined in other <sound/xxx.h> header files. They have to be included after <sound/core.h>.

For each soundcard, a “card” record must be allocated.

A card record is the headquarters of the soundcard. It manages the whole list of devices (components) on the soundcard, such as PCM, mixers, MIDI, synthesizer, and so on. Also, the card record holds the ID and the name strings of the card, manages the root of proc files, and controls the power-management states and hotplug disconnections. The component list on the card record is used to manage the correct release of resources at destruction.

As mentioned above, to create a card instance, call snd_card_create().

  struct snd_card *card;
  int err;
  err = snd_card_create(index, id, module, extra_size, &card);

          

The function takes five arguments, the card-index number, the id string, the module pointer (usually THIS_MODULE), the size of extra-data space, and the pointer to return the card instance. The extra_size argument is used to allocate card->private_data for the chip-specific data. Note that these data are allocated by snd_card_create().

After the card is created, you can attach the components (devices) to the card instance. In an ALSA driver, a component is represented as a struct snd_device object. A component can be a PCM instance, a control interface, a raw MIDI interface, etc. Each such instance has one component entry.

A component can be created via snd_device_new() function.

  snd_device_new(card, SNDRV_DEV_XXX, chip, &ops);

          

This takes the card pointer, the device-level (SNDRV_DEV_XXX), the data pointer, and the callback pointers (&ops). The device-level defines the type of components and the order of registration and de-registration. For most components, the device-level is already defined. For a user-defined component, you can use SNDRV_DEV_LOWLEVEL.

This function itself doesn't allocate the data space. The data must be allocated manually beforehand, and its pointer is passed as the argument. This pointer is used as the (chip identifier in the above example) for the instance.

Each pre-defined ALSA component such as ac97 and pcm calls snd_device_new() inside its constructor. The destructor for each component is defined in the callback pointers. Hence, you don't need to take care of calling a destructor for such a component.

If you wish to create your own component, you need to set the destructor function to the dev_free callback in the ops, so that it can be released automatically via snd_card_free(). The next example will show an implementation of chip-specific data.

Chip-specific information, e.g. the I/O port address, its resource pointer, or the irq number, is stored in the chip-specific record.

  struct mychip {
          ....
  };

          

In general, there are two ways of allocating the chip record.

  err = snd_card_create(index[dev], id[dev], THIS_MODULE,
                        sizeof(struct mychip), &card);

            

  struct mychip *chip = card->private_data;

            

  struct snd_card *card;
  struct mychip *chip;
  err = snd_card_create(index[dev], id[dev], THIS_MODULE, 0, &card);
  .....
  chip = kzalloc(sizeof(*chip), GFP_KERNEL);

            

  struct mychip {
          struct snd_card *card;
          ....
  };

            

  chip->card = card;

            

  static struct snd_device_ops ops = {
          .dev_free =        snd_mychip_dev_free,
  };
  ....
  snd_device_new(card, SNDRV_DEV_LOWLEVEL, chip, &ops);

            

  static int snd_mychip_dev_free(struct snd_device *device)
  {
          return snd_mychip_free(device->device_data);
  }

            

After all components are assigned, register the card instance by calling snd_card_register(). Access to the device files is enabled at this point. That is, before snd_card_register() is called, the components are safely inaccessible from external side. If this call fails, exit the probe function after releasing the card via snd_card_free().

For releasing the card instance, you can call simply snd_card_free(). As mentioned earlier, all components are released automatically by this call.

As further notes, the destructors (both snd_mychip_dev_free and snd_mychip_free) cannot be defined with the __devexit prefix, because they may be called from the constructor, too, at the false path.

For a device which allows hotplugging, you can use snd_card_free_when_closed. This one will postpone the destruction until all devices are closed.

In this section, we'll complete the chip-specific constructor, destructor and PCI entries. Example code is shown first, below.

  struct mychip {
          struct snd_card *card;
          struct pci_dev *pci;

          unsigned long port;
          int irq;
  };

  static int snd_mychip_free(struct mychip *chip)
  {
          /* disable hardware here if any */
          .... /* (not implemented in this document) */

          /* release the irq */
          if (chip->irq >= 0)
                  free_irq(chip->irq, chip);
          /* release the I/O ports & memory */
          pci_release_regions(chip->pci);
          /* disable the PCI entry */
          pci_disable_device(chip->pci);
          /* release the data */
          kfree(chip);
          return 0;
  }

  /* chip-specific constructor */
  static int __devinit snd_mychip_create(struct snd_card *card,
                                         struct pci_dev *pci,
                                         struct mychip **rchip)
  {
          struct mychip *chip;
          int err;
          static struct snd_device_ops ops = {
                 .dev_free = snd_mychip_dev_free,
          };

          *rchip = NULL;

          /* initialize the PCI entry */
          err = pci_enable_device(pci);
          if (err < 0)
                  return err;
          /* check PCI availability (28bit DMA) */
          if (pci_set_dma_mask(pci, DMA_BIT_MASK(28)) < 0 ||
              pci_set_consistent_dma_mask(pci, DMA_BIT_MASK(28)) < 0) {
                  printk(KERN_ERR "error to set 28bit mask DMA\n");
                  pci_disable_device(pci);
                  return -ENXIO;
          }

          chip = kzalloc(sizeof(*chip), GFP_KERNEL);
          if (chip == NULL) {
                  pci_disable_device(pci);
                  return -ENOMEM;
          }

          /* initialize the stuff */
          chip->card = card;
          chip->pci = pci;
          chip->irq = -1;

          /* (1) PCI resource allocation */
          err = pci_request_regions(pci, "My Chip");
          if (err < 0) {
                  kfree(chip);
                  pci_disable_device(pci);
                  return err;
          }
          chip->port = pci_resource_start(pci, 0);
          if (request_irq(pci->irq, snd_mychip_interrupt,
                          IRQF_SHARED, "My Chip", chip)) {
                  printk(KERN_ERR "cannot grab irq %d\n", pci->irq);
                  snd_mychip_free(chip);
                  return -EBUSY;
          }
          chip->irq = pci->irq;

          /* (2) initialization of the chip hardware */
          .... /*   (not implemented in this document) */

          err = snd_device_new(card, SNDRV_DEV_LOWLEVEL, chip, &ops);
          if (err < 0) {
                  snd_mychip_free(chip);
                  return err;
          }

          snd_card_set_dev(card, &pci->dev);

          *rchip = chip;
          return 0;
  }        

  /* PCI IDs */
  static struct pci_device_id snd_mychip_ids[] = {
          { PCI_VENDOR_ID_FOO, PCI_DEVICE_ID_BAR,
            PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0, },
          ....
          { 0, }
  };
  MODULE_DEVICE_TABLE(pci, snd_mychip_ids);

  /* pci_driver definition */
  static struct pci_driver driver = {
          .name = "My Own Chip",
          .id_table = snd_mychip_ids,
          .probe = snd_mychip_probe,
          .remove = __devexit_p(snd_mychip_remove),
  };

  /* module initialization */
  static int __init alsa_card_mychip_init(void)
  {
          return pci_register_driver(&driver);
  }

  /* module clean up */
  static void __exit alsa_card_mychip_exit(void)
  {
          pci_unregister_driver(&driver);
  }

  module_init(alsa_card_mychip_init)
  module_exit(alsa_card_mychip_exit)

  EXPORT_NO_SYMBOLS; /* for old kernels only */

          

The allocation of PCI resources is done in the probe() function, and usually an extra xxx_create() function is written for this purpose.

In the case of PCI devices, you first have to call the pci_enable_device() function before allocating resources. Also, you need to set the proper PCI DMA mask to limit the accessed I/O range. In some cases, you might need to call pci_set_master() function, too.

Suppose the 28bit mask, and the code to be added would be like:

  err = pci_enable_device(pci);
  if (err < 0)
          return err;
  if (pci_set_dma_mask(pci, DMA_BIT_MASK(28)) < 0 ||
      pci_set_consistent_dma_mask(pci, DMA_BIT_MASK(28)) < 0) {
          printk(KERN_ERR "error to set 28bit mask DMA\n");
          pci_disable_device(pci);
          return -ENXIO;
  }
  

          

The allocation of I/O ports and irqs is done via standard kernel functions. Unlike ALSA ver.0.5.x., there are no helpers for that. And these resources must be released in the destructor function (see below). Also, on ALSA 0.9.x, you don't need to allocate (pseudo-)DMA for PCI like in ALSA 0.5.x.

Now assume that the PCI device has an I/O port with 8 bytes and an interrupt. Then struct mychip will have the following fields:

  struct mychip {
          struct snd_card *card;

          unsigned long port;
          int irq;
  };

          

For an I/O port (and also a memory region), you need to have the resource pointer for the standard resource management. For an irq, you have to keep only the irq number (integer). But you need to initialize this number as -1 before actual allocation, since irq 0 is valid. The port address and its resource pointer can be initialized as null by kzalloc() automatically, so you don't have to take care of resetting them.

The allocation of an I/O port is done like this:

  err = pci_request_regions(pci, "My Chip");
  if (err < 0) { 
          kfree(chip);
          pci_disable_device(pci);
          return err;
  }
  chip->port = pci_resource_start(pci, 0);

          

It will reserve the I/O port region of 8 bytes of the given PCI device. The returned value, chip->res_port, is allocated via kmalloc() by request_region(). The pointer must be released via kfree(), but there is a problem with this. This issue will be explained later.

The allocation of an interrupt source is done like this:

  if (request_irq(pci->irq, snd_mychip_interrupt,
                  IRQF_SHARED, "My Chip", chip)) {
          printk(KERN_ERR "cannot grab irq %d\n", pci->irq);
          snd_mychip_free(chip);
          return -EBUSY;
  }
  chip->irq = pci->irq;

          

where snd_mychip_interrupt() is the interrupt handler defined later. Note that chip->irq should be defined only when request_irq() succeeded.

On the PCI bus, interrupts can be shared. Thus, IRQF_SHARED is used as the interrupt flag of request_irq().

The last argument of request_irq() is the data pointer passed to the interrupt handler. Usually, the chip-specific record is used for that, but you can use what you like, too.

I won't give details about the interrupt handler at this point, but at least its appearance can be explained now. The interrupt handler looks usually like the following:

  static irqreturn_t snd_mychip_interrupt(int irq, void *dev_id)
  {
          struct mychip *chip = dev_id;
          ....
          return IRQ_HANDLED;
  }

          

Now let's write the corresponding destructor for the resources above. The role of destructor is simple: disable the hardware (if already activated) and release the resources. So far, we have no hardware part, so the disabling code is not written here.

To release the resources, the “check-and-release” method is a safer way. For the interrupt, do like this:

  if (chip->irq >= 0)
          free_irq(chip->irq, chip);

          

Since the irq number can start from 0, you should initialize chip->irq with a negative value (e.g. -1), so that you can check the validity of the irq number as above.

When you requested I/O ports or memory regions via pci_request_region() or pci_request_regions() like in this example, release the resource(s) using the corresponding function, pci_release_region() or pci_release_regions().

  pci_release_regions(chip->pci);

          

When you requested manually via request_region() or request_mem_region, you can release it via release_resource(). Suppose that you keep the resource pointer returned from request_region() in chip->res_port, the release procedure looks like:

  release_and_free_resource(chip->res_port);

          

Don't forget to call pci_disable_device() before the end.

And finally, release the chip-specific record.

  kfree(chip);

          

Again, remember that you cannot use the __devexit prefix for this destructor.

We didn't implement the hardware disabling part in the above. If you need to do this, please note that the destructor may be called even before the initialization of the chip is completed. It would be better to have a flag to skip hardware disabling if the hardware was not initialized yet.

When the chip-data is assigned to the card using snd_device_new() with SNDRV_DEV_LOWLELVEL , its destructor is called at the last. That is, it is assured that all other components like PCMs and controls have already been released. You don't have to stop PCMs, etc. explicitly, but just call low-level hardware stopping.

The management of a memory-mapped region is almost as same as the management of an I/O port. You'll need three fields like the following:

  struct mychip {
          ....
          unsigned long iobase_phys;
          void __iomem *iobase_virt;
  };

          

and the allocation would be like below:

  if ((err = pci_request_regions(pci, "My Chip")) < 0) {
          kfree(chip);
          return err;
  }
  chip->iobase_phys = pci_resource_start(pci, 0);
  chip->iobase_virt = ioremap_nocache(chip->iobase_phys,
                                      pci_resource_len(pci, 0));

          

and the corresponding destructor would be:

  static int snd_mychip_free(struct mychip *chip)
  {
          ....
          if (chip->iobase_virt)
                  iounmap(chip->iobase_virt);
          ....
          pci_release_regions(chip->pci);
          ....
  }

          

At some point, typically after calling snd_device_new(), you need to register the struct device of the chip you're handling for udev and co. ALSA provides a macro for compatibility with older kernels. Simply call like the following:

  snd_card_set_dev(card, &pci->dev);

          

so that it stores the PCI's device pointer to the card. This will be referred by ALSA core functions later when the devices are registered.

In the case of non-PCI, pass the proper device struct pointer of the BUS instead. (In the case of legacy ISA without PnP, you don't have to do anything.)

So far, so good. Let's finish the missing PCI stuff. At first, we need a pci_device_id table for this chipset. It's a table of PCI vendor/device ID number, and some masks.

For example,

  static struct pci_device_id snd_mychip_ids[] = {
          { PCI_VENDOR_ID_FOO, PCI_DEVICE_ID_BAR,
            PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0, },
          ....
          { 0, }
  };
  MODULE_DEVICE_TABLE(pci, snd_mychip_ids);

          

The first and second fields of the pci_device_id structure are the vendor and device IDs. If you have no reason to filter the matching devices, you can leave the remaining fields as above. The last field of the pci_device_id struct contains private data for this entry. You can specify any value here, for example, to define specific operations for supported device IDs. Such an example is found in the intel8x0 driver.

The last entry of this list is the terminator. You must specify this all-zero entry.

Then, prepare the pci_driver record:

  static struct pci_driver driver = {
          .name = "My Own Chip",
          .id_table = snd_mychip_ids,
          .probe = snd_mychip_probe,
          .remove = __devexit_p(snd_mychip_remove),
  };

          

The probe and remove functions have already been defined in the previous sections. The remove function should be defined with the __devexit_p() macro, so that it's not defined for built-in (and non-hot-pluggable) case. The name field is the name string of this device. Note that you must not use a slash “/” in this string.

And at last, the module entries:

  static int __init alsa_card_mychip_init(void)
  {
          return pci_register_driver(&driver);
  }

  static void __exit alsa_card_mychip_exit(void)
  {
          pci_unregister_driver(&driver);
  }

  module_init(alsa_card_mychip_init)
  module_exit(alsa_card_mychip_exit)

          

Note that these module entries are tagged with __init and __exit prefixes, not __devinit nor __devexit.

Oh, one thing was forgotten. If you have no exported symbols, you need to declare it in 2.2 or 2.4 kernels (it's not necessary in 2.6 kernels).

  EXPORT_NO_SYMBOLS;

          

That's all!

The PCM middle layer of ALSA is quite powerful and it is only necessary for each driver to implement the low-level functions to access its hardware.

For accessing to the PCM layer, you need to include <sound/pcm.h> first. In addition, <sound/pcm_params.h> might be needed if you access to some functions related with hw_param.

Each card device can have up to four pcm instances. A pcm instance corresponds to a pcm device file. The limitation of number of instances comes only from the available bit size of the Linux's device numbers. Once when 64bit device number is used, we'll have more pcm instances available.

A pcm instance consists of pcm playback and capture streams, and each pcm stream consists of one or more pcm substreams. Some soundcards support multiple playback functions. For example, emu10k1 has a PCM playback of 32 stereo substreams. In this case, at each open, a free substream is (usually) automatically chosen and opened. Meanwhile, when only one substream exists and it was already opened, the successful open will either block or error with EAGAIN according to the file open mode. But you don't have to care about such details in your driver. The PCM middle layer will take care of such work.

The example code below does not include any hardware access routines but shows only the skeleton, how to build up the PCM interfaces.

  #include <sound/pcm.h>
  ....

  /* hardware definition */
  static struct snd_pcm_hardware snd_mychip_playback_hw = {
          .info = (SNDRV_PCM_INFO_MMAP |
                   SNDRV_PCM_INFO_INTERLEAVED |
                   SNDRV_PCM_INFO_BLOCK_TRANSFER |
                   SNDRV_PCM_INFO_MMAP_VALID),
          .formats =          SNDRV_PCM_FMTBIT_S16_LE,
          .rates =            SNDRV_PCM_RATE_8000_48000,
          .rate_min =         8000,
          .rate_max =         48000,
          .channels_min =     2,
          .channels_max =     2,
          .buffer_bytes_max = 32768,
          .period_bytes_min = 4096,
          .period_bytes_max = 32768,
          .periods_min =      1,
          .periods_max =      1024,
  };

  /* hardware definition */
  static struct snd_pcm_hardware snd_mychip_capture_hw = {
          .info = (SNDRV_PCM_INFO_MMAP |
                   SNDRV_PCM_INFO_INTERLEAVED |
                   SNDRV_PCM_INFO_BLOCK_TRANSFER |
                   SNDRV_PCM_INFO_MMAP_VALID),
          .formats =          SNDRV_PCM_FMTBIT_S16_LE,
          .rates =            SNDRV_PCM_RATE_8000_48000,
          .rate_min =         8000,
          .rate_max =         48000,
          .channels_min =     2,
          .channels_max =     2,
          .buffer_bytes_max = 32768,
          .period_bytes_min = 4096,
          .period_bytes_max = 32768,
          .periods_min =      1,
          .periods_max =      1024,
  };

  /* open callback */
  static int snd_mychip_playback_open(struct snd_pcm_substream *substream)
  {
          struct mychip *chip = snd_pcm_substream_chip(substream);
          struct snd_pcm_runtime *runtime = substream->runtime;

          runtime->hw = snd_mychip_playback_hw;
          /* more hardware-initialization will be done here */
          ....
          return 0;
  }

  /* close callback */
  static int snd_mychip_playback_close(struct snd_pcm_substream *substream)
  {
          struct mychip *chip = snd_pcm_substream_chip(substream);
          /* the hardware-specific codes will be here */
          ....
          return 0;

  }

  /* open callback */
  static int snd_mychip_capture_open(struct snd_pcm_substream *substream)
  {
          struct mychip *chip = snd_pcm_substream_chip(substream);
          struct snd_pcm_runtime *runtime = substream->runtime;

          runtime->hw = snd_mychip_capture_hw;
          /* more hardware-initialization will be done here */
          ....
          return 0;
  }

  /* close callback */
  static int snd_mychip_capture_close(struct snd_pcm_substream *substream)
  {
          struct mychip *chip = snd_pcm_substream_chip(substream);
          /* the hardware-specific codes will be here */
          ....
          return 0;

  }

  /* hw_params callback */
  static int snd_mychip_pcm_hw_params(struct snd_pcm_substream *substream,
                               struct snd_pcm_hw_params *hw_params)
  {
          return snd_pcm_lib_malloc_pages(substream,
                                     params_buffer_bytes(hw_params));
  }

  /* hw_free callback */
  static int snd_mychip_pcm_hw_free(struct snd_pcm_substream *substream)
  {
          return snd_pcm_lib_free_pages(substream);
  }

  /* prepare callback */
  static int snd_mychip_pcm_prepare(struct snd_pcm_substream *substream)
  {
          struct mychip *chip = snd_pcm_substream_chip(substream);
          struct snd_pcm_runtime *runtime = substream->runtime;

          /* set up the hardware with the current configuration
           * for example...
           */
          mychip_set_sample_format(chip, runtime->format);
          mychip_set_sample_rate(chip, runtime->rate);
          mychip_set_channels(chip, runtime->channels);
          mychip_set_dma_setup(chip, runtime->dma_addr,
                               chip->buffer_size,
                               chip->period_size);
          return 0;
  }

  /* trigger callback */
  static int snd_mychip_pcm_trigger(struct snd_pcm_substream *substream,
                                    int cmd)
  {
          switch (cmd) {
          case SNDRV_PCM_TRIGGER_START:
                  /* do something to start the PCM engine */
                  ....
                  break;
          case SNDRV_PCM_TRIGGER_STOP:
                  /* do something to stop the PCM engine */
                  ....
                  break;
          default:
                  return -EINVAL;
          }
  }

  /* pointer callback */
  static snd_pcm_uframes_t
  snd_mychip_pcm_pointer(struct snd_pcm_substream *substream)
  {
          struct mychip *chip = snd_pcm_substream_chip(substream);
          unsigned int current_ptr;

          /* get the current hardware pointer */
          current_ptr = mychip_get_hw_pointer(chip);
          return current_ptr;
  }

  /* operators */
  static struct snd_pcm_ops snd_mychip_playback_ops = {
          .open =        snd_mychip_playback_open,
          .close =       snd_mychip_playback_close,
          .ioctl =       snd_pcm_lib_ioctl,
          .hw_params =   snd_mychip_pcm_hw_params,
          .hw_free =     snd_mychip_pcm_hw_free,
          .prepare =     snd_mychip_pcm_prepare,
          .trigger =     snd_mychip_pcm_trigger,
          .pointer =     snd_mychip_pcm_pointer,
  };

  /* operators */
  static struct snd_pcm_ops snd_mychip_capture_ops = {
          .open =        snd_mychip_capture_open,
          .close =       snd_mychip_capture_close,
          .ioctl =       snd_pcm_lib_ioctl,
          .hw_params =   snd_mychip_pcm_hw_params,
          .hw_free =     snd_mychip_pcm_hw_free,
          .prepare =     snd_mychip_pcm_prepare,
          .trigger =     snd_mychip_pcm_trigger,
          .pointer =     snd_mychip_pcm_pointer,
  };

  /*
   *  definitions of capture are omitted here...
   */

  /* create a pcm device */
  static int __devinit snd_mychip_new_pcm(struct mychip *chip)
  {
          struct snd_pcm *pcm;
          int err;

          err = snd_pcm_new(chip->card, "My Chip", 0, 1, 1, &pcm);
          if (err < 0) 
                  return err;
          pcm->private_data = chip;
          strcpy(pcm->name, "My Chip");
          chip->pcm = pcm;
          /* set operators */
          snd_pcm_set_ops(pcm, SNDRV_PCM_STREAM_PLAYBACK,
                          &snd_mychip_playback_ops);
          snd_pcm_set_ops(pcm, SNDRV_PCM_STREAM_CAPTURE,
                          &snd_mychip_capture_ops);
          /* pre-allocation of buffers */
          /* NOTE: this may fail */
          snd_pcm_lib_preallocate_pages_for_all(pcm, SNDRV_DMA_TYPE_DEV,
                                                snd_dma_pci_data(chip->pci),
                                                64*1024, 64*1024);
          return 0;
  }

          

A pcm instance is allocated by the snd_pcm_new() function. It would be better to create a constructor for pcm, namely,

  static int __devinit snd_mychip_new_pcm(struct mychip *chip)
  {
          struct snd_pcm *pcm;
          int err;

          err = snd_pcm_new(chip->card, "My Chip", 0, 1, 1, &pcm);
          if (err < 0) 
                  return err;
          pcm->private_data = chip;
          strcpy(pcm->name, "My Chip");
          chip->pcm = pcm;
	  ....
          return 0;
  }

          

The snd_pcm_new() function takes four arguments. The first argument is the card pointer to which this pcm is assigned, and the second is the ID string.

The third argument (index, 0 in the above) is the index of this new pcm. It begins from zero. If you create more than one pcm instances, specify the different numbers in this argument. For example, index = 1 for the second PCM device.

The fourth and fifth arguments are the number of substreams for playback and capture, respectively. Here 1 is used for both arguments. When no playback or capture substreams are available, pass 0 to the corresponding argument.

If a chip supports multiple playbacks or captures, you can specify more numbers, but they must be handled properly in open/close, etc. callbacks. When you need to know which substream you are referring to, then it can be obtained from struct snd_pcm_substream data passed to each callback as follows:

  struct snd_pcm_substream *substream;
  int index = substream->number;

          

After the pcm is created, you need to set operators for each pcm stream.

  snd_pcm_set_ops(pcm, SNDRV_PCM_STREAM_PLAYBACK,
                  &snd_mychip_playback_ops);
  snd_pcm_set_ops(pcm, SNDRV_PCM_STREAM_CAPTURE,
                  &snd_mychip_capture_ops);

          

The operators are defined typically like this:

  static struct snd_pcm_ops snd_mychip_playback_ops = {
          .open =        snd_mychip_pcm_open,
          .close =       snd_mychip_pcm_close,
          .ioctl =       snd_pcm_lib_ioctl,
          .hw_params =   snd_mychip_pcm_hw_params,
          .hw_free =     snd_mychip_pcm_hw_free,
          .prepare =     snd_mychip_pcm_prepare,
          .trigger =     snd_mychip_pcm_trigger,
          .pointer =     snd_mychip_pcm_pointer,
  };

          

All the callbacks are described in the Operators subsection.

After setting the operators, you probably will want to pre-allocate the buffer. For the pre-allocation, simply call the following:

  snd_pcm_lib_preallocate_pages_for_all(pcm, SNDRV_DMA_TYPE_DEV,
                                        snd_dma_pci_data(chip->pci),
                                        64*1024, 64*1024);

          

It will allocate a buffer up to 64kB as default. Buffer management details will be described in the later section Buffer and Memory Management.

Additionally, you can set some extra information for this pcm in pcm->info_flags. The available values are defined as SNDRV_PCM_INFO_XXX in <sound/asound.h>, which is used for the hardware definition (described later). When your soundchip supports only half-duplex, specify like this:

  pcm->info_flags = SNDRV_PCM_INFO_HALF_DUPLEX;

          

The destructor for a pcm instance is not always necessary. Since the pcm device will be released by the middle layer code automatically, you don't have to call the destructor explicitly.

The destructor would be necessary if you created special records internally and needed to release them. In such a case, set the destructor function to pcm->private_free:

  static void mychip_pcm_free(struct snd_pcm *pcm)
  {
          struct mychip *chip = snd_pcm_chip(pcm);
          /* free your own data */
          kfree(chip->my_private_pcm_data);
          /* do what you like else */
          ....
  }

  static int __devinit snd_mychip_new_pcm(struct mychip *chip)
  {
          struct snd_pcm *pcm;
          ....
          /* allocate your own data */
          chip->my_private_pcm_data = kmalloc(...);
          /* set the destructor */
          pcm->private_data = chip;
          pcm->private_free = mychip_pcm_free;
          ....
  }

          

When the PCM substream is opened, a PCM runtime instance is allocated and assigned to the substream. This pointer is accessible via substream->runtime. This runtime pointer holds most information you need to control the PCM: the copy of hw_params and sw_params configurations, the buffer pointers, mmap records, spinlocks, etc.

The definition of runtime instance is found in <sound/pcm.h>. Here are the contents of this file:

struct _snd_pcm_runtime {
	/* -- Status -- */
	struct snd_pcm_substream *trigger_master;
	snd_timestamp_t trigger_tstamp;	/* trigger timestamp */
	int overrange;
	snd_pcm_uframes_t avail_max;
	snd_pcm_uframes_t hw_ptr_base;	/* Position at buffer restart */
	snd_pcm_uframes_t hw_ptr_interrupt; /* Position at interrupt time*/

	/* -- HW params -- */
	snd_pcm_access_t access;	/* access mode */
	snd_pcm_format_t format;	/* SNDRV_PCM_FORMAT_* */
	snd_pcm_subformat_t subformat;	/* subformat */
	unsigned int rate;		/* rate in Hz */
	unsigned int channels;		/* channels */
	snd_pcm_uframes_t period_size;	/* period size */
	unsigned int periods;		/* periods */
	snd_pcm_uframes_t buffer_size;	/* buffer size */
	unsigned int tick_time;		/* tick time */
	snd_pcm_uframes_t min_align;	/* Min alignment for the format */
	size_t byte_align;
	unsigned int frame_bits;
	unsigned int sample_bits;
	unsigned int info;
	unsigned int rate_num;
	unsigned int rate_den;

	/* -- SW params -- */
	struct timespec tstamp_mode;	/* mmap timestamp is updated */
  	unsigned int period_step;
	unsigned int sleep_min;		/* min ticks to sleep */
	snd_pcm_uframes_t start_threshold;
	snd_pcm_uframes_t stop_threshold;
	snd_pcm_uframes_t silence_threshold; /* Silence filling happens when
						noise is nearest than this */
	snd_pcm_uframes_t silence_size;	/* Silence filling size */
	snd_pcm_uframes_t boundary;	/* pointers wrap point */

	snd_pcm_uframes_t silenced_start;
	snd_pcm_uframes_t silenced_size;

	snd_pcm_sync_id_t sync;		/* hardware synchronization ID */

	/* -- mmap -- */
	volatile struct snd_pcm_mmap_status *status;
	volatile struct snd_pcm_mmap_control *control;
	atomic_t mmap_count;

	/* -- locking / scheduling -- */
	spinlock_t lock;
	wait_queue_head_t sleep;
	struct timer_list tick_timer;
	struct fasync_struct *fasync;

	/* -- private section -- */
	void *private_data;
	void (*private_free)(struct snd_pcm_runtime *runtime);

	/* -- hardware description -- */
	struct snd_pcm_hardware hw;
	struct snd_pcm_hw_constraints hw_constraints;

	/* -- interrupt callbacks -- */
	void (*transfer_ack_begin)(struct snd_pcm_substream *substream);
	void (*transfer_ack_end)(struct snd_pcm_substream *substream);

	/* -- timer -- */
	unsigned int timer_resolution;	/* timer resolution */

	/* -- DMA -- */           
	unsigned char *dma_area;	/* DMA area */
	dma_addr_t dma_addr;		/* physical bus address (not accessible from main CPU) */
	size_t dma_bytes;		/* size of DMA area */

	struct snd_dma_buffer *dma_buffer_p;	/* allocated buffer */

#if defined(CONFIG_SND_PCM_OSS) || defined(CONFIG_SND_PCM_OSS_MODULE)
	/* -- OSS things -- */
	struct snd_pcm_oss_runtime oss;
#endif
};

            

For the operators (callbacks) of each sound driver, most of these records are supposed to be read-only. Only the PCM middle-layer changes / updates them. The exceptions are the hardware description (hw), interrupt callbacks (transfer_ack_xxx), DMA buffer information, and the private data. Besides, if you use the standard buffer allocation method via snd_pcm_lib_malloc_pages(), you don't need to set the DMA buffer information by yourself.

In the sections below, important records are explained.

          struct snd_pcm_runtime *runtime = substream->runtime;
          ...
          runtime->hw = snd_mychip_playback_hw; /* common definition */
          if (chip->model == VERY_OLD_ONE)
                  runtime->hw.channels_max = 1;

            

  static struct snd_pcm_hardware snd_mychip_playback_hw = {
          .info = (SNDRV_PCM_INFO_MMAP |
                   SNDRV_PCM_INFO_INTERLEAVED |
                   SNDRV_PCM_INFO_BLOCK_TRANSFER |
                   SNDRV_PCM_INFO_MMAP_VALID),
          .formats =          SNDRV_PCM_FMTBIT_S16_LE,
          .rates =            SNDRV_PCM_RATE_8000_48000,
          .rate_min =         8000,
          .rate_max =         48000,
          .channels_min =     2,
          .channels_max =     2,
          .buffer_bytes_max = 32768,
          .period_bytes_min = 4096,
          .period_bytes_max = 32768,
          .periods_min =      1,
          .periods_max =      1024,
  };

            

  period_bytes = frames_to_bytes(runtime, runtime->period_size);

            

  static int snd_xxx_open(struct snd_pcm_substream *substream)
  {
          struct my_pcm_data *data;
          ....
          data = kmalloc(sizeof(*data), GFP_KERNEL);
          substream->runtime->private_data = data;
          ....
  }

            

OK, now let me give details about each pcm callback (ops). In general, every callback must return 0 if successful, or a negative error number such as -EINVAL. To choose an appropriate error number, it is advised to check what value other parts of the kernel return when the same kind of request fails.

The callback function takes at least the argument with snd_pcm_substream pointer. To retrieve the chip record from the given substream instance, you can use the following macro.

  int xxx() {
          struct mychip *chip = snd_pcm_substream_chip(substream);
          ....
  }

          

The macro reads substream->private_data, which is a copy of pcm->private_data. You can override the former if you need to assign different data records per PCM substream. For example, the cmi8330 driver assigns different private_data for playback and capture directions, because it uses two different codecs (SB- and AD-compatible) for different directions.

  static int snd_xxx_open(struct snd_pcm_substream *substream);

            

  static int snd_xxx_open(struct snd_pcm_substream *substream)
  {
          struct mychip *chip = snd_pcm_substream_chip(substream);
          struct snd_pcm_runtime *runtime = substream->runtime;

          runtime->hw = snd_mychip_playback_hw;
          return 0;
  }

            

  static int snd_xxx_close(struct snd_pcm_substream *substream);

            

  static int snd_xxx_close(struct snd_pcm_substream *substream)
  {
          ....
          kfree(substream->runtime->private_data);
          ....
  }

            

  static int snd_xxx_hw_params(struct snd_pcm_substream *substream,
                               struct snd_pcm_hw_params *hw_params);

            

  snd_pcm_lib_malloc_pages(substream, params_buffer_bytes(hw_params));

            

  static int snd_xxx_hw_free(struct snd_pcm_substream *substream);

            

  snd_pcm_lib_free_pages(substream);

            

  static int snd_xxx_prepare(struct snd_pcm_substream *substream);

            

  static int snd_xxx_trigger(struct snd_pcm_substream *substream, int cmd);

            

  switch (cmd) {
  case SNDRV_PCM_TRIGGER_START:
          /* do something to start the PCM engine */
          break;
  case SNDRV_PCM_TRIGGER_STOP:
          /* do something to stop the PCM engine */
          break;
  default:
          return -EINVAL;
  }

            

  static snd_pcm_uframes_t snd_xxx_pointer(struct snd_pcm_substream *substream)

            

The rest of pcm stuff is the PCM interrupt handler. The role of PCM interrupt handler in the sound driver is to update the buffer position and to tell the PCM middle layer when the buffer position goes across the prescribed period size. To inform this, call the snd_pcm_period_elapsed() function.

There are several types of sound chips to generate the interrupts.

  static irqreturn_t snd_mychip_interrupt(int irq, void *dev_id)
  {
          struct mychip *chip = dev_id;
          spin_lock(&chip->lock);
          ....
          if (pcm_irq_invoked(chip)) {
                  /* call updater, unlock before it */
                  spin_unlock(&chip->lock);
                  snd_pcm_period_elapsed(chip->substream);
                  spin_lock(&chip->lock);
                  /* acknowledge the interrupt if necessary */
          }
          ....
          spin_unlock(&chip->lock);
          return IRQ_HANDLED;
  }

            

  static irqreturn_t snd_mychip_interrupt(int irq, void *dev_id)
  {
          struct mychip *chip = dev_id;
          spin_lock(&chip->lock);
          ....
          if (pcm_irq_invoked(chip)) {
                  unsigned int last_ptr, size;
                  /* get the current hardware pointer (in frames) */
                  last_ptr = get_hw_ptr(chip);
                  /* calculate the processed frames since the
                   * last update
                   */
                  if (last_ptr < chip->last_ptr)
                          size = runtime->buffer_size + last_ptr 
                                   - chip->last_ptr; 
                  else
                          size = last_ptr - chip->last_ptr;
                  /* remember the last updated point */
                  chip->last_ptr = last_ptr;
                  /* accumulate the size */
                  chip->size += size;
                  /* over the period boundary? */
                  if (chip->size >= runtime->period_size) {
                          /* reset the accumulator */
                          chip->size %= runtime->period_size;
                          /* call updater */
                          spin_unlock(&chip->lock);
                          snd_pcm_period_elapsed(substream);
                          spin_lock(&chip->lock);
                  }
                  /* acknowledge the interrupt if necessary */
          }
          ....
          spin_unlock(&chip->lock);
          return IRQ_HANDLED;
  }

            

One of the most important (and thus difficult to debug) problems in kernel programming are race conditions. In the Linux kernel, they are usually avoided via spin-locks, mutexes or semaphores. In general, if a race condition can happen in an interrupt handler, it has to be managed atomically, and you have to use a spinlock to protect the critical session. If the critical section is not in interrupt handler code and if taking a relatively long time to execute is acceptable, you should use mutexes or semaphores instead.

As already seen, some pcm callbacks are atomic and some are not. For example, the hw_params callback is non-atomic, while trigger callback is atomic. This means, the latter is called already in a spinlock held by the PCM middle layer. Please take this atomicity into account when you choose a locking scheme in the callbacks.

In the atomic callbacks, you cannot use functions which may call schedule or go to sleep. Semaphores and mutexes can sleep, and hence they cannot be used inside the atomic callbacks (e.g. trigger callback). To implement some delay in such a callback, please use udelay() or mdelay().

All three atomic callbacks (trigger, pointer, and ack) are called with local interrupts disabled.

If your chip supports unconventional sample rates, or only the limited samples, you need to set a constraint for the condition.

For example, in order to restrict the sample rates in the some supported values, use snd_pcm_hw_constraint_list(). You need to call this function in the open callback.

  static unsigned int rates[] =
          {4000, 10000, 22050, 44100};
  static struct snd_pcm_hw_constraint_list constraints_rates = {
          .count = ARRAY_SIZE(rates),
          .list = rates,
          .mask = 0,
  };

  static int snd_mychip_pcm_open(struct snd_pcm_substream *substream)
  {
          int err;
          ....
          err = snd_pcm_hw_constraint_list(substream->runtime, 0,
                                           SNDRV_PCM_HW_PARAM_RATE,
                                           &constraints_rates);
          if (err < 0)
                  return err;
          ....
  }

          

There are many different constraints. Look at sound/pcm.h for a complete list. You can even define your own constraint rules. For example, let's suppose my_chip can manage a substream of 1 channel if and only if the format is S16_LE, otherwise it supports any format specified in the snd_pcm_hardware structure (or in any other constraint_list). You can build a rule like this:

  static int hw_rule_format_by_channels(struct snd_pcm_hw_params *params,
                                        struct snd_pcm_hw_rule *rule)
  {
          struct snd_interval *c = hw_param_interval(params,
                SNDRV_PCM_HW_PARAM_CHANNELS);
          struct snd_mask *f = hw_param_mask(params, SNDRV_PCM_HW_PARAM_FORMAT);
          struct snd_mask fmt;

          snd_mask_any(&fmt);    /* Init the struct */
          if (c->min < 2) {
                  fmt.bits[0] &= SNDRV_PCM_FMTBIT_S16_LE;
                  return snd_mask_refine(f, &fmt);
          }
          return 0;
  }

          

Then you need to call this function to add your rule:

  snd_pcm_hw_rule_add(substream->runtime, 0, SNDRV_PCM_HW_PARAM_CHANNELS,
                      hw_rule_channels_by_format, 0, SNDRV_PCM_HW_PARAM_FORMAT,
                      -1);

          

The rule function is called when an application sets the number of channels. But an application can set the format before the number of channels. Thus you also need to define the inverse rule:

  static int hw_rule_channels_by_format(struct snd_pcm_hw_params *params,
                                        struct snd_pcm_hw_rule *rule)
  {
          struct snd_interval *c = hw_param_interval(params,
                        SNDRV_PCM_HW_PARAM_CHANNELS);
          struct snd_mask *f = hw_param_mask(params, SNDRV_PCM_HW_PARAM_FORMAT);
          struct snd_interval ch;

          snd_interval_any(&ch);
          if (f->bits[0] == SNDRV_PCM_FMTBIT_S16_LE) {
                  ch.min = ch.max = 1;
                  ch.integer = 1;
                  return snd_interval_refine(c, &ch);
          }
          return 0;
  }

          

...and in the open callback:

  snd_pcm_hw_rule_add(substream->runtime, 0, SNDRV_PCM_HW_PARAM_FORMAT,
                      hw_rule_format_by_channels, 0, SNDRV_PCM_HW_PARAM_CHANNELS,
                      -1);

          

I won't give more details here, rather I would like to say, “Luke, use the source.”

The control interface is used widely for many switches, sliders, etc. which are accessed from user-space. Its most important use is the mixer interface. In other words, since ALSA 0.9.x, all the mixer stuff is implemented on the control kernel API.

ALSA has a well-defined AC97 control module. If your chip supports only the AC97 and nothing else, you can skip this section.

The control API is defined in <sound/control.h>. Include this file if you want to add your own controls.

To create a new control, you need to define the following three callbacks: info, get and put. Then, define a struct snd_kcontrol_new record, such as:

  static struct snd_kcontrol_new my_control __devinitdata = {
          .iface = SNDRV_CTL_ELEM_IFACE_MIXER,
          .name = "PCM Playback Switch",
          .index = 0,
          .access = SNDRV_CTL_ELEM_ACCESS_READWRITE,
          .private_value = 0xffff,
          .info = my_control_info,
          .get = my_control_get,
          .put = my_control_put
  };

          

Most likely the control is created via snd_ctl_new1(), and in such a case, you can add the __devinitdata prefix to the definition as above.

The iface field specifies the control type, SNDRV_CTL_ELEM_IFACE_XXX, which is usually MIXER. Use CARD for global controls that are not logically part of the mixer. If the control is closely associated with some specific device on the sound card, use HWDEP, PCM, RAWMIDI, TIMER, or SEQUENCER, and specify the device number with the device and subdevice fields.

The name is the name identifier string. Since ALSA 0.9.x, the control name is very important, because its role is classified from its name. There are pre-defined standard control names. The details are described in the Control Names subsection.

The index field holds the index number of this control. If there are several different controls with the same name, they can be distinguished by the index number. This is the case when several codecs exist on the card. If the index is zero, you can omit the definition above.

The access field contains the access type of this control. Give the combination of bit masks, SNDRV_CTL_ELEM_ACCESS_XXX, there. The details will be explained in the Access Flags subsection.

The private_value field contains an arbitrary long integer value for this record. When using the generic info, get and put callbacks, you can pass a value through this field. If several small numbers are necessary, you can combine them in bitwise. Or, it's possible to give a pointer (casted to unsigned long) of some record to this field, too.

The tlv field can be used to provide metadata about the control; see the Metadata subsection.

The other three are callback functions.

There are some standards to define the control names. A control is usually defined from the three parts as “SOURCE DIRECTION FUNCTION”.

The first, SOURCE, specifies the source of the control, and is a string such as “Master”, “PCM”, “CD” and “Line”. There are many pre-defined sources.

The second, DIRECTION, is one of the following strings according to the direction of the control: “Playback”, “Capture”, “Bypass Playback” and “Bypass Capture”. Or, it can be omitted, meaning both playback and capture directions.

The third, FUNCTION, is one of the following strings according to the function of the control: “Switch”, “Volume” and “Route”.

The example of control names are, thus, “Master Capture Switch” or “PCM Playback Volume”.

There are some exceptions:

The access flag is the bitmask which specifies the access type of the given control. The default access type is SNDRV_CTL_ELEM_ACCESS_READWRITE, which means both read and write are allowed to this control. When the access flag is omitted (i.e. = 0), it is considered as READWRITE access as default.

When the control is read-only, pass SNDRV_CTL_ELEM_ACCESS_READ instead. In this case, you don't have to define the put callback. Similarly, when the control is write-only (although it's a rare case), you can use the WRITE flag instead, and you don't need the get callback.

If the control value changes frequently (e.g. the VU meter), VOLATILE flag should be given. This means that the control may be changed without notification. Applications should poll such a control constantly.

When the control is inactive, set the INACTIVE flag, too. There are LOCK and OWNER flags to change the write permissions.

When everything is ready, finally we can create a new control. To create a control, there are two functions to be called, snd_ctl_new1() and snd_ctl_add().

In the simplest way, you can do like this:

  err = snd_ctl_add(card, snd_ctl_new1(&my_control, chip));
  if (err < 0)
          return err;

          

where my_control is the struct snd_kcontrol_new object defined above, and chip is the object pointer to be passed to kcontrol->private_data which can be referred to in callbacks.

snd_ctl_new1() allocates a new snd_kcontrol instance (that's why the definition of my_control can be with the __devinitdata prefix), and snd_ctl_add assigns the given control component to the card.

If you need to change and update a control in the interrupt routine, you can call snd_ctl_notify(). For example,

  snd_ctl_notify(card, SNDRV_CTL_EVENT_MASK_VALUE, id_pointer);

          

This function takes the card pointer, the event-mask, and the control id pointer for the notification. The event-mask specifies the types of notification, for example, in the above example, the change of control values is notified. The id pointer is the pointer of struct snd_ctl_elem_id to be notified. You can find some examples in es1938.c or es1968.c for hardware volume interrupts.

To provide information about the dB values of a mixer control, use on of the DECLARE_TLV_xxx macros from <sound/tlv.h> to define a variable containing this information, set thetlv.p field to point to this variable, and include the SNDRV_CTL_ELEM_ACCESS_TLV_READ flag in the access field; like this:

  static DECLARE_TLV_DB_SCALE(db_scale_my_control, -4050, 150, 0);

  static struct snd_kcontrol_new my_control __devinitdata = {
          ...
          .access = SNDRV_CTL_ELEM_ACCESS_READWRITE |
                    SNDRV_CTL_ELEM_ACCESS_TLV_READ,
          ...
          .tlv.p = db_scale_my_control,
  };

        

The DECLARE_TLV_DB_SCALE macro defines information about a mixer control where each step in the control's value changes the dB value by a constant dB amount. The first parameter is the name of the variable to be defined. The second parameter is the minimum value, in units of 0.01 dB. The third parameter is the step size, in units of 0.01 dB. Set the fourth parameter to 1 if the minimum value actually mutes the control.

The DECLARE_TLV_DB_LINEAR macro defines information about a mixer control where the control's value affects the output linearly. The first parameter is the name of the variable to be defined. The second parameter is the minimum value, in units of 0.01 dB. The third parameter is the maximum value, in units of 0.01 dB. If the minimum value mutes the control, set the second parameter to TLV_DB_GAIN_MUTE.

The ALSA AC97 codec layer is a well-defined one, and you don't have to write much code to control it. Only low-level control routines are necessary. The AC97 codec API is defined in <sound/ac97_codec.h>.

To create an ac97 instance, first call snd_ac97_bus with an ac97_bus_ops_t record with callback functions.

  struct snd_ac97_bus *bus;
  static struct snd_ac97_bus_ops ops = {
        .write = snd_mychip_ac97_write,
        .read = snd_mychip_ac97_read,
  };

  snd_ac97_bus(card, 0, &ops, NULL, &pbus);

          

The bus record is shared among all belonging ac97 instances.

And then call snd_ac97_mixer() with an struct snd_ac97_template record together with the bus pointer created above.

  struct snd_ac97_template ac97;
  int err;

  memset(&ac97, 0, sizeof(ac97));
  ac97.private_data = chip;
  snd_ac97_mixer(bus, &ac97, &chip->ac97);

          

where chip->ac97 is a pointer to a newly created ac97_t instance. In this case, the chip pointer is set as the private data, so that the read/write callback functions can refer to this chip instance. This instance is not necessarily stored in the chip record. If you need to change the register values from the driver, or need the suspend/resume of ac97 codecs, keep this pointer to pass to the corresponding functions.

The standard callbacks are read and write. Obviously they correspond to the functions for read and write accesses to the hardware low-level codes.

The read callback returns the register value specified in the argument.

  static unsigned short snd_mychip_ac97_read(struct snd_ac97 *ac97,
                                             unsigned short reg)
  {
          struct mychip *chip = ac97->private_data;
          ....
          return the_register_value;
  }

          

Here, the chip can be cast from ac97->private_data.

Meanwhile, the write callback is used to set the register value.

  static void snd_mychip_ac97_write(struct snd_ac97 *ac97,
                       unsigned short reg, unsigned short val)

          

These callbacks are non-atomic like the control API callbacks.

There are also other callbacks: reset, wait and init.

The reset callback is used to reset the codec. If the chip requires a special kind of reset, you can define this callback.

The wait callback is used to add some waiting time in the standard initialization of the codec. If the chip requires the extra waiting time, define this callback.

The init callback is used for additional initialization of the codec.

If you need to access to the codec from the driver, you can call the following functions: snd_ac97_write(), snd_ac97_read(), snd_ac97_update() and snd_ac97_update_bits().

Both snd_ac97_write() and snd_ac97_update() functions are used to set a value to the given register (AC97_XXX). The difference between them is that snd_ac97_update() doesn't write a value if the given value has been already set, while snd_ac97_write() always rewrites the value.

  snd_ac97_write(ac97, AC97_MASTER, 0x8080);
  snd_ac97_update(ac97, AC97_MASTER, 0x8080);

          

snd_ac97_read() is used to read the value of the given register. For example,

  value = snd_ac97_read(ac97, AC97_MASTER);

          

snd_ac97_update_bits() is used to update some bits in the given register.

  snd_ac97_update_bits(ac97, reg, mask, value);

          

Also, there is a function to change the sample rate (of a given register such as AC97_PCM_FRONT_DAC_RATE) when VRA or DRA is supported by the codec: snd_ac97_set_rate().

  snd_ac97_set_rate(ac97, AC97_PCM_FRONT_DAC_RATE, 44100);

          

The following registers are available to set the rate: AC97_PCM_MIC_ADC_RATE, AC97_PCM_FRONT_DAC_RATE, AC97_PCM_LR_ADC_RATE, AC97_SPDIF. When AC97_SPDIF is specified, the register is not really changed but the corresponding IEC958 status bits will be updated.

In some chips, the clock of the codec isn't 48000 but using a PCI clock (to save a quartz!). In this case, change the field bus->clock to the corresponding value. For example, intel8x0 and es1968 drivers have their own function to read from the clock.

The ALSA AC97 interface will create a proc file such as /proc/asound/card0/codec97#0/ac97#0-0 and ac97#0-0+regs. You can refer to these files to see the current status and registers of the codec.

When there are several codecs on the same card, you need to call snd_ac97_mixer() multiple times with ac97.num=1 or greater. The num field specifies the codec number.

If you set up multiple codecs, you either need to write different callbacks for each codec or check ac97->num in the callback routines.

Many soundcards have built-in MIDI (MPU401-UART) interfaces. When the soundcard supports the standard MPU401-UART interface, most likely you can use the ALSA MPU401-UART API. The MPU401-UART API is defined in <sound/mpu401.h>.

Some soundchips have a similar but slightly different implementation of mpu401 stuff. For example, emu10k1 has its own mpu401 routines.

To create a rawmidi object, call snd_mpu401_uart_new().

  struct snd_rawmidi *rmidi;
  snd_mpu401_uart_new(card, 0, MPU401_HW_MPU401, port, info_flags,
                      irq, irq_flags, &rmidi);

          

The first argument is the card pointer, and the second is the index of this component. You can create up to 8 rawmidi devices.

The third argument is the type of the hardware, MPU401_HW_XXX. If it's not a special one, you can use MPU401_HW_MPU401.

The 4th argument is the I/O port address. Many backward-compatible MPU401 have an I/O port such as 0x330. Or, it might be a part of its own PCI I/O region. It depends on the chip design.

The 5th argument is a bitflag for additional information. When the I/O port address above is part of the PCI I/O region, the MPU401 I/O port might have been already allocated (reserved) by the driver itself. In such a case, pass a bit flag MPU401_INFO_INTEGRATED, and the mpu401-uart layer will allocate the I/O ports by itself.

When the controller supports only the input or output MIDI stream, pass the MPU401_INFO_INPUT or MPU401_INFO_OUTPUT bitflag, respectively. Then the rawmidi instance is created as a single stream.

MPU401_INFO_MMIO bitflag is used to change the access method to MMIO (via readb and writeb) instead of iob and outb. In this case, you have to pass the iomapped address to snd_mpu401_uart_new().

When MPU401_INFO_TX_IRQ is set, the output stream isn't checked in the default interrupt handler. The driver needs to call snd_mpu401_uart_interrupt_tx() by itself to start processing the output stream in the irq handler.

Usually, the port address corresponds to the command port and port + 1 corresponds to the data port. If not, you may change the cport field of struct snd_mpu401 manually afterward. However, snd_mpu401 pointer is not returned explicitly by snd_mpu401_uart_new(). You need to cast rmidi->private_data to snd_mpu401 explicitly,

  struct snd_mpu401 *mpu;
  mpu = rmidi->private_data;

          

and reset the cport as you like:

  mpu->cport = my_own_control_port;

          

The 6th argument specifies the irq number for UART. If the irq is already allocated, pass 0 to the 7th argument (irq_flags). Otherwise, pass the flags for irq allocation (SA_XXX bits) to it, and the irq will be reserved by the mpu401-uart layer. If the card doesn't generate UART interrupts, pass -1 as the irq number. Then a timer interrupt will be invoked for polling.

When the interrupt is allocated in snd_mpu401_uart_new(), the private interrupt handler is used, hence you don't have anything else to do than creating the mpu401 stuff. Otherwise, you have to call snd_mpu401_uart_interrupt() explicitly when a UART interrupt is invoked and checked in your own interrupt handler.

In this case, you need to pass the private_data of the returned rawmidi object from snd_mpu401_uart_new() as the second argument of snd_mpu401_uart_interrupt().

  snd_mpu401_uart_interrupt(irq, rmidi->private_data, regs);

          

The raw MIDI interface is used for hardware MIDI ports that can be accessed as a byte stream. It is not used for synthesizer chips that do not directly understand MIDI.

ALSA handles file and buffer management. All you have to do is to write some code to move data between the buffer and the hardware.

The rawmidi API is defined in <sound/rawmidi.h>.

To create a rawmidi device, call the snd_rawmidi_new function:

  struct snd_rawmidi *rmidi;
  err = snd_rawmidi_new(chip->card, "MyMIDI", 0, outs, ins, &rmidi);
  if (err < 0)
          return err;
  rmidi->private_data = chip;
  strcpy(rmidi->name, "My MIDI");
  rmidi->info_flags = SNDRV_RAWMIDI_INFO_OUTPUT |
                      SNDRV_RAWMIDI_INFO_INPUT |
                      SNDRV_RAWMIDI_INFO_DUPLEX;

          

The first argument is the card pointer, the second argument is the ID string.

The third argument is the index of this component. You can create up to 8 rawmidi devices.

The fourth and fifth arguments are the number of output and input substreams, respectively, of this device (a substream is the equivalent of a MIDI port).

Set the info_flags field to specify the capabilities of the device. Set SNDRV_RAWMIDI_INFO_OUTPUT if there is at least one output port, SNDRV_RAWMIDI_INFO_INPUT if there is at least one input port, and SNDRV_RAWMIDI_INFO_DUPLEX if the device can handle output and input at the same time.

After the rawmidi device is created, you need to set the operators (callbacks) for each substream. There are helper functions to set the operators for all the substreams of a device:

  snd_rawmidi_set_ops(rmidi, SNDRV_RAWMIDI_STREAM_OUTPUT, &snd_mymidi_output_ops);
  snd_rawmidi_set_ops(rmidi, SNDRV_RAWMIDI_STREAM_INPUT, &snd_mymidi_input_ops);

          

The operators are usually defined like this:

  static struct snd_rawmidi_ops snd_mymidi_output_ops = {
          .open =    snd_mymidi_output_open,
          .close =   snd_mymidi_output_close,
          .trigger = snd_mymidi_output_trigger,
  };

          

These callbacks are explained in the Callbacks section.

If there are more than one substream, you should give a unique name to each of them:

  struct snd_rawmidi_substream *substream;
  list_for_each_entry(substream,
                      &rmidi->streams[SNDRV_RAWMIDI_STREAM_OUTPUT].substreams,
                      list {
          sprintf(substream->name, "My MIDI Port %d", substream->number + 1);
  }
  /* same for SNDRV_RAWMIDI_STREAM_INPUT */

          

In all the callbacks, the private data that you've set for the rawmidi device can be accessed as substream->rmidi->private_data.

If there is more than one port, your callbacks can determine the port index from the struct snd_rawmidi_substream data passed to each callback:

  struct snd_rawmidi_substream *substream;
  int index = substream->number;

          

  static int snd_xxx_open(struct snd_rawmidi_substream *substream);

          

  static int snd_xxx_close(struct snd_rawmidi_substream *substream);

          

  static void snd_xxx_output_trigger(struct snd_rawmidi_substream *substream, int up);

          

  unsigned char data;
  while (snd_rawmidi_transmit_peek(substream, &data, 1) == 1) {
          if (snd_mychip_try_to_transmit(data))
                  snd_rawmidi_transmit_ack(substream, 1);
          else
                  break; /* hardware FIFO full */
  }

            

  while (snd_mychip_transmit_possible()) {
          unsigned char data;
          if (snd_rawmidi_transmit(substream, &data, 1) != 1)
                  break; /* no more data */
          snd_mychip_transmit(data);
  }

            

  static void snd_xxx_input_trigger(struct snd_rawmidi_substream *substream, int up);

          

  void snd_mychip_midi_interrupt(...)
  {
          while (mychip_midi_available()) {
                  unsigned char data;
                  data = mychip_midi_read();
                  snd_rawmidi_receive(substream, &data, 1);
          }
  }

            

  static void snd_xxx_drain(struct snd_rawmidi_substream *substream);

          

The FM OPL3 is still used in many chips (mainly for backward compatibility). ALSA has a nice OPL3 FM control layer, too. The OPL3 API is defined in <sound/opl3.h>.

FM registers can be directly accessed through the direct-FM API, defined in <sound/asound_fm.h>. In ALSA native mode, FM registers are accessed through the Hardware-Dependant Device direct-FM extension API, whereas in OSS compatible mode, FM registers can be accessed with the OSS direct-FM compatible API in /dev/dmfmX device.

To create the OPL3 component, you have two functions to call. The first one is a constructor for the opl3_t instance.

  struct snd_opl3 *opl3;
  snd_opl3_create(card, lport, rport, OPL3_HW_OPL3_XXX,
                  integrated, &opl3);

          

The first argument is the card pointer, the second one is the left port address, and the third is the right port address. In most cases, the right port is placed at the left port + 2.

The fourth argument is the hardware type.

When the left and right ports have been already allocated by the card driver, pass non-zero to the fifth argument (integrated). Otherwise, the opl3 module will allocate the specified ports by itself.

When the accessing the hardware requires special method instead of the standard I/O access, you can create opl3 instance separately with snd_opl3_new().

  struct snd_opl3 *opl3;
  snd_opl3_new(card, OPL3_HW_OPL3_XXX, &opl3);

          

Then set command, private_data and private_free for the private access function, the private data and the destructor. The l_port and r_port are not necessarily set. Only the command must be set properly. You can retrieve the data from the opl3->private_data field.

After creating the opl3 instance via snd_opl3_new(), call snd_opl3_init() to initialize the chip to the proper state. Note that snd_opl3_create() always calls it internally.

If the opl3 instance is created successfully, then create a hwdep device for this opl3.

  struct snd_hwdep *opl3hwdep;
  snd_opl3_hwdep_new(opl3, 0, 1, &opl3hwdep);

          

The first argument is the opl3_t instance you created, and the second is the index number, usually 0.

The third argument is the index-offset for the sequencer client assigned to the OPL3 port. When there is an MPU401-UART, give 1 for here (UART always takes 0).

Some chips need user-space access for special controls or for loading the micro code. In such a case, you can create a hwdep (hardware-dependent) device. The hwdep API is defined in <sound/hwdep.h>. You can find examples in opl3 driver or isa/sb/sb16_csp.c.

The creation of the hwdep instance is done via snd_hwdep_new().

  struct snd_hwdep *hw;
  snd_hwdep_new(card, "My HWDEP", 0, &hw);

          

where the third argument is the index number.

You can then pass any pointer value to the private_data. If you assign a private data, you should define the destructor, too. The destructor function is set in the private_free field.

  struct mydata *p = kmalloc(sizeof(*p), GFP_KERNEL);
  hw->private_data = p;
  hw->private_free = mydata_free;

          

and the implementation of the destructor would be:

  static void mydata_free(struct snd_hwdep *hw)
  {
          struct mydata *p = hw->private_data;
          kfree(p);
  }

          

The arbitrary file operations can be defined for this instance. The file operators are defined in the ops table. For example, assume that this chip needs an ioctl.

  hw->ops.open = mydata_open;
  hw->ops.ioctl = mydata_ioctl;
  hw->ops.release = mydata_release;

          

And implement the callback functions as you like.

Usually the controls for IEC958 devices are implemented via the control interface. There is a macro to compose a name string for IEC958 controls, SNDRV_CTL_NAME_IEC958() defined in <include/asound.h>.

There are some standard controls for IEC958 status bits. These controls use the type SNDRV_CTL_ELEM_TYPE_IEC958, and the size of element is fixed as 4 bytes array (value.iec958.status[x]). For the info callback, you don't specify the value field for this type (the count field must be set, though).

“IEC958 Playback Con Mask” is used to return the bit-mask for the IEC958 status bits of consumer mode. Similarly, “IEC958 Playback Pro Mask” returns the bitmask for professional mode. They are read-only controls, and are defined as MIXER controls (iface = SNDRV_CTL_ELEM_IFACE_MIXER).

Meanwhile, “IEC958 Playback Default” control is defined for getting and setting the current default IEC958 bits. Note that this one is usually defined as a PCM control (iface = SNDRV_CTL_ELEM_IFACE_PCM), although in some places it's defined as a MIXER control.

In addition, you can define the control switches to enable/disable or to set the raw bit mode. The implementation will depend on the chip, but the control should be named as “IEC958 xxx”, preferably using the SNDRV_CTL_NAME_IEC958() macro.

You can find several cases, for example, pci/emu10k1, pci/ice1712, or pci/cmipci.c.

ALSA provides several different buffer allocation functions depending on the bus and the architecture. All these have a consistent API. The allocation of physically-contiguous pages is done via snd_malloc_xxx_pages() function, where xxx is the bus type.

The allocation of pages with fallback is snd_malloc_xxx_pages_fallback(). This function tries to allocate the specified pages but if the pages are not available, it tries to reduce the page sizes until enough space is found.

The release the pages, call snd_free_xxx_pages() function.

Usually, ALSA drivers try to allocate and reserve a large contiguous physical space at the time the module is loaded for the later use. This is called “pre-allocation”. As already written, you can call the following function at pcm instance construction time (in the case of PCI bus).

  snd_pcm_lib_preallocate_pages_for_all(pcm, SNDRV_DMA_TYPE_DEV,
                                        snd_dma_pci_data(pci), size, max);

          

where size is the byte size to be pre-allocated and the max is the maximum size to be changed via the prealloc proc file. The allocator will try to get an area as large as possible within the given size.

The second argument (type) and the third argument (device pointer) are dependent on the bus. In the case of the ISA bus, pass snd_dma_isa_data() as the third argument with SNDRV_DMA_TYPE_DEV type. For the continuous buffer unrelated to the bus can be pre-allocated with SNDRV_DMA_TYPE_CONTINUOUS type and the snd_dma_continuous_data(GFP_KERNEL) device pointer, where GFP_KERNEL is the kernel allocation flag to use. For the PCI scatter-gather buffers, use SNDRV_DMA_TYPE_DEV_SG with snd_dma_pci_data(pci) (see the Non-Contiguous Buffers section).

Once the buffer is pre-allocated, you can use the allocator in the hw_params callback:

  snd_pcm_lib_malloc_pages(substream, size);

          

Note that you have to pre-allocate to use this function.

Some chips have their own hardware buffers and the DMA transfer from the host memory is not available. In such a case, you need to either 1) copy/set the audio data directly to the external hardware buffer, or 2) make an intermediate buffer and copy/set the data from it to the external hardware buffer in interrupts (or in tasklets, preferably).

The first case works fine if the external hardware buffer is large enough. This method doesn't need any extra buffers and thus is more effective. You need to define the copy and silence callbacks for the data transfer. However, there is a drawback: it cannot be mmapped. The examples are GUS's GF1 PCM or emu8000's wavetable PCM.

The second case allows for mmap on the buffer, although you have to handle an interrupt or a tasklet to transfer the data from the intermediate buffer to the hardware buffer. You can find an example in the vxpocket driver.

Another case is when the chip uses a PCI memory-map region for the buffer instead of the host memory. In this case, mmap is available only on certain architectures like the Intel one. In non-mmap mode, the data cannot be transferred as in the normal way. Thus you need to define the copy and silence callbacks as well, as in the cases above. The examples are found in rme32.c and rme96.c.

The implementation of the copy and silence callbacks depends upon whether the hardware supports interleaved or non-interleaved samples. The copy callback is defined like below, a bit differently depending whether the direction is playback or capture:

  static int playback_copy(struct snd_pcm_substream *substream, int channel,
               snd_pcm_uframes_t pos, void *src, snd_pcm_uframes_t count);
  static int capture_copy(struct snd_pcm_substream *substream, int channel,
               snd_pcm_uframes_t pos, void *dst, snd_pcm_uframes_t count);

          

In the case of interleaved samples, the second argument (channel) is not used. The third argument (pos) points the current position offset in frames.

The meaning of the fourth argument is different between playback and capture. For playback, it holds the source data pointer, and for capture, it's the destination data pointer.

The last argument is the number of frames to be copied.

What you have to do in this callback is again different between playback and capture directions. In the playback case, you copy the given amount of data (count) at the specified pointer (src) to the specified offset (pos) on the hardware buffer. When coded like memcpy-like way, the copy would be like:

  my_memcpy(my_buffer + frames_to_bytes(runtime, pos), src,
            frames_to_bytes(runtime, count));

          

For the capture direction, you copy the given amount of data (count) at the specified offset (pos) on the hardware buffer to the specified pointer (dst).

  my_memcpy(dst, my_buffer + frames_to_bytes(runtime, pos),
            frames_to_bytes(runtime, count));

          

Note that both the position and the amount of data are given in frames.

In the case of non-interleaved samples, the implementation will be a bit more complicated.

You need to check the channel argument, and if it's -1, copy the whole channels. Otherwise, you have to copy only the specified channel. Please check isa/gus/gus_pcm.c as an example.

The silence callback is also implemented in a similar way.

  static int silence(struct snd_pcm_substream *substream, int channel,
                     snd_pcm_uframes_t pos, snd_pcm_uframes_t count);

          

The meanings of arguments are the same as in the copy callback, although there is no src/dst argument. In the case of interleaved samples, the channel argument has no meaning, as well as on copy callback.

The role of silence callback is to set the given amount (count) of silence data at the specified offset (pos) on the hardware buffer. Suppose that the data format is signed (that is, the silent-data is 0), and the implementation using a memset-like function would be like:

  my_memcpy(my_buffer + frames_to_bytes(runtime, pos), 0,
            frames_to_bytes(runtime, count));

          

In the case of non-interleaved samples, again, the implementation becomes a bit more complicated. See, for example, isa/gus/gus_pcm.c.

If your hardware supports the page table as in emu10k1 or the buffer descriptors as in via82xx, you can use the scatter-gather (SG) DMA. ALSA provides an interface for handling SG-buffers. The API is provided in <sound/pcm.h>.

For creating the SG-buffer handler, call snd_pcm_lib_preallocate_pages() or snd_pcm_lib_preallocate_pages_for_all() with SNDRV_DMA_TYPE_DEV_SG in the PCM constructor like other PCI pre-allocator. You need to pass snd_dma_pci_data(pci), where pci is the struct pci_dev pointer of the chip as well. The struct snd_sg_buf instance is created as substream->dma_private. You can cast the pointer like:

  struct snd_sg_buf *sgbuf = (struct snd_sg_buf *)substream->dma_private;

          

Then call snd_pcm_lib_malloc_pages() in the hw_params callback as well as in the case of normal PCI buffer. The SG-buffer handler will allocate the non-contiguous kernel pages of the given size and map them onto the virtually contiguous memory. The virtual pointer is addressed in runtime->dma_area. The physical address (runtime->dma_addr) is set to zero, because the buffer is physically non-contigous. The physical address table is set up in sgbuf->table. You can get the physical address at a certain offset via snd_pcm_sgbuf_get_addr().

When a SG-handler is used, you need to set snd_pcm_sgbuf_ops_page as the page callback. (See page callback section.)

To release the data, call snd_pcm_lib_free_pages() in the hw_free callback as usual.

It's possible to use a buffer allocated via vmalloc, for example, for an intermediate buffer. Since the allocated pages are not contiguous, you need to set the page callback to obtain the physical address at every offset.

The implementation of page callback would be like this:

  #include <linux/vmalloc.h>

  /* get the physical page pointer on the given offset */
  static struct page *mychip_page(struct snd_pcm_substream *substream,
                                  unsigned long offset)
  {
          void *pageptr = substream->runtime->dma_area + offset;
          return vmalloc_to_page(pageptr);
  }

          

So far, you've learned how to write the driver codes. And you might have a question now: how to put my own driver into the ALSA driver tree? Here (finally :) the standard procedure is described briefly.

Suppose that you create a new PCI driver for the card “xyz”. The card module name would be snd-xyz. The new driver is usually put into the alsa-driver tree, alsa-driver/pci directory in the case of PCI cards. Then the driver is evaluated, audited and tested by developers and users. After a certain time, the driver will go to the alsa-kernel tree (to the corresponding directory, such as alsa-kernel/pci) and eventually will be integrated into the Linux 2.6 tree (the directory would be linux/sound/pci).

In the following sections, the driver code is supposed to be put into alsa-driver tree. The two cases are covered: a driver consisting of a single source file and one consisting of several source files.

Suppose that the driver snd-xyz have several source files. They are located in the new subdirectory, pci/xyz.

ALSA provides a verbose version of the printk() function. If a kernel config CONFIG_SND_VERBOSE_PRINTK is set, this function prints the given message together with the file name and the line of the caller. The KERN_XXX prefix is processed as well as the original printk() does, so it's recommended to add this prefix, e.g.

  snd_printk(KERN_ERR "Oh my, sorry, it's extremely bad!\n");

          

There are also printk()'s for debugging. snd_printd() can be used for general debugging purposes. If CONFIG_SND_DEBUG is set, this function is compiled, and works just like snd_printk(). If the ALSA is compiled without the debugging flag, it's ignored.

snd_printdd() is compiled in only when CONFIG_SND_DEBUG_VERBOSE is set. Please note that CONFIG_SND_DEBUG_VERBOSE is not set as default even if you configure the alsa-driver with --with-debug=full option. You need to give explicitly --with-debug=detect option instead.

It shows the BUG? message and stack trace as well as snd_BUG_ON at the point. It's useful to show that a fatal error happens there.

When no debug flag is set, this macro is ignored.

snd_BUG_ON() macro is similar with WARN_ON() macro. For example,

  snd_BUG_ON(!pointer);

          

or it can be used as the condition,

  if (snd_BUG_ON(non_zero_is_bug))
          return -EINVAL;

          

The macro takes an conditional expression to evaluate. When CONFIG_SND_DEBUG, is set, the expression is actually evaluated. If it's non-zero, it shows the warning message such as BUG? (xxx) normally followed by stack trace. It returns the evaluated value. When no CONFIG_SND_DEBUG is set, this macro always returns zero.