powerpc/mm: Cleanup management of kmem_caches for pagetables

[firewire-audio.git] / drivers / staging / vme / vme_api.txt

                        VME Device Driver API
                        =====================

Driver registration
===================

As with other subsystems within the Linux kernel, VME device drivers register
with the VME subsystem, typically called from the devices init routine.  This is
achieved via a call to the follwoing function:

        int vme_register_driver (struct vme_driver *driver);

If driver registration is successful this function returns zero, if an error
occurred a negative error code will be returned.

A pointer to a structure of type 'vme_driver' must be provided to the
registration function. The structure is as follows:

        struct vme_driver {
                struct list_head node;
                char *name;
                const struct vme_device_id *bind_table;
                int (*probe)  (struct device *, int, int);
                int (*remove) (struct device *, int, int);
                void (*shutdown) (void);
                struct device_driver    driver;
        };

At the minimum, the '.name', '.probe' and '.bind_table' elements of this
structure should be correctly set. The '.name' element is a pointer to a string
holding the device driver's name. The '.probe' element should contain a pointer
to the probe routine.

The arguments of the probe routine are as follows:

        probe(struct device *dev, int bus, int slot);

The '.bind_table' is a pointer to an array of type 'vme_device_id':

        struct vme_device_id {
                int bus;
                int slot;
        };

Each structure in this array should provide a bus and slot number where the core
should probe, using the driver's probe routine, for a device on the specified
VME bus.

The VME subsystem supports a single VME driver per 'slot'. There are considered
to be 32 slots per bus, one for each slot-ID as defined in the ANSI/VITA 1-1994
specification and are analogious to the physical slots on the VME backplane.

A function is also provided to unregister the driver from the VME core and is
usually called from the device driver's exit routine:

        void vme_unregister_driver (struct vme_driver *driver);


Resource management
===================

Once a driver has registered with the VME core the provided probe routine will
be called for each of the bus/slot combination that becomes valid as VME buses
are themselves registered.  The probe routine is passed a pointer to the devices
device structure. This pointer should be saved, it will be required for
requesting VME resources.

The driver can request ownership of one or more master windows, slave windows
and/or dma channels. Rather than allowing the device driver to request a
specific window or DMA channel (which may be used by a different driver) this
driver allows a resource to be assigned based on the required attributes of the
driver in question:

        struct vme_resource * vme_master_request(struct device *dev,
                vme_address_t aspace, vme_cycle_t cycle, vme_width_t width);

        struct vme_resource * vme_slave_request(struct device *dev,
                vme_address_t aspace, vme_cycle_t cycle);

        struct vme_resource *vme_request_dma(struct device *dev);

For slave windows these attributes are split into those of type 'vme_address_t'
and 'vme_cycle_t'. Master windows add a further set of attributes 'vme_cycle_t'.
These attributes are defined as bitmasks and as such any combination of the
attributes can be requested for a single window, the core will assign a window
that meets the requirements, returning a pointer of type vme_resource that
should be used to identify the allocated resource when it is used. If an
unallocated window fitting the requirements can not be found a NULL pointer will
be returned.

Functions are also provided to free window allocations once they are no longer
required. These functions should be passed the pointer to the resource provided
during resource allocation:

        void vme_master_free(struct vme_resource *res);

        void vme_slave_free(struct vme_resource *res);

        void vme_dma_free(struct vme_resource *res);


Master windows
==============

Master windows provide access from the local processor[s] out onto the VME bus.
The number of windows available and the available access modes is dependant on
the underlying chipset. A window must be configured before it can be used.


Master window configuration
---------------------------

Once a master window has been assigned the following functions can be used to
configure it and retrieve the current settings:

        int vme_master_set (struct vme_resource *res, int enabled,
                unsigned long long base, unsigned long long size,
                vme_address_t aspace, vme_cycle_t cycle, vme_width_t width);

        int vme_master_get (struct vme_resource *res, int *enabled,
                unsigned long long *base, unsigned long long *size,
                vme_address_t *aspace, vme_cycle_t *cycle, vme_width_t *width);

The address spaces, transfer widths and cycle types are the same as described
under resource management, however some of the options are mutually exclusive.
For example, only one address space may be specified.

These functions return 0 on success or an error code should the call fail.


Master window access
--------------------

The following functions can be used to read from and write to configured master
windows. These functions return the number of bytes copied:

        ssize_t vme_master_read(struct vme_resource *res, void *buf,
                size_t count, loff_t offset);

        ssize_t vme_master_write(struct vme_resource *res, void *buf,
                size_t count, loff_t offset);

In addition to simple reads and writes, a function is provided to do a
read-modify-write transaction. This function returns the original value of the
VME bus location :

        unsigned int vme_master_rmw (struct vme_resource *res,
                unsigned int mask, unsigned int compare, unsigned int swap,
                loff_t offset);

This functions by reading the offset, applying the mask. If the bits selected in
the mask match with the values of the corresponding bits in the compare field,
the value of swap is written the specified offset.


Slave windows
=============

Slave windows provide devices on the VME bus access into mapped portions of the
local memory. The number of windows available and the access modes that can be
used is dependant on the underlying chipset. A window must be configured before
it can be used.


Slave window configuration
--------------------------

Once a slave window has been assigned the following functions can be used to
configure it and retrieve the current settings:

        int vme_slave_set (struct vme_resource *res, int enabled,
                unsigned long long base, unsigned long long size,
                dma_addr_t mem, vme_address_t aspace, vme_cycle_t cycle);

        int vme_slave_get (struct vme_resource *res, int *enabled,
                unsigned long long *base, unsigned long long *size,
                dma_addr_t *mem, vme_address_t *aspace, vme_cycle_t *cycle);

The address spaces, transfer widths and cycle types are the same as described
under resource management, however some of the options are mutually exclusive.
For example, only one address space may be specified.

These functions return 0 on success or an error code should the call fail.


Slave window buffer allocation
------------------------------

Functions are provided to allow the user to allocate and free a contiguous
buffers which will be accessible by the VME bridge. These functions do not have
to be used, other methods can be used to allocate a buffer, though care must be
taken to ensure that they are contiguous and accessible by the VME bridge:

        void * vme_alloc_consistent(struct vme_resource *res, size_t size,
                dma_addr_t *mem);

        void vme_free_consistent(struct vme_resource *res, size_t size,
                void *virt,     dma_addr_t mem);


Slave window access
-------------------

Slave windows map local memory onto the VME bus, the standard methods for
accessing memory should be used.


DMA channels
============

The VME DMA transfer provides the ability to run link-list DMA transfers. The
API introduces the concept of DMA lists. Each DMA list is a link-list which can
be passed to a DMA controller. Multiple lists can be created, extended,
executed, reused and destroyed.


List Management
---------------

The following functions are provided to create and destroy DMA lists. Execution
of a list will not automatically destroy the list, thus enabling a list to be
reused for repetitive tasks:

        struct vme_dma_list *vme_new_dma_list(struct vme_resource *res);

        int vme_dma_list_free(struct vme_dma_list *list);


List Population
---------------

An item can be added to a list using the following function ( the source and
destination attributes need to be created before calling this function, this is
covered under "Transfer Attributes"):

        int vme_dma_list_add(struct vme_dma_list *list,
                struct vme_dma_attr *src, struct vme_dma_attr *dest,
                size_t count);


Transfer Attributes
-------------------

The attributes for the source and destination are handled separately from adding
an item to a list. This is due to the diverse attributes required for each type
of source and destination. There are functions to create attributes for PCI, VME
and pattern sources and destinations (where appropriate):

Pattern source:

        struct vme_dma_attr *vme_dma_pattern_attribute(u32 pattern,
                vme_pattern_t type);

PCI source or destination:

        struct vme_dma_attr *vme_dma_pci_attribute(dma_addr_t mem);

VME source or destination:

        struct vme_dma_attr *vme_dma_vme_attribute(unsigned long long base,
                vme_address_t aspace, vme_cycle_t cycle, vme_width_t width);

The following function should be used to free an attribute:

        void vme_dma_free_attribute(struct vme_dma_attr *attr);


List Execution
--------------

The following function queues a list for execution. The function will return
once the list has been executed:

        int vme_dma_list_exec(struct vme_dma_list *list);


Interrupts
==========

The VME API provides functions to attach and detach callbacks to specific VME
level and status ID combinations and for the generation of VME interrupts with
specific VME level and status IDs.


Attaching Interrupt Handlers
----------------------------

The following functions can be used to attach and free a specific VME level and
status ID combination. Any given combination can only be assigned a single
callback function. A void pointer parameter is provided, the value of which is
passed to the callback function, the use of this pointer is user undefined:

        int vme_request_irq(struct device *dev, int level, int statid,
                void (*callback)(int, int, void *), void *priv);

        void vme_free_irq(struct device *dev, int level, int statid);

The callback parameters are as follows. Care must be taken in writing a callback
function, callback functions run in interrupt context:

        void callback(int level, int statid, void *priv);


Interrupt Generation
--------------------

The following function can be used to generate a VME interrupt at a given VME
level and VME status ID:

        int vme_generate_irq(struct device *dev, int level, int statid);


Location monitors
=================

The VME API provides the following functionality to configure the location
monitor.


Location Monitor Management
---------------------------

The following functions are provided to request the use of a block of location
monitors and to free them after they are no longer required:

        struct vme_resource * vme_lm_request(struct device *dev);

        void vme_lm_free(struct vme_resource * res);

Each block may provide a number of location monitors, monitoring adjacent
locations. The following function can be used to determine how many locations
are provided:

        int vme_lm_count(struct vme_resource * res);


Location Monitor Configuration
------------------------------

Once a bank of location monitors has been allocated, the following functions
are provided to configure the location and mode of the location monitor:

        int vme_lm_set(struct vme_resource *res, unsigned long long base,
                vme_address_t aspace, vme_cycle_t cycle);

        int vme_lm_get(struct vme_resource *res, unsigned long long *base,
                vme_address_t *aspace, vme_cycle_t *cycle);


Location Monitor Use
--------------------

The following functions allow a callback to be attached and detached from each
location monitor location. Each location monitor can monitor a number of
adjacent locations:

        int vme_lm_attach(struct vme_resource *res, int num,
                void (*callback)(int));

        int vme_lm_detach(struct vme_resource *res, int num);

The callback function is declared as follows.

        void callback(int num);


Slot Detection
==============

This function returns the slot ID of the provided bridge.

        int vme_slot_get(struct device *dev);