Linux PCMCIA Programmers Guide David Hinds, dhinds@allegro.stanford.edu v1.21, 1995/06/01 02:50:34 This document describes how to write kernel device drivers for the Linux PCMCIA Card Services interface. It also describes how to write user-mode utilities for communicating with Card Services. The latest version of this document can always be found at cb-iris.stanford.edu in /pub/pcmcia/doc. 1. Introduction The Linux kernel PCMCIA system has three main components. At the lowest level are the socket drivers. Next is the Card Services module. Drivers for specific cards are layered on top of Card Services. One special Card Services client, called Driver Services, provides a link betweek user level PCMCIA utility programs and the kernel PCMCIA facilities. The socket driver layer is loosely based on the Socket Services API. There are two socket driver modules. The tcic module supports the Databook TCIC-2 family of PCMCIA controllers. The i82365 module supports the Intel i82365sl family and various Intel-compatible controllers, including Cirrus, VLSI, Ricoh, and Vadem chips. Card Services is the largest single component of the PCMCIA package. It provides an API somewhat similar to DOS Card Services, adapted to a Unix environment. The Linux implementation was based in part on the Solaris PCMCIA interface specification. It is implemented in the pcmcia_core module. The Driver Services layer implements a user mode pseudo-device for accessing some Card Services functions from PCMCIA utility programs. It is responsible for keeping track of all PCMCIA client drivers, and for matching up drivers with physical sockets. It is implemented in the ds module. This document describes the kernel interface to the Card Services and Driver Services modules, and the user interface to Driver Services. It is intended for use by PCMCIA device driver developers. The Linux PCMCIA-HOWTO describes how to install and use Linux PCMCIA support. It is available from cb-iris.stanford.edu in /pub/pcmcia. 1.1. Copyright notice and disclaimer Copyright (c) 1995 David A. Hinds This document may be reproduced or distributed in any form without my prior permission. Parts of this document may be distributed, provided that this copyright message and a pointer to the complete document are included. Specifically, it may be included in commercial distributions without my prior consent. However, I would like to be informed of such usage. This document may be translated into any language, provided this copyright statement is left intact. This document is provided ``as is'', with no explicit or implied warranties. Use the information in this document at your own risk. 1.2. Acknowledgements I'd like to thank all the Linux users who have helped test and debug this PCMCIA software, and who have helped with driver development. I should also thank Linus Torvalds, Donald Becker, Alan Cox, and Bjorn Ekwall for Linux kernel development help. I'm especially grateful to Michael Bender for many helpful discussions about the Solaris PCMCIA implementation. 2. Basic PCMCIA Concepts 2.1. The PCMCIA socket interface The PCMCIA card bus has two basic operating modes: ``memory-only'' and ``memory and IO''. The first mode was defined by the original Version 1.0 specification and only supports simple memory cards. The second mode, defined in Version 2.0, redefines a few of the memory card control signals to support IO port addressing and IO interrupt signalling. PCMCIA cards have two memory spaces: ``attribute memory'' and ``common memory''. The PCMCIA interface can address up to 16MB of each type of memory. Attribute memory is typically used for holding descriptive information and configuration registers. Common memory may include the bulk storage of a memory card, or device buffers in the case of IO cards. All cards that are compliant with the version 2.0 PCMCIA specification should have a Card Information Structure (or ``CIS'') in attribute memory, which describes the card and how it should be configured. Separate control signals allow cards to signal their operating status to the host. These signals include card detect, ready/busy, write protect, battery low, and battery dead. The ``memory and IO'' interface mode allows cards to address up to 64K of IO ports. It also allows cards to signal IO interrupts, and routes one card output to the host system's speaker. In this mode, several of the memory card control signals are unavailable because those pins are used to carry the extra IO card signals. On some cards, these signals can instead be read from a special configuration register in attribute memory, the ``Pin Replacement Register''. 2.2. The PCMCIA socket controller The PCMCIA controller serves as a bridge between PCMCIA cards and the system bus. There are several varieties of controllers, but all share the same basic functionality. The Socket Services software layer takes care of all the details of how to program the PCMCIA controller. The PCMCIA controller has the job of mapping windows of addresses in the host memory and IO spaces to windows of addresses in card space. All controllers support at least four independent memory windows and two IO windows per socket. Each memory window is defined by a base address in the host address space, a base address in the card address space, and a window size. Some controllers differ in their alignment rules for memory windows, but all controllers will support windows whose size is at least 4K and also a power of two, and where the base address is a multiple of the window size. Each window can be programmed to point to either attribute or common memory. IO windows differ from memory windows in that host addresses that fall within an IO window are not modified before they are passed on to an IO card. Effectively, the base addresses of the window in the host and card address spaces are always equal. IO windows also have no alignment or size restrictions; an IO window can start and end on any byte boundary in the 64K IO address space. The PCMCIA card bus defines a single interrupt signal from the card to the controller. The controller then has the responsibility of steering this interrupt to an appropriate interrupt request (``irq'') line. All controllers support steering card IO interrupts to essentially any free interrupt line. Because steering happens in the controller, the card itself is unaware of which interrupt it uses. All PCMCIA controllers can generate interrupts in response to card status changes. These interrupts are distinct from the IO interrupts generated by an IO card, and use a separate interrupt line. Signals that can generate interrupts include card detect, ready/busy, write protect, battery low, and battery dead. 3. Card Services subfunction descriptions Card Services calls have the general form: #include "cs_types.h" #include "cs.h" int CardServices(int subfunc, void *arg1, void *arg2, ...); Some Card Services functions require additional #include statements. The particular subfunction determines the number of expected arguments. A return code of CS_SUCCESS indicates that a call succeeded. Other return codes indicate errors. 3.1. Client management functions Device drivers that use Card Services functions are called ``clients''. A device driver should use the RegisterClient call to get a client handle before using other services. Most Card Services functions will take this client handle as an argument. Before unloading, drivers should also unregister with DeregisterClient. 3.1.1. RegisterClient int CardServices(RegisterClient, client_handle_t *client, client_reg_t *reg); The client_reg_t data structure is given by: typedef struct client_reg_t { dev_info_t *dev_info; u_long Attributes; u_long EventMask; int (*event_handler)(event_t event, int priority, event_callback_args_t *args); event_callback_args_t event_callback_args; u_long Version; } client_reg_t; RegisterClient establishes a link between a client driver and Card Services, and connects the client with an appropriate socket. The dev_info parameter is used by Card Services to match the client with a socket; this correspondence is normally established by Driver Services via a call to BindDevice. If successful, a client handle will be returned in client. The following flags can be specified in Attributes: INFO_DRIVER_SERVICES For use only by the Driver Services client, among other things, specifies that this client should not be automatically unbound when a card is ejected from this socket. INFO_IO_CLIENT Specifies that this client is an IO card driver. INFO_MTD_CLIENT Specifies that this client is a Memory Technology Driver. INFO_MEM_CLIENT Specifies that this client is a memory card driver. INFO_CARD_SHARE Included for compatibility, has no effect. INFO_CARD_EXCL Included for compatibility, has no effect. EventMask specifies what events this client should be notified of. The event_handler entry point will be called by Card Services when an event in EventMask is processed. The event_handler_args structure is a template for the structure that will be passed to the event handler. The Version parameter identifies the Card Services version level that this driver expects; it is currently ignored. A driver should be prepared to handle Card Services events before calling RegisterClient. This call will always generate a CS_REGISTRATION_COMPLETE event, and may also generate an artificial CS_CARD_INSERTION event if the socket is currently occupied. Return codes: CS_OUT_OF_RESOURCE An appropriate socket could not be found for this driver. 3.1.2. DeregisterClient int CardServices(DeregisterClient, client_handle_t client); DeregisterClient severs the connection between a client and Card Services. It should be called after the client has freed any resources it has allocated. Once a connection is broken, it cannot be reestablished until after another call to BindDevice. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_IN_USE The client still has allocated resources, such as IO port windows or an interrupt, or the socket configuration is locked. 3.1.3. SetEventMask int CardServices(SetEventMask, client_handle_t client, eventmask_t *mask); The eventmask_t structure is given by: typedef struct eventmask_t { u_long Attributes; u_long EventMask; } eventmask_t; SetEventMask updates the mask that determines which events this client will be notified of. Return codes: CS_BAD_HANDLE The client handle is invalid. 3.1.4. BindDevice int CardServices(BindDevice, bind_req_t *req); The bind_req structure is given by: typedef struct bind_req_t { socket_t Socket; dev_info_t *dev_info; } bind_req_t; BindDevice associates a device driver with a particular socket. It is normally called by Driver Services after a newly inserted card has been identified. Once a driver has been bound to a socket, it will be eligible to register as a client of that socket. Note that this call does not take a client handle as an argument. This is the only Card Services call that takes a socket number as an argument. Return codes: CS_BAD_SOCKET The specified socket number is invalid. 3.2. Socket state control These functions are more or less concerned with getting and setting the current operating state of a socket. GetStatus returns the current socket state. ResetCard is used to send a hard reset signal to a socket. SuspendCard and ResumeCard can be used to power down and power up a socket without releasing the drivers currently bound to that socket. EjectCard and InsertCard essentially mimic real card ejection and insertion events. 3.2.1. GetStatus int CardServices(GetStatus, client_handle_t client, status_t *status); The status_t data structure is given by: typedef struct status_t { u_long CardState; u_long SocketState; } status_t; GetStatus returns the current status of a client's socket. The following flags are defined in CardState: CS_EVENT_CARD_DETECT Specifies that the socket is occupied. CS_EVENT_WRITE_PROTECT Specifies that the card is currently write protected. CS_EVENT_BATTERY_LOW Specifies that the card battery is low. CS_EVENT_BATTERY_DEAD Specifies that the card battery is dead. CS_EVENT_READY_CHANGE Specifies that the card is ready. CS_EVENT_PM_SUSPEND Specifies that the socket is suspended. SocketState is currently unused, but in theory, it should latch changes in the state of the fields in CardState. Return codes: CS_BAD_HANDLE The client handle is invalid. 3.2.2. ResetCard int CardServices(ResetCard, client_handle_t client); ResetCard requests that a client's socket be reset. When this call is made, Card Services sends all clients a CS_EVENT_RESET_REQUEST event. If any client rejects the request, Card Services sends the initiating client a CS_EVENT_RESET_COMPLETE event with event_callback_args.info set to the return code of the client that rejected the request. If all clients agree to the request, Card Services sends a CS_EVENT_RESET_PHYSICAL event, then resets the socket. When the socket signals that it is ready, a CS_EVENT_CARD_RESET event is generated. Finally, a CS_EVENT_RESET_COMPLETE event is sent to the initiating client, with event_callback_args.info set to zero. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_CARD The socket assigned to this client is currently vacant. CS_IN_USE This socket is currently being reset. 3.2.3. SuspendCard int CardServices(SuspendCard, client_handle_t client); Card Services sends all clients CS_EVENT_PM_SUSPEND events, then shuts down and turns off power to the socket. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_CARD The socket assigned to this client is currently vacant. CS_IN_USE This socket is already suspended. 3.2.4. ResumeCard int CardServices(ResumeCard, client_handle_t client); After restoring power to the socket, Card Services will notify all clients with CS_EVENT_PM_RESUME events. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_CARD The socket assigned to this client is currently vacant. CS_IN_USE This socket is not currently suspended. 3.2.5. EjectCard int CardServices(EjectCard, client_handle_t client); Card Services sends eject events to all clients, then shuts down and turns off power to the socket. All clients except for Driver Services will be unlinked from the socket. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_CARD The socket assigned to this client is currently vacant. 3.2.6. InsertCard int CardServices(InsertCard, client_handle_t client); Card Services sends insertion events to all clients of this socket (normally, only Driver Services). Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_CARD The socket assigned to this client is currently vacant. CS_IN_USE The socket has already been configured. 3.3. IO card configuration calls The normal order of events is for a driver to reserve IO ports and an interrupt line with calls to RequestIO and RequestIRQ, then to call RequestConfiguration to actually configure the socket. If any of these calls fails, a driver should be sure to release any resources it successfully reserved. 3.3.1. RequestIO int CardServices(RequestIO, client_handle_t client, io_req_t *req); The io_req_t data structure is given by: typedef struct io_req_t { ioaddr_t BasePort1; ioaddr_t NumPorts1; u_long Attributes1; ioaddr_t BasePort2; ioaddr_t NumPorts2; u_long Attributes2; u_long IOAddrLines; } io_req_t; RequestIO reserves IO port windows for a card. If BasePort1 is non- zero, it specifies the IO port address of the window to be reserved; if it is zero, Card Services will find an available window and set BasePort1 to this address. If NumPorts2 is non-zero, a second IO port window will also be reserved. IOAddrLines specifies the number of address lines that are actually decoded by the PCMCIA card; this is not currently used. This call does not actually configure a socket's IO windows: this is done by a subsequent call to RequestConfiguration. The following flags can be specified in Attributes1 and Attributes2: IO_DATA_PATH_WIDTH This field may either be IO_DATA_PATH_WIDTH_16 for 16-bit access, or IO_DATA_PATH_WIDTH_8 for 8-bit access, or IO_DATA_PATH_WIDTH_AUTO to dynamically size the bus based on the access size. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_CARD The socket assigned to this client is currently vacant. CS_IN_USE This socket's IO windows have already been reserved. CS_CONFIGURATION_LOCKED This socket's configuration has been locked by a call to RequestConfiguration. CS_BAD_ATTRIBUTE An unsupported attribute flag was specified. 3.3.2. ReleaseIO int CardServices(ReleaseIO, client_handle_t client, io_req_t *req); ReleaseIO un-reserves IO port windows allocated by a previous call to RequestIO. The req parameter should be the same one passed to RequestIO. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_CONFIGURATION_LOCKED This socket's configuration has been locked by a call to RequestConfiguration. The configuration should be released before calling ReleaseIO. CS_BAD_ARGS The parameters in req do not match the parameters passed to RequestIO. 3.3.3. RequestIRQ int CardServices(RequestIRQ, client_handle_t client, irq_req_t *req); The irq_req_t structure is given by: typedef struct irq_req_t { u_long Attributes; u_long AssignedIRQ; u_long IRQInfo1, IRQInfo2; } irq_req_t; RequestIRQ reserves an interrupt line for use by a PCMCIA card. The IRQInfo1 and IRQInfo2 fields correspond to the interrupt description bytes in a CFTABLE_ENTRY tuple. If IRQ_INFO2_VALID is set in IRQInfo1, then IRQInfo2 is a bit-mapped mask of allowed interrupt values. Each bit corresponds to one interrupt line: bit 0 = irq 0, bit 1 = irq 1, etc. So, a mask of 0x1100 would mean that interrupts 12 and 8 could be used. If IRQ_INFO2_VALID is not set, IRQInfo1 is just the desired interrupt number. If the call is successful, the reserved interrupt is returned in AssignedIRQ. Note that this call does not install an interrupt handler. The calling driver should use the request_irq() kernel function to actually catch the assigned interrupt. Also, the interrupt will not be enabled until a call to RequestConfiguration. The following flags can be specified in Attributes: IRQ_FORCED_PULSE Specifies that the interrupt should be configured for pulsed mode, rather than the default level mode. IRQ_TYPE_TIME Specifies that this interrupt can be time-shared with other Card Services drivers. Only one driver should enable the interrupt at any time. IRQ_FIRST_SHARED In conjunction with IRQ_TYPE_TIME, this should be set by the first driver requesting a shared interrupt. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_CARD The socket assigned to this client is currently vacant. CS_IN_USE An interrupt has already been reserved for this socket, or the requested interrupt is unavailable. CS_CONFIGURATION_LOCKED This socket's configuration has been locked by a call to RequestConfiguration. CS_BAD_ATTRIBUTE An unsupported attribute flag was specified. 3.3.4. ReleaseIRQ int CardServices(ReleaseIRQ, client_handle_t client, irq_req_t *req); ReleaseIRQ un-reserves an interrupt assigned by an earlier call to RequestIRQ. The req structure should be the same structure that was passed to RequestIRQ. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_CONFIGURATION_LOCKED This socket's configuration has been locked by a call to RequestConfiguration. The configuration should be released before calling ReleaseIRQ. CS_BAD_IRQ The parameters in req do not match the parameters passed to RequestIRQ. 3.3.5. RequestConfiguration int CardServices(RequestConfiguration, client_handle_t client, config_req_t *req); The config_req_t structure is given by: typedef struct config_req_t { u_long Attributes; u_long Vcc, Vpp1, Vpp2; u_long IntType; caddr_t ConfigBase; u_char Status, Pin, Copy; u_char ConfigIndex; u_long Present; } config_req_t; RequestConfiguration is responsible for actually configuring a socket. This includes setting voltages, setting CIS configuration registers, setting up IO port windows, and setting up interrupts. IntType specifies the type of interface to use for this card. It may either be INT_MEMORY or INT_MEMORY_AND_IO. Voltages are specified in units of 1/10 volt. Currently, Vpp1 must equal Vpp2. The following flags can be specified in Attributes. DMA and speaker control are not supported on all systems. CONF_ENABLE_IRQ Enable the IO interrupt reserved by a previous call to RequestIRQ. CONF_ENABLE_DMA Enable DMA accesses for this socket. CONF_ENABLE_SPKR Enable speaker output from this socket. The Present parameter is a bit map specifying which CIS configuration registers are implemented by this card. ConfigBase gives the offset of the configuration registers in attribute memory. The following registers can be specified: PRESENT_OPTION Specifies that the Configuration Option Register is present. The COR register will be set using the ConfigIndex parameter. PRESENT_STATUS Specifies that the Card Configuration and Status Register is present. The CCSR will be initialized with the Status parameter. PRESENT_PIN_REPLACE Specifies that the Pin Replacement Register is present. The PRR will be initialized with the Pin parameter. PRESENT_COPY Specifies that the Socket and Copy Register is present. The SCR will be initialized with the Copy parameter. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_CARD The socket assigned to this client is currently vacant. CS_OUT_OF_RESOURCE Card Services was unable to allocate a memory window to access the card's configuration registers. CS_CONFIGURATION_LOCKED This socket's configuration has already been locked by another call to RequestConfiguration. CS_BAD_VCC The requested Vcc voltage is not supported. CS_BAD_VPP The requested Vpp1/Vpp2 voltage is not supported. 3.3.6. ModifyConfiguration int CardServices(ModifyConfiguration, client_handle_t client, modconf_t *mod); The modconf_t structure is given by: typedef struct modconf_t { u_long Attributes; u_long Vcc, Vpp1, Vpp2; } modconf_t; ModifyConfiguration modifies some attributes of a socket that has been configured by a call to RequestConfiguration. The following flags can be specified in Attributes: CONF_IRQ_CHANGE_VALID Indicates that the CONF_ENABLE_IRQ setting should be updated. CONF_ENABLE_IRQ Specifies that IO interrupts should be enabled for this socket. CONF_VCC_CHANGE_VALID Indicates that Vcc should be updated. CONF_VPP1_CHANGE_VALID Indicates that Vpp1 should be updated. CONF_VPP2_CHANGE_VALID Indicates that Vpp2 should be updated. Currently, Vpp1 and Vpp2 must always have the same value. So, the two values must always be changed at the same time. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_CARD The socket assigned to this client is currently vacant. CS_CONFIGURATION_LOCKED This actually means that this socket has not been locked. CS_BAD_VCC The requested Vcc voltage is not supported. CS_BAD_VPP The requested Vpp1/Vpp2 voltage is not supported. 3.3.7. ReleaseConfiguration int CardServices(ReleaseConfiguration, client_handle_t client, config_req_t *req); ReleaseConfiguration un-configures a socket previously set up by a call to RequestConfiguration. The req parameter should be the same one used to configure the socket. Return codes: CS_BAD_HANDLE The window handle is invalid, or the socket is not configured. 3.3.8. GetConfigurationInfo int CardServices(GetConfigurationInfo, client_handle_t client, config_t *config); The config_t structure is given by: typedef struct config_t { u_long Attributes; u_long Vcc, Vpp1, Vpp2; u_long IntType; caddr_t ConfigBase; u_char Status, Pin, Copy, Option; u_long Present; u_long AssignedIRQ; u_long IRQAttributes; ioaddr_t BasePort1; ioaddr_t NumPorts1; u_long Attributes1; ioaddr_t BasePort2; ioaddr_t NumPorts2; u_long Attributes2; u_long IOAddrLines; } config_t; GetConfigurationInfo returns the current socket configuration as it was set up by RequestIO, RequestIRQ, and RequestConfiguration. It can only be applied to a fully configured socket. Return codes: CS_BAD_HANDLE The window handle is invalid, or the socket is not configured. CS_NO_CARD The socket assigned to this client is currently vacant. CS_CONFIGURATION_LOCKED This actually means that the configuration has not been locked. 3.4. Card Information Structure (CIS) calls The definition of the Card Information Structure (CIS) is the darkest chapter of the PCMCIA standard. All version 2 PCMCIA cards should have a CIS, which describes the card and how it should be configured. The CIS is a linked list of ``tuples'' in the card's attribute memory space. Each tuple consists of an identification code, a length byte, and a series of data bytes. The layout of the data bytes for some tuple types is absurdly complicated, in an apparent effort to use every last bit. The ValidateCIS call checks to see if a card has a reasonable CIS. The GetFirstTuple and GetNextTuple calls are used to step through CIS tuple lists. GetTupleData extracts data bytes from a tuple. And ParseTuple interprets a limited number of particularly important tuple types. 3.4.1. GetFirstTuple, GetNextTuple #include "cistpl.h" int CardServices(GetFirstTuple, client_handle_t client, tuple_t *tuple); int CardServices(GetNextTuple, client_handle_t client, tuple_t *tuple); The tuple_t data structure is given by: typedef struct tuple_t { u_long Attributes; cis_data_t DesiredTuple; u_long Flags; cisdata_t TupleCode; u_long TupleLink; cisdata_t TupleOffset; cisdata_t TupleDataMax; cisdata_t TupleDataLen; cisdata_t *TupleData; } tuple_t; GetFirstTuple searches a card's CIS for the first tuple code matching DesiredTuple. The special code RETURN_FIRST_TUPLE will match the first tuple of any kind. If successful, TupleCode is set to the code of the first matching tuple found, and TupleLink is the address of this tuple in attribute memory. GetNextTuple is like GetFirstTuple, except that given a tuple_t structure returned by a previous call to GetFirstTuple or GetNextTuple, it will return the next tuple matching DesiredTuple. The following flags can be specified in Attributes: TUPLE_RETURN_LINK Indicates that link tuples (CISTPL_LONGLINK_A, CISTPL_LONGLINK_C, CISTPL_NOLINK, CISTPL_LINKTARGET) should be returned. Normally these tuples are processed silently. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_OUT_OF_RESOURCE Card Services was unable to set up a memory window to map the card's CIS. CS_NO_MORE_ITEMS There were no tuples matching DesiredTuple. 3.4.2. GetTupleData #include "cistpl.h" int CardServices(GetTupleData, client_handle_t client, tuple_t *tuple); GetTupleData extracts a series of data bytes from the specified tuple, which must have been returned by a previous call to GetFirstTuple or GetNextTuple. A maximum of TupleDataMax bytes will be copied into the TupleData buffer, starting at an offset of TupleOffset bytes. The number of bytes copied is returned in TupleDataLen. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_OUT_OF_RESOURCE Card Services was unable to set up a memory window to map the card's CIS. CS_NO_MORE_ITEMS The tuple does not contain any more data. TuppleOffset is greater than or equal to the length of the tuple. 3.4.3. ParseTuple #include "cistpl.h" int CardServices(ParseTuple, client_handle_t client, tuple_t *tuple, cisparse_t *parse); The cisparse_t data structure is given by: typedef union cisparse_t { cistpl_device_t device; cistpl_checksum_t checksum; cistpl_longlink_t longlink; cistpl_vers_1_t version_1; cistpl_altstr_t altstr; cistpl_jedec_t jedec; cistpl_manfid_t manfid; cistpl_funcid_t funcid; cistpl_config_t config; cistpl_cftable_entry_t cftable_entry; cistpl_device_geo_t device_geo; cistpl_vers_2_t version_2; cistpl_org_t org; } cisparse_t; ParseTuple interprets tuple data returned by a previous call to GetTupleData. The structure returned depends on the type of the parsed tuple. See the cistpl.h file for these structure definitions; some of them are quite complex. Return codes: CS_BAD_TUPLE An error was encounted during parsing of this tuple. The tuple may be incomplete, or may be formatted incorrectly. CS_UNSUPPORTED_FUNCTION ParseTuple cannot parse the specified tuple type. 3.4.4. ValidateCIS int CardServices(ValidateCIS, client_handle_t client, cisinfo_t *cisinfo); The cisinfo_t structure is given by: typedef struct cisinfo_t { u_long Chains; } cisinfo_t; ValidateCIS attempts to verify that a card has a reasonable Card Information Structure. It returns the number of tuples found in Chains. If the CIS appears to be uninterpretable, Chains will be set to 0. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_OUT_OF_RESOURCE Card Services was unable to set up a memory window to map the card's CIS. 3.5. Memory window control Each socket can have up to four active memory windows, mapping portions of PCMCIA memory into the host system address space. A PCMCIA device can address at most 16MB of both common and attribute memory. Windows should typically be sized to a power of two, and be aligned on a boundary that is a multiple of the window size in both the host and card address spaces. A memory window is initialized by a call to RequestWindow. Some window attributes can be modified using ModifyWindow. The segment of card memory mapped to the window can be modified using MapMemPage. And windows are released with ReleaseWindow. Unlike almost all other Card Services subfunctions, the memory window functions normally act on window_handle_t handles, rather than client_handle_t handles. 3.5.1. RequestWindow int CardServices(RequestWindow, client_handle_t *handle, win_req_t *req); The win_req_t structure is given by: typedef struct win_req_t { u_long Attributes; caddr_t Base; u_long Size; u_long AccessSpeed; } win_req_t; RequestWindow maps a window of card memory into system memory. On entry, the handle parameter should point to a valid client handle. On return, this will be replaced by a window_handle_t handle that should be used in subsequent calls to ModifyWindow, MapMemPage, and ReleaseWindow. The following flags can be specified in Attributes: WIN_MEMORY_TYPE This field can be either WIN_MEMORY_TYPE_CM for common memory, or WIN_MEMORY_TYPE_AM for attribute memory. WIN_DATA_WIDTH Either WIN_DATA_WIDTH_16 for 16-bit accesses, or WIN_DATA_WIDTH_8 for 8-bit access. WIN_ENABLE If this is set, the window is turned on. WIN_USE_WAIT Specifies that the controller should observe the card's MWAIT signal. Base specifies the base address of the window in system memory. If NULL, Card Services will set Base to the first available window address. Size specifies the window size in bytes. AccessSpeed specifies the memory access speed, in nanoseconds. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_CARD The socket assigned to this client is currently vacant. CS_BAD_ATTRIBUTE An unsupported window attribute was requested. CS_OUT_OF_RESOURCE The maximum number of memory windows for this socket are already being used. CS_IN_USE RequestWindow was unable to find a free window of system memory. 3.5.2. ModifyWindow int CardServices(ModifyWindow, window_handle_t, modwin_t *); The modwin_t structure is given by: typedef struct modwin_t { u_long Attributes; u_long AccessSpeed; } modwin_t; ModifyWindow modifies the attributes of a window handle returned by a previous call to RequestWindow. The following attributes can be changed: WIN_MEMORY_TYPE This field can be either WIN_MEMORY_TYPE_CM for common memory, or WIN_MEMORY_TYPE_AM for attribute memory. WIN_DATA_WIDTH Either WIN_DATA_WIDTH_16 for 16-bit accesses, or WIN_DATA_WIDTH_8 for 8-bit access. WIN_ENABLE If this is set, the window is turned on. AccessSpeed gives the new memory access speed, in nanoseconds. Return codes: CS_BAD_HANDLE The window handle is invalid. 3.5.3. MapMemPage int CardServices(MapMemPage, window_handle_t, memreq_t *); The memreq_t structure is given by: typedef struct memreq_t { u_long CardOffset; page_t Page; } memreq_t; MapMemPage sets the address of card memory that is mapped to the base of a memory window to CardOffset. The window should have been created by a call to RequestWindow. The Page parameter is not implemented in this version and should be set to 0. Return codes: CS_BAD_HANDLE The window handle is invalid. CS_BAD_PAGE The Page value was non-zero. 3.5.4. ReleaseWindow int CardServices(ReleaseWindow, window_handle_t handle); ReleaseWindow releases a memory window previously allocated with RequestWindow. Return codes: CS_BAD_HANDLE The window handle is invalid. 3.6. Bulk Memory Services Bulk memory services provide a higher level interface for accessing memory regions than that provided by the memory window services. A client using bulk memory calls does not need to know anything about the underlying memory organization or access methods. The device- specific code is packaged into a special Card Services client called a Memory Technology Driver. 3.6.1. RegisterMTD int CardServices(RegisterMTD, client_handle_t handle, mtd_reg_t *reg); The mtd_reg_t data structure is given by: typedef union mtd_reg_t { u_long Attributes; u_long Offset; u_long MediaID; } mtd_reg_t; RegisterMTD informs Card Services that this client MTD will handle requests for a specified memory region. The Offset field specifies the starting address of the memory region. The following fields are defined in Attributes: REGION_TYPE Either REGION_TYPE_CM for common memory, or REGION_TYPE_AM for attribute memory. The MediaID field is recorded by Card Services, and will be passed to the MTD as part of any request that references this memory region. Once an MTD is bound to a memory region by a call to RegisterMTD, it will remain bound until the MTD calls DeregisterClient. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_BAD_OFFSET Either the offset does not match a valid memory region for this card, or another MTD has already registered for this region. 3.6.2. GetFirstRegion, GetNextRegion int CardServices(GetFirstRegion, client_handle_t handle, region_info_t *region); int CardServices(GetNextRegion, client_handle_t handle, region_info_t *region); The region_info_t data structure is given by: typedef union region_info_t { u_long Attributes; u_long CardOffset; u_long RegionSize; u_long AccessSpeed; u_long BlockSize; u_long PartMultiple; u_char JedecMfr, JedecInfo; memory_handle_t next; } region_info_t; GetFirstRegion and GetNextRegion summarize the information in a card's CISTPL_DEVICE, CISTPL_JEDEC, and CISTPL_DEVICE_GEO tuples. CardOffset gives the starting address of a region. RegionSize gives the length of the region in bytes. AccessSpeed gives the device's cycle time in nanoseconds. BlockSize gives the erase block size in bytes, and PartMultiple gives the minimum granularity of partitions on this device, in units of BlockSize. JedecMfr and JedecInfo give the JEDEC identification bytes for this region. The following fields are defined in Attributes: REGION_TYPE Either REGION_TYPE_CM for common memory, or REGION_TYPE_AM for attribute memory. When these calls are made by an MTD client, only regions that have been bound to this client through calls to BindMTD will be returned. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_NO_MORE_ITEMS No more memory regions are defined. 3.6.3. OpenMemory int CardServices(OpenMemory, client_handle_t *handle, open_mem_t *req); The open_mem_t structure is given by: typedef struct open_mem_t { u_long Attributes; u_long Offset; } open_mem_t; OpenMemory is used to obtain a handle for accessing a memory region via the other bulk memory services. The Offset field specifies the base address of the region to be accessed. If successful, the client handle argument is replaced by the new memory handle. The following fields are defined in Attributes: MEMORY_TYPE Either MEMORY_TYPE_CM for common memory, or MEMORY_TYPE_AM for attribute memory. MEMORY_EXCLUSIVE Specifies that this client should have exclusive access to this memory region. Return codes: CS_BAD_HANDLE The window handle is invalid. CS_BAD_OFFSET Either the offset does not specify a valid region, or the region does not have an associated MTD to service bulk memory requests. 3.6.4. CloseMemory int CardServices(CloseMemory, memory_handle_t handle); CloseMemory releases a memory handle returned by a previous call to OpenMemory. A client should release all memory handles before calling DeregisterClient. Return codes: CS_BAD_HANDLE The memory handle is invalid. 3.6.5. ReadMemory, WriteMemory int CardServices(ReadMemory, memory_handle_t handle mem_op_t *req, caddr_t buf); int CardServices(WriteMemory, memory_handle_t handle, mem_op_t *req, caddr_t buf); The mem_io_t structure is given by: typedef struct mem_op_t { u_long Attributes; u_long Offset; u_long Count; } mem_op_t; ReadMemory and WriteMemory read from and write to a card memory area defined by the specified memory handle, returned by a previous call to OpenMemory. The Offset field gives the offset of the operation from the start of the card memory region. The Count field gives the number of bytes to be transferred. The buf field points to a host memory buffer to be the destination for a ReadMemory operation, or the source for a WriteMemory operation. The following fields are defined in Attributes: MEM_OP_BUFFER Either MEM_OP_BUFFER_USER if the host buffer is in a user memory segment, or MEM_OP_BUFFER_KERNEL if the host buffer is in kernel memory. MEM_OP_DISABLE_ERASE Specifies that a card area should not be erased before it is written. MEM_OP_VERIFY Specifies verification of write operations. Return codes: CS_BAD_HANDLE The window handle is invalid. CS_BAD_OFFSET The specified card offset is beyond the end of the memory region. CS_BAD_SIZE The specified transfer size extends past the end of the memory region. 3.6.6. RegisterEraseQueue int CardServices(RegisterEraseQueue, client_handle_t *handle, eraseq_hdr_t *header); The eraseq_hdr_t structure is given by: typedef struct erase_queue_header_t { int QueueEntryCount; eraseq_entry_t *QueueEntryArray; } eraseq_hdr_t; This call registers a queue of erase requests with Card Services. An eraseq_handle_t handle will be returned in *handle. When this client calls CheckEraseQueue, Card Services will scan the queue and begin asynchronous processing of any new requests. The eraseq_entry_t structure is given by: typedef struct eraseq_entry_t { memory_handle_t Handle; u_char State; u_long Size; u_long Offset; u_long *Optional; } eraseq_entry_t; In an erase queue entry, the Header field should be a memory handle returned by a previous call to OpenMemory. The State field indicates the state of the erase request. The following values are defined: ERASE_QUEUED Set by the client to indicate that this is a new request. ERASE_IDLE Set by the client to indicate that this entry is not active. ERASE_PASSED Set by the MTD to indicate successful completion. ERASE_FAILED Set by the MTD to indicate that the erase failed. ERASE_MEDIA_WRPROT Indicates that the region is write protected. ERASE_NOT_ERASABLE Indicates that this region does not support erase operations. ERASE_BAD_OFFSET Indicates that the erase does not start on an erase block boundary. ERASE_BAD_SIZE Indicates that the requested erase size is not a multiple of the erase block size. ERASE_BAD_SOCKET Set by the MTD to indicate that there is no card present. Additionally, the macro ERASE_IN_PROGRESS() will return a true condition for values of State that indicate an erase is being processed. The Size field gives the size of the erase request in bytes. The Offset field gives the offset from the start of the region. The size and offset should be aligned to erase block boundaries. The Optional field is not used by Card Services and may be used by the client driver. Return codes: CS_BAD_HANDLE The client handle is invalid. 3.6.7. DeregisterEraseQueue int CardServices(DeregisterEraseQueue, eraseq_handle_t handle); DeregisterEraseQueue frees a queue previously registered by a call to RegisterEraseQueue. If there are any pending requests in the specified queue, the call will fail. Return codes: CS_BAD_HANDLE The erase queue handle is invalid. CS_BUSY The erase queue has erase requests pending. 3.6.8. CheckEraseQueue int CardServices(CheckEraseQueue, eraseq_handle_t handle); This call notifies Card Services that there are new erase requests in a queue previously registered with RegisterEraseQueue. Typically, a client will initially assign each erase queue entry the state value ERASE_IDLE. When new requests are added to the queue, the client will set their states to ERASE_QUEUED, and call CheckEraseQueue. When the client is notified of an erase completion event, it will check the state field to determine whether the request was successful. Return codes: CS_BAD_HANDLE The erase queue handle is invalid. 3.7. Miscellaneous calls 3.7.1. GetCardServicesInfo int CardServices(GetCardServicesInfo, servinfo_t *info); The servinfo_t structure is given by: typedef struct servinfo_t { char Signature[2]; u_long Count; u_long Revision; u_long CSLevel; char *VendorString; } servinfo_t; GetCardServicesInfo returns revision information about this version of Card Services. Signature is set to ``CS''. Count is set to the number of sockets currently configured. Revision is set to the revision level of the Card Services package, and CSLevel is set to the level of compliance with the PCMCIA standard. These are encoded as BCD numbers. VendorString is set to point to an RCS identification string. This call always succeeds. 3.7.2. AccessConfigurationRegister #include "cisreg.h" int CardServices(AccessConfigurationRegister, client_handle_t handle, conf_reg_t *reg); The conf_reg_t structure is given by: typedef struct conf_reg_t { u_long Action; off_t Offset; u_long Value; } conf_reg_t; The Action parameter can be one of the following: CS_READ Read the specified configuration register and return Value. CS_WRITE Write Value to the specified configuration register. AccessConfigurationRegister either reads or writes the one-byte CIS configuration register at offset Offset from the start of the config register area. It can only be used for a socket that has been configured with RequestConfiguration. The following values for Offset are defined in cistpl.h: CISREG_COR The Configuration Option Register. CISREG_CCSR The Card Configuration and Status Register. CISREG_PRR The Pin Replacement Register. CISREG_SCR The Socket and Copy Register. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_BAD_ARGS The specified Action is not supported. CS_CONFIGURATION_LOCKED This actually means that the configuration has not been locked. CS_OUT_OF_RESOURCE Card Services was unable to allocate a memory window to access the card's configuration registers. 3.7.3. AdjustResourceInfo int CardServices(AdjustResourceInfo, client_handle_t handle, adjust_t *adj); The adjust_t structure is given by: typedef struct adjust_t { u_long Action; u_long Resource; u_long Attributes; union { struct memory { caddr_t Base; u_long Size; } memory; struct io { ioaddr_t BasePort; ioaddr_t NumPorts; u_long IOAddrLines; } io; struct irq { u_long IRQ; } irq; } resource; } adjust_t; AdjustResourceInfo is used to tell Card Services what resources may or may not be allocated by PCMCIA devices. The normal Linux resource management systems (the *_region calls for IO ports, interrupt allocation) are respected by Card Services, but this call gives the user another level of control. The Action parameter can have the following values: ADD_MANAGED_RESOURCE Place the specified resource under Card Services control, so that it may be allocated by PCMCIA devices. REMOVE_MANAGED_RESOURCE Remove the specified resource from Card Services control. At initialization time, Card Services assumes that it can use all available interrupts, but IO ports and memory regions must be explicitly enabled with ADD_MANAGED_RESOURCE. The Resource parameter can have the following values: RES_MEMORY_RANGE Specifies a memory range resource, described by adj->resource.memory. RES_IO_RANGE Specifies an IO port resource, described by adj->resource.io. RES_IRQ Specifies an interrupt resource, described by adj->resource.irq. The following flags may be specified in Attributes: RES_RESERVED Indicates that the resource should be reserved for PCMCIA devices that specifically request it. The resource will not be allocated for a device that asks Card Services for any available location. This is not implemented yet. Return codes: CS_UNSUPPORTED_FUNCTION The specified Action or Resource is not supported. CS_BAD_BASE The specified IO address is out of range. CS_BAD_SIZE The specified memory or IO window size is out of range. CS_IN_USE The specified interrupt is currently allocated by a Card Services client. 3.7.4. ReportError int CardServices(ReportError, char *prefix, int func, int ret); ReportError generates a kernel error message given a Card Services function code and its return code, with an optional prefix string. For example: CardServices(ReportError, "serial_cs", RequestIO, CS_BAD_HANDLE); would generate the following message: serial_cs: RequestIO: Bad handle This call always succeeds. 4. Card Information Structure Definitions 4.1. CIS Tuple Definitions The Card Services ParseTuple function interprets raw CIS tuple data from a call to GetTupleData and returns the tuple contents in a form dependant on the tuple type. #include "cistpl.h" 4.1.1. CISTPL_CHECKSUM The cistpl_checksum_t structure is given by: typedef struct cistpl_checksum_t { u_short addr; u_short len; u_char sum; } cistpl_checksum_t; 4.1.2. CISTPL_LINKTARGET, CISTPL_NOLINK CISTPL_LONGLINK_A, CISTPL_LONGLINK_C, The cistpl_longlink_t structure is given by: typedef struct cistpl_longlink_t { u_long addr; } cistpl_longlink_t; These tuples are pointers to additional chains of CIS tuples, either in attribute or common memory. Each CIS tuple chain can have at most one long link. CISTPL_LONGLINK_A tuples point to attribute memory, and CISTPL_LONGLINK_C tuples point to common memory. The standard CIS chain starting at address 0 in attribute memory has an implied long link to address 0 in common memory. A CISTPL_NOLINK tuple can be used to cancel this default link. The first tuple of a chain pointed to by a long link must be a CISTPL_LINKTARGET. The CS tuple handling code will automatically follow long links and verify link targets; these tuples are normally invisible unless the TUPLE_RETURN_LINK attribute is specified in GetNextTuple. 4.1.3. CISTPL_DEVICE, CISTPL_DEVICE_A The cistpl_device_t structure is given by: typedef struct cistpl_device_t { int ndev; struct device_info { u_char type; u_char wp; u_long speed; u_long size; } dev[CISTPL_MAX_DEVICES]; } cistpl_device_t; The CISTPL_DEVICE tuple describes address regions in a card's common memory. The CISTPL_DEVICE_A tuple describes regions in attribute memory. The type flag indicates the type of memory device for this region. The wp flag indicates if this region is write protected. The speed field is in nanoseconds, and size is in bytes. Address regions are assumed to be ordered consecutively starting with address 0. The following device types are defined: CISTPL_DTYPE_NULL Specifies that there is no device, or a ``hole'' in the card address space. CISTPL_DTYPE_ROM Masked ROM CISTPL_DTYPE_OTPROM One-type programmable ROM. CISTPL_DTYPE_EPROM UV erasable PROM. CISTPL_DTYPE_EEPROM Electrically erasable PROM. CISTPL_DTYPE_FLASH Flash EPROM. CISTPL_DTYPE_SRAM Static or non-volatile RAM. CISTPL_DTYPE_DRAM Dynamic or volatile RAM. CISTPL_DTYPE_FUNCSPEC Specifies a function-specific device, such as a memory-mapped IO device or buffer, as opposed to general purpose storage. CISTPL_DTYPE_EXTEND Specifies an extended device type. This type is reserved for future use. 4.1.4. CISTPL_VERS_1 The cistpl_vers_1_t structure is given by: typedef struct cistpl_vers_1_t { u_char major; u_char minor; int ns; int ofs[CISTPL_VERS_1_MAX_PROD_STRINGS]; char str[254]; } cistpl_vers_1_t; The ns field specifies the number of product information strings in the tuple. The string data is contained in the str array. Each string is null terminated, and ofs gives the offset to the start of each string. 4.1.5. CISTPL_ALTSTR The cistpl_altstr_t structure is given by: typedef struct cistpl_altstr_t { int ns; int ofs[CISTPL_ALTSTR_MAX_STRINGS]; char str[254]; } cistpl_altstr_t; The ns field specifies the number of alternate language strings in the tuple. The string data is contained in the str array. Each string is null terminated, and ofs gives the offset to the start of each string. 4.1.6. CISTPL_JEDEC_C, CISTPL_JEDEC_A The cistpl_jedec_t structure is given by: typedef struct cistpl_jedec_t { int nid; struct jedec_id { u_char mfr; u_char info; } id[CISTPL_MAX_DEVICES]; } cistpl_jedec_t; The nid field specifies the number of JEDEC identifiers in the tuple. There should be a one-to-one correspondence between JEDEC identifiers and device descriptions in the corresponding CISTPL_DEVICE tuple. 4.1.7. CISTPL_CONFIG The cistpl_config_t structure is given by: typedef struct cistpl_config_t { u_char last_idx; u_long base; u_long rmask[4]; u_char subtuples; } cistpl_config_t; The last_idx field gives the index of the highest numbered configuration table entry. The base field gives the offset of a card's configuration registers in attribute memory. The rmask array is a series of bit masks indicating which configuration registers are present. The subtuples field gives the number of bytes of subtuples following the normal tuple contents. 4.1.8. CISTPL_CFTABLE_ENTRY The cistpl_cftable_entry_t structure is given by: typedef struct cistpl_cftable_entry_t { u_char index; u_char flags; u_char interface; cistpl_power_t vcc, vpp1, vpp2; cistpl_timing_t timing; cistpl_io_t io; cistpl_irq_t irq; cistpl_mem_t mem; u_char subtuples; } cistpl_cftable_entry_t; A CISTPL_CFTABLE_ENTRY structure describes a complete operating mode for a card. Many sections are optional. The index field gives the configuration index for this operating mode; writing this value to the card's Configuration Option Register selects this mode. The following fields are defined in flags: CISTPL_CFTABLE_DEFAULT Specifies that this is the default configuration table entry. CISTPL_CFTABLE_BVDS Specifies that this configuration implements the BVD1 and BVD2 signals in the Pin Replacement Register. CISTPL_CFTABLE_WP Specifies that this configuration implements the write protect signal in the Pin Replacement Register. CISTPL_CFTABLE_RDYBSY Specifies that this configuration implements the Ready/Busy signal in the Pin Replacement Register. CISTPL_CFTABLE_MWAIT Specifies that the WAIT signal should be observed during memory access cycles. CISTPL_CFTABLE_AUDIO Specifies that this configuration generates an audio signal that can be routed to the host system speaker. CISTPL_CFTABLE_READONLY Specifies that the card has a memory region that is read-only in this configuration. CISTPL_CFTABLE_PWRDOWN Specifies that this configuration supports a power down mode, via the Card Configuration and Status Register. The cistpl_power_t structure is given by: typedef struct cistpl_power_t { u_char present; u_char flags; u_long param[7]; } cistpl_power_t; The present field is bit mapped and indicates which parameters are present for this power signal. The following indices are defined: CISTPL_POWER_VNOM The nominal supply voltage. CISTPL_POWER_VMIN The minimum supply voltage. CISTPL_POWER_VMAX The maximum supply voltage. CISTPL_POWER_ISTATIC The continuous supply current required. CISTPL_POWER_IAVG The maximum current averaged over one second. CISTPL_POWER_IPEAK The maximum current averaged over 10 ms. CISTPL_POWER_IDOWN The current required in power down mode. Voltages are given in units of 10 microvolts. Currents are given in units of 100 nanoamperes. The cistpl_timing_t structure is given by: typedef cistpl_timing_t { u_long wait, waitscale; u_long ready, rdyscale; u_long reserved, rsvscale; } cistpl_timing_t; Each time consists of a base time in nanoseconds, and a scale multiplier. Unspecified times have values of 0. The cistpl_io_t structure is given by: typedef struct cistpl_io_t { u_char flags; int nwin; struct { u_long base; u_long len; } win[CISTPL_IO_MAX_WIN; } cistpl_io_t; The number of IO windows is given by nwin. Each window is described by a base address, base, and a length in bytes, len. The following bit fields are defined in flags: CISTPL_IO_LINES_MASK The number of IO lines decoded by this card. CISTPL_IO_8BIT Indicates that the card supports split 8-bit accesses to 16-bit IO registers. CISTPL_IO_16BIT Indicates that the card supports full 16-bit accesses to IO registers. The cistpl_irq_t structure is given by: typedef struct cistpl_irq_t { u_long IRQInfo1; u_long IRQInfo2; } cistpl_irq_t; If IRQInfo1 is 0, then no interrupt information is available. IRQInfo2 is a bit mask of allowed interrupt request numbers. The cistpl_mem_t structure is given by: typedef struct cistpl_mem_t { u_char nwin; struct { u_long len; u_long card_addr; caddr_t host_addr; } win[CISTPL_MEM_MAX_WIN; } cistpl_mem_t; The number of memory windows is given by nwin. Each window is described by an address in the card memory space, card_addr, an address in the host memory space, host_addr, and a length in bytes, len. If the host address is 0, the position of the window is arbitrary. 4.1.9. CISTPL_MANFID The cistpl_manfid_t structure is given by: typedef struct cistpl_manfid_t { u_short manf; u_short card; } cistpl_manfid_t; The manf field identifies the card manufacturer. The card field is chosen by the vendor and should identify the card type and model. 4.1.10. CISTPL_FUNCID The cistpl_funcid_t structure is given by: typedef struct cistpl_funcid_t { u_char func; u_char sysinit; } cistpl_funcid_t; The func field identifies the card function. The sysinit field contains several bit-mapped flags describing how the card should be configured at boot time. 4.1.11. CISTPL_DEVICE_GEO The cistpl_device_geo_t structure is given by: typedef struct cistpl_device_geo_t { int ngeo; struct geo { u_char buswidth; u_long erase_block; u_long read_block; u_long write_block; u_long partition; u_long interleave; } geo[CISTPL_MAX_DEVICES]; } cistpl_device_geo_t; The erase_block, read_block, and write_block sizes are in units of buswidth bytes times interleave. The partition size is in units of erase_block. 4.1.12. CISTPL_VERS_2 The cistpl_vers_2_t structure is given by: typedef struct cistpl_vers_2_t { u_char vers; u_char comply; u_short dindex; u_char vspec8, vspec9; u_char nhdr; int vendor, info; char str[244]; } cistpl_vers_2_t; The vspec8 and vspec9 fields may contain vendor-specific information. The nhdr field gives the number of copies of the CIS that are present on this card. The str array contains two strings: a vendor name, and an informational message describing the card. The offset of the vendor string is given by vendor, and the offset of the product info string is in info. 4.1.13. CISTPL_ORG The cistpl_org_t structure is given by: typedef struct cistpl_org_t { u_char data_org; char desc[30]; 4.2. CIS configuration registers The PCMCIA standard defines a few standard configuration registers located in a card's attribute memory space. A card's CONFIG tuple specifies which of these registers are implemented. #include "cisreg.h" 4.2.1. Configuration Option Register This register should be present for virtually all IO cards. Writing to this register selects a configuration table entry and enables a card's IO functions. The following bit fields are defined: COR_CONFIG_MASK Specifies the configuration table index describing the card's current operating mode. COR_LEVEL_REQ Specifies that the card should generate level mode (edge- triggered) interrupts, the default. COR_SOFT_RESET Setting this bit performs a ``soft'' reset operation. Drivers should use the ResetCard call to reset a card, rather than writing directly to this register. 4.2.2. Card Configuration and Status Register The following bit fields are defined: CCSR_INTR_PENDING Signals that the card is currently asserting an interrupt request. This signal can be used to support interrupt sharing. CCSR_POWER_DOWN Setting this bit signals that the card should enter a power down state. CCSR_AUDIO_ENA Specifies that the card's audio output should be enabled. CCSR_IOIS8 This is used by the host to indicate that it can only perform 8-bit IO operations and that 16-bit accesses will be carried out as two 8-bit accesses. CCSR_SIGCHG_ENA This indicates to the card that it should use the SIGCHG signal to indicate changes in the WP, READY, BVD1, and BVD2 signals. CCSR_CHANGED This bit signals to the host that one of the signals in the Pin Replacement Register has changed state. 4.2.3. Pin Replacement Register Signals in this register replace signals that are not available when a socket is operating in memory and IO mode. An IO card will normally assert the SIGCHG signal to indicate that one of these signals has changed state, then a driver can poll this register to find out specifically what happened. The following bit fields are defined: PRR_WP_STATUS The current state of the write protect signal. PRR_READY_STATUS The current state of the ready signal. PRR_BVD2_STATUS The current state of the battery warn signal. PRR_BVD1_STATUS The current state of the battery dead signal. PRR_WP_EVENT Indicates that the write protect signal has changed state since the PRR register was last read. PRR_READY_EVENT Indicates that the ready signal has changed state since the PRR register was last read. PRR_BVD2_EVENT Indicates that the battery warn signal has changed state since the PRR register was last read. PRR_BVD1_EVENT Indicates that the battery dead signal has changed state since the PRR register was last read. This register can also be written. In this case, the STATUS bits act as a mask; if a STATUS bit is set, the corresponding EVENT bit is updated by the write. 4.2.4. Socket and Copy Register This register is used when several identical cards may be set up to share the same range of IO ports, to emulate an ISA bus card that would control several devices. For example, an ISA hard drive controller might control several drives, selectable by writing a drive number to an IO port. For several PCMCIA drives to emulate this controller interface, each needs to ``know'' which drive it is, so that it can identify which IO operations are intended for it. The following bit fields are defined: SCR_SOCKET_NUM This should indicate the socket number in which the card is located. SCR_COPY_NUM If several identical cards are installed in a system, this field should be set to a unique number identifying which of the identical cards this is. 5. Card Services Event Handling Card Services events have several sources: o Card status changes reported by the low-level socket drivers. o Artificial events generated by Card Services itself. o Advanced Power Management (APM) events. o Events generated by other Card Services clients. Socket driver events may be either interrupt-driven or polled. 5.1. Event handler operations When Card Services recognizes that an event has occurred, it checks the event mask of each client to determine which clients should receive an event notification. When a client registers with Card Services, it specifies an event handler callback function. This handler should have the form: int (*event_handler)(event_t event, int priority, event_callback_args_t *args); The priority parameter is set to either CS_EVENT_PRI_LOW for ordinary events, or CS_EVENT_PRI_HIGH for events that require an immediate response. The only high priority event is CS_EVENT_CARD_REMOVAL. A client event handler should process this event as efficiently as possible so that Card Services can quickly notify other clients. The event_callback_args_t structure is given by: typedef struct event_callback_args_t { client_handle_t client_handle; void *info; void *mtdrequest; void *buffer; void *misc; void *client_data; } event_callback_args_t; The client_handle member is set to the handle of the client whose socket was responsible for the event. This is useful if a driver is bound to several sockets. The info field is currently only used to return an exit status from a call to ResetCard. The client_data field may be used by a driver to point to a local data structure associated with this device. The remaining fields are currently unused. 5.2. Event descriptions CS_EVENT_CARD_INSERTION This event signals that a card has been inserted. If a driver is bound to an already occupied socket, Card Services will send the driver an artificial insertion event. CS_EVENT_CARD_REMOVAL This event signals that a card has been removed. This event should be handled with minimum delay so that Card Services can notify all clients as quickly as possible. CS_EVENT_BATTERY_LOW This event signals a change of state of the ``battery low'' signal. CS_EVENT_BATTERY_DEAD This event signals a change of state of the ``battery dead'' signal. CS_EVENT_READY_CHANGE This event signals a change of state of the ``ready'' signal. CS_EVENT_WRITE_PROTECT This event signals a change of state of the ``write protect'' signal. CS_EVENT_REGISTRATION_COMPLETE This event is sent to a driver after a successful call to RegisterClient. CS_EVENT_RESET_REQUEST This event is sent when a client calls ResetCard. An event handler can veto the reset operation by returning failure. CS_EVENT_RESET_PHYSICAL This is sent to all clients just before a reset signal is sent to a card. CS_EVENT_CARD_RESET This event signals that a reset operation is finished. The success or failure of the reset should be determined using GetStatus. CS_EVENT_RESET_COMPLETE This event is sent to a client that has called ResetCard to signal the end of reset processing. CS_EVENT_PM_SUSPEND This event signals that Card Services has received either a user initiated or APM suspend request. An event handler can veto the suspend by returning failure. CS_EVENT_PM_RESUME This signals that the system is back on line after a suspend/resume cycle. CS_EVENT_MTD_REQUEST This is used to initiate an MTD memory operation. CS_EVENT_ERASE_COMPLETE This is used to signal a client that a queued erase operation has completed. 5.3. Client driver event handling responsibilities A client driver should respond to CS_EVENT_CARD_INSERTION and CS_EVENT_CARD_REMOVAL events by configuring and un-configuring the socket. Because card removal is a high priority event, the driver should immediately block IO to the socket, perhaps by setting a flag in a device structure, and schedule all other shutdown processing to happen later using a timer interrupt. When a CS_EVENT_PM_RESET_REQUEST event is received, a driver should block IO and release a locked socket configuration. When a CS_EVENT_CARD_RESET is received, a driver should restore the socket configuration and unblock IO. A CS_EVENT_PM_SUSPEND event should be handled somewhat like a CS_EVENT_PM_RESET_REQUEST event, in that IO should be blocked and the socket configuration should be released. When a CS_EVENT_PM_RESUME event is received, a driver can expect a card to be ready to be reconfigured, similar to when a CS_EVENT_CARD_RESET event is received. 6. Memory Technology Drivers A Memory Technology Driver (``MTD'') is used by Card Services to implement bulk memory services for a particular type of memory device. An MTD should register as a normal Card Services client with a call to RegisterClient. When it receives a card insertion event, it should use GetFirstRegion and GetNextRegion to identify memory regions that it will administer. Then, it should use RegisterMTD to take control of these regions. MTD read, write, copy, and erase requests are packaged into CS_EVENT_MTD_REQUEST events by Card Services, and passed to the MTD's event handler for processing. 6.1. MTD request handling An MTD receives requests from Card Services in the form of CS_EVENT_MTD_REQUEST events. Card Services passes a description of the request in the mtdrequest field of the event callback arguments. For requests that transfer data to or from the host, the host buffer address is passed in the buffer field. The mtd_request_t structure is given by: typedef struct mtd_request_t { u_long SrcCardOffset; u_long DestCardOffset; u_long TransferLength; u_long Function; u_long MediaID; u_long Status; u_long Timeout; } mtd_request_t; The Function field is bit mapped and describes the action to be performed by this request: MTD_REQ_ACTION Either MTD_REQ_ERASE, MTD_REQ_READ, MTD_REQ_WRITE, or MTD_REQ_COPY. MTD_REQ_NOERASE For a write command that is sized and aligned on erase block boundaries, this specifies that no erase should be performed. MTD_REQ_VERIFY Specifies that writes should be verified. MTD_REQ_READY Indicates that this request is a retry of a previously request that was delayed until the card asserted READY. MTD_REQ_TIMEOUT Indicates that this request is a retry of a previously request that was delayed by a timeout. MTD_REQ_FIRST Indicates that this request is the first in a series of requests. MTD_REQ_LAST Indicates that this request is the last of a series of requests. MTD_REQ_KERNEL Indicates that the host buffer for a read or write command is located in kernel memory, as opposed to user memory. The MediaID field is the value specified in the RegisterMTD request for this region. The Status field is used by the MTD when it is unable to satisfy a request because a device is busy. MTD requests normally run without blocking. If an MTD request would block, it should return an error code of CS_BUSY, and set Status to one of the have the following values: MTD_WAITREQ Specifies that the request should be retried after another MTD request currently in progress completes. MTD_WAITTIMER Specifies that the request should be continued after the time specified in the timeout field. MTD_WAITRDY Specifies that the request should be continued when the card signals READY, or when the time specified in Timeout elapses, whichever happens first. MTD_WAITPOWER Specifies that the request should be retried after something happens that affects power availability to the socket. For MTD_WAITTIMER and MTD_WAITRDY, the Timeout field will specify the timeout interval in milliseconds. 6.2. MTD helper functions Since an MTD processes requests generated by Card Services, there may be some restrictions on the sorts of Card Services calls that can be safely made from the MTD event handler. The MTD helper functions provide a limited set of special services that may be needed by an MTD but would be tricky to implement using the normal Card Services calls. In the Linux PCMCIA implementation, most CS calls can be safely made from an MTD event handler, but the MTD helper interface is included for compatibility. #include "cs_types.h" #include "cs.h" #include "bulkmem.h" int MTDHelperEntry(int subfunc, void *arg1, void *arg2); 6.2.1. MTDModifyWindow int MTDHelperEntry(MTDModifyWindow, memory_handle_t handle, mtd_mod_req_t *mod); The mtd_mod_req_t structure is give by: typedef struct mtd_mod_req_t { u_long Attributes; u_long AccessSpeed; u_long CardOffset; } mtd_mod_req_t; MTDModifyWindow is essentially equivalent to using the normal ModifyWindow and MapMemPage calls. The following flags can be specified in Attributes: WIN_MEMORY_TYPE Either WIN_MEMORY_TYPE_CM for common memory, or WIN_MEMORY_TYPE_AM for attribute memory. WIN_USE_WAIT Specifies that the controller should observe the card's MWAIT signal. A window configured with MTDModifyWindow will always be enabled, and have a 16 bit data width. Return codes: CS_BAD_HANDLE The memory handle is invalid. 6.2.2. MTDSetVpp int MTDHelperEntry(MTDSetVpp, client_handle_t client, mtd_vpp_req_t *req); typedef struct mtd_vpp_req_t { u_char Vpp1, Vpp2; } mtd_vpp_req_t; MTDSetVpp changes the programming voltage for a socket. Vpp1 and Vpp2 should be given in units of 1/10 volt. Currently, Vpp1 should always equal Vpp2. Return codes: CS_BAD_HANDLE The client handle is invalid. CS_BAD_VPP The specified Vpp is not available, or Vpp1 does not equal Vpp2. 6.2.3. MTDRDYMask int MTDHelperEntry(MTDRDYMask, client_handle_t client, mtd_rdy_req_t *req); typedef struct mtd_rdy_req_t { u_long Mask; } mtd_rdy_req_t; MTDRDYMask selects whether or not CS_EVENT_READY_CHANGE events will be enabled. The client should already have indicated to Card Services that it should receive ready change events, via a call to either RegisterClient or SetEventMask. Ready change events will be enabled if the CS_EVENT_READY_CHANGE bit is set in the Mask argument. Return codes: CS_BAD_HANDLE The client handle is invalid. 7. Driver Services interface Driver Services provides a link between Card Services client drivers and user mode utilities like the cardmgr daemon. It is a sort of Card Services ``super-client''. Driver Services uses the BindDevice function to link other client drivers with their corresponding PCMCIA cards. Unlike other clients, Driver Services remains permanently bound to all sockets as cards are inserted and removed. 7.1. Interface to other client drivers Driver Services keeps track of all client drivers that are installed and ready to attach to a socket. Client drivers need to have entry points for creating and deleting device ``instances'', where one device instance is everything needed to manage one PCMCIA card. Each client driver is identified by a unique 16-character tag that has the special type dev_info_t, defined in cs_types.h. Each device instance is described by a dev_link_t structure. 7.1.1. The dev_link_t structure The dev_link_t data structure is given by: #include "ds.h" typedef struct dev_link_t { char dev_name[8]; u_char major, minor; u_long state; u_long open; struct wait_queue *pending struct timer_list release client_handle_t handle; io_req_t io; irq_req_t irq; config_req_t conf; window_handle_t win; void *priv; struct dev_link_t *next; } dev_link_t; The dev_name field should be filled in by the driver with a device file name for accessing this device, if appropriate. For example, the serial_cs driver uses names like ``cua1''. The major and minor fields give major and minor device numbers for accessing this device. Driver Services relays these fields to user mode programs via the DS_GET_DEVICE_INFO ioctl. The state field should be used to keep track of the current device state. The following flags are defined: DEV_PRESENT Indicates that the card is present. This bit should be set and cleared by the driver's event handler in response to card insertion and removal events. DEV_CONFIG Indicates that the card is configured for use. DEV_CONFIG_PENDING Indicates that configuration is in progress. DEV_SUSPEND Indicates that the card is suspended. DEV_BUSY Indicates that an IO operation is in progress. This bit may be used as an interlock to prevent access conflicts. DEV_STALE_CONFIG For some drivers, when a running card is ejected, the socket should not be unconfigured until any devices corresponding to this card are closed. This flag indicates that the socket should be unconfigured when the device is closed. DEV_STALE_LINK A driver instance should not be deleted until all its PCMCIA resources are released. This flag indicates that this driver instance should be freed as soon as the socket is unconfigured. The open field is a usage count for this device. The device should only be freed when the open count is zero. The pending field can be used to manage a queue of processes waiting to use the device. The release field is used to schedule device shutdown processing when a card is ejected. A card removal event needs to be handled at high priority, so a driver's event handler will typically deal with an eject by resetting the DEV_PRESENT bit in the device state, then scheduling the shutdown processing to run at a later time. The handle, io, irq, conf, and win fields comprise all the normal PCMCIA data structures needed to configure one PCMCIA IO card. The priv field can be used for any sort of private data structure needed to manage the device. The next field can be used to build linked lists of dev_link_t structures, for drivers that can handle multiple instances. 7.1.2. register_pcmcia_driver int register_pcmcia_driver(dev_info_t *dev_info, dev_link_t *(*attach)(void), void (*detach)(dev_link_t *)); register_pcmcia_driver informs Driver Services that a client driver is present and ready to be bound to sockets. When Driver Services receives a DS_BIND_REQUEST ioctl that matches this driver's dev_info string, it will call the driver's attach() entry point. When it gets a DS_UNBIND_REQUEST ioctl, it will call detach(). 7.1.3. unregister_pcmcia_driver int unregister_pcmcia_driver(dev_info_t *dev_info); This informs Driver Services that it should no longer bind sockets to the specified client driver. 7.2. Interface to user mode PCMCIA utilities Driver Services creates a pseudo-device for communicating with user mode PCMCIA utilities. The major number of the device is chosen dynamically, and PCMCIA utilities should read /proc/devices to determine it. Minor device numbers correspond to socket numbers, starting with 0. Only one process is allowed to open a socket for read/write access. Other processes can open the socket in read-only mode. A read-only connection to Driver Services can perform a subset of ioctl calls. A read/write connection can issue all ioctl calls, and can also receive Card Services event notifications. 7.2.1. Card Services event notifications Driver Services implements read() and select() functions for event notification. Reading from a pcmcia device returns an unsigned long value containing all the events received by Driver Services since the previous read(). If no events have been received, the call will block until the next event. A select() call can be used to monitor several sockets for new events. The following events are monitored by Driver Services: CS_EVENT_CARD_INSERTION, CS_EVENT_CARD_REMOVAL, CS_EVENT_RESET_PHYSICAL, CS_EVENT_CARD_RESET, and CS_EVENT_RESET_COMPLETE. 7.2.2. ioctl descriptions Most Driver Services ioctl operations directly map to Card Services functions. An ioctl call has the form: int ioctl(int fd, int cmd, ds_ioctl_arg_t *arg); The ds_ioctl_arg_t structure is given by: typedef union ds_ioctl_arg_t { servinfo_t servinfo; adjust_t adjust; config_t config; tuple_t tuple; tuple_parse_t tuple_parse; client_req_t client_req; status_t status; conf_reg_t conf_reg; cisinfo_t cisinfo; region_info_t region; bind_info_t bind_info; mtd_info_t mtd_info; } ds_ioctl_arg_t; The following ioctl commands execute the corresponding Card Services function: DS_GET_CARD_SERVICES_INFO Calls CardServices(GetCardServicesInfo, ..., &arg->servinfo). DS_ADJUST_RESOURCE_INFO Calls CardServices(AdjustResourceInfo, ..., &arg->adjust). DS_GET_CONFIGURATION_INFO Calls CardServices(GetConfigurationInfo, ..., &arg->config). DS_GET_FIRST_TUPLE Calls CardServices(GetFirstTuple, ..., &arg->tuple). DS_GET_NEXT_TUPLE Calls CardServices(GetNextTuple, ..., &arg->tuple). DS_GET_TUPLE_DATA Calls CardServices(GetTupleData, ..., &arg->tuple_parse.tuple). The tuple data is returned in arg->tuple_parse.data. DS_PARSE_TUPLE Calls CardServices(ParseTuple, ..., &arg->tuple_parse.tuple, &arg->tuple_parse.parse). DS_RESET_CARD Calls CardServices(ResetCard, ...). DS_GET_STATUS Calls CardServices(GetStatus, ..., &arg->status). DS_ACCESS_CONFIGURATION_REGISTER Calls CardServices(AccessConfigurationRegister, ..., &arg->conf_reg). DS_VALIDATE_CIS Calls CardServices(ValidateCIS, ..., &arg->cisinfo). DS_SUSPEND_CARD Calls CardServices(SuspendCard, ...). DS_RESUME_CARD Calls CardServices(ResumeCard, ...). DS_EJECT_CARD Calls CardServices(EjectCard, ...). DS_INSERT_CARD Calls CardServices(InsertCard, ...). DS_GET_FIRST_REGION Calls CardServices(GetFirstRegion, ..., &arg->region). DS_GET_NEXT_REGION Calls CardServices(GetNextRegion, ..., &arg->region). The following ioctl commands invoke special Driver Services functions. They act on bind_info_t structures: typedef struct bind_info_t { dev_info_t dev_info; struct dev_info_t *instance; char name[8]; u_char major, minor; void *next; } bind_info_t; DS_BIND_REQUEST This call connects a socket to a client driver. The specified device ID dev_info is looked up in the list of registered drivers. If found, the driver is bound to this socket using the BindDevice call. Then, Driver Services calls the client driver's attach() entry point to create a device instance. The new dev_link_t pointer is returned in instance. DS_GET_DEVICE_INFO This call retrieves the dev_name, major, and minor entries from the dev_link_t structure pointed to by instance. DS_UNBIND_REQUEST This call calls the detach() function for the specified driver and instance, shutting down this device. Finally, the DS_BIND_MTD request takes an argument of type mtd_info_t: typedef struct mtd_info_t { dev_info_t dev_info; u_long Attributes; u_long CardOffset; This call associates an MTD identified by dev_info with a memory region described by Attributes and CardOffset, which have the same meanings as in the Card Services BindMTD call. 8. Anatomy of a Card Services Client Driver Each release of the Linux PCMCIA package comes with a well-commented ``skeleton'' client driver that should be used as a starting point for writing a new driver. Look for it in modules/skeleton.c. 8.1. Module initialization and cleanup All loadable modules must supply init_module() and cleanup_module() functions, which are invoked by the module support code when the module is installed and removed. A PCMCIA client driver's init function should register the driver with Driver Services, via the register_pcmcia_driver() call. The cleanup function should use unregister_pcmcia_driver() to unregister with Driver Services. Depending on the driver, the cleanup function may also need to free any device structures that still exist at shutdown time. 8.2. The *_attach() and *_detach() functions The *_attach() entry point is responsible for creating an ``instance'' of the driver, setting up any data structures needed to manage one PCMCIA card. The *_attach() function should allocate and initialize a dev_link_t structure, and call RegisterClient to establish a link with Card Services. It returns a pointer to the new dev_link_t structure, or NULL if the new instance could not be created. The *_detach() entry point deletes a driver instance created by a previous call to *_attach. It also breaks the link with Card Services, using DeregisterClient. The *_attach() entry point is called by Driver Services when a card has been successfully identified and mapped to a matching driver by a DS_BIND_REQUEST ioctl(). The *_detach() entry point is called in response to a DS_UNBIND_REQUEST ioctl() call. 8.3. The *_config() and *_release() functions The *_config() function is called to prepare a card for IO. Most drivers read some configuration details from the card itsef, but most have at least some built-in knowledge of how the device should be set up. For example, the serial card driver reads a card's CFTABLE_ENTRY tuples to determine appropriate IO port base addresses and corresponding configuration indices, but the driver ignores the interrupt information in the CIS. The *_config function will parse relevant parts of a card's CIS, then make calls to RequestIO, RequestIRQ, and/or RequestWindow, then call RequestConfiguration. When a card is successfully configured, the *_config() routine should fill in the dev_name, major, and minor fields in the dev_link_t structure. These fields will be returned by Driver Services in response to a DS_GET_DEVICE_INFO ioctl(). The *_release() function should release any resource allocated by a previous call to *_config(), and blank out the device's dev_name field. The *_config() and *_release functions are normally called in response to card status change events or from timer interrupts. Thus, they cannot sleep, and should not call other kernel functions that might block. 8.4. The PCMCIA event handler The *_event() entry point is called from Card Services to notify a driver of card status change events. 8.5. Locking and synchronization issues A configured socket should only be released when all associated devices are closed. Releasing a socket allows its system resources to be allocated for use by another device. If the released resources are reallocated while IO to the original device is still in progress, the original driver may interfere with use of the new device. A driver instance should only be freed after its corresponding socket configuration has been released. Card Services requires that a client explicitly release any allocated resources before a call to DeregisterClient will succeed. All loadable modules have a ``use count'' that is used by the system to determine when it is safe to unload a module. The convention in the PCMCIA client drivers is to increment the use count when a device is opened, and to decrement the count when a device is closed. So, a driver can be unloaded whenever all associated devices are closed. in particular, a driver can be unloaded even if it is still bound to a socket, and the module cleanup code needs to be able to appropriately free any such resources that are still allocated. This should always be safe, because if the driver has a use count of zero, all devices are closed, which means all active sockets can be released, and all device instances can be detached. If a driver's *_release() function is called while a device is still open, it should set the DEV_STALE_CONFIG flag in the device state, to signal that the device should be released when the driver's close() function is called. If *_detach() is called for a configured device, the DEV_STALE_LINK flag should be set to signal that the instance should be detached when the *_release() function is called. 8.6. Using existing Linux drivers to access PCMCIA devices Many of the current PCMCIA client drivers use existing Linux driver code to perform device IO operations. The Card Services client module handles card configuration and responds to card status change events, but delegates device IO to a compatible driver for a conventional ISA bus card. In some cases, a conventional driver can be used without modification. However, to fully support PCMCIA features like hot swapping and power management, there needs to be some communication between the PCMCIA client code and the device IO code. Most Linux drivers expect to probe for devices at boot time, and are not designed to handle adding and removing devices. One side-effect of the move towards driver modularization is that it is usually easier to adapt a modularized driver to handle removable devices. It is important that a device driver be able to recover from having a device disappear at an inappropriate time. At best, the driver should check for device presence before attempting any IO operation or before handling an IO interrupt. Loops that check device status should have timeouts so they will eventually exit if a device never responds. 9. Where to Go for More Information The Linux Kernel Hackers' Guide, written by Michael Johnson, is a good source of general information about writing Linux device drivers. It is available from the usual Linux FTP sites. The genuine PCMCIA standard is only available for a steep price from the PCMCIA association itself. The new PC Card Standard is supposed to cost $399, but you may still be able to get old copies of the 2.1 standard at a deep discount. Personal Computer Memory Card International Association 1030 East Duane Avenue, Suite G Sunnyvale, CA 94086 USA (408) 720-0107, (408) 720-9416 FAX, (408) 720-9388 BBS One alternative is the PCMCIA Developer's Guide, written by Michael Mori, available from Sycard Technology for approximately $89.95: Sycard Technology 1180-F Miraloma Way Sunnyvale, CA 94086 USA (408) 749-0130, (408) 749-1323 FAX Larry Levine has written a more general introduction to PCMCIA called the PCMCIA Primer, which should be on bookstore shelves in the summer of 1995. It is published by M & T Books, ISBN 1-55828-437-0. Programming information for various PCMCIA controllers is available from the corresponding chip vendors: Intel Corporation (800) 628-8686 Cirrus Logic (510) 623-8300 Vadem (408) 943-9301 Databook Inc. (716) 889-4204