1 How To Write Linux PCI Drivers
3 by Martin Mares <mj@ucw.cz> on 07-Feb-2000
5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
6 The world of PCI is vast and it's full of (mostly unpleasant) surprises.
7 Different PCI devices have different requirements and different bugs --
8 because of this, the PCI support layer in Linux kernel is not as trivial
9 as one would wish. This short pamphlet tries to help all potential driver
10 authors find their way through the deep forests of PCI handling.
13 0. Structure of PCI drivers
14 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
15 There exist two kinds of PCI drivers: new-style ones (which leave most of
16 probing for devices to the PCI layer and support online insertion and removal
17 of devices [thus supporting PCI, hot-pluggable PCI and CardBus in a single
18 driver]) and old-style ones which just do all the probing themselves. Unless
19 you have a very good reason to do so, please don't use the old way of probing
20 in any new code. After the driver finds the devices it wishes to operate
21 on (either the old or the new way), it needs to perform the following steps:
24 Access device configuration space
25 Discover resources (addresses and IRQ numbers) provided by the device
26 Allocate these resources
27 Communicate with the device
30 Most of these topics are covered by the following sections, for the rest
31 look at <linux/pci.h>, it's hopefully well commented.
33 If the PCI subsystem is not configured (CONFIG_PCI is not set), most of
34 the functions described below are defined as inline functions either completely
35 empty or just returning an appropriate error codes to avoid lots of ifdefs
41 The new-style drivers just call pci_register_driver during their initialization
42 with a pointer to a structure describing the driver (struct pci_driver) which
45 name Name of the driver
46 id_table Pointer to table of device ID's the driver is
47 interested in. Most drivers should export this
48 table using MODULE_DEVICE_TABLE(pci,...).
49 probe Pointer to a probing function which gets called (during
50 execution of pci_register_driver for already existing
51 devices or later if a new device gets inserted) for all
52 PCI devices which match the ID table and are not handled
53 by the other drivers yet. This function gets passed a
54 pointer to the pci_dev structure representing the device
55 and also which entry in the ID table did the device
56 match. It returns zero when the driver has accepted the
57 device or an error code (negative number) otherwise.
58 This function always gets called from process context,
60 remove Pointer to a function which gets called whenever a
61 device being handled by this driver is removed (either
62 during deregistration of the driver or when it's
63 manually pulled out of a hot-pluggable slot). This
64 function always gets called from process context, so it
66 save_state Save a device's state before it's suspend.
67 suspend Put device into low power state.
68 resume Wake device from low power state.
69 enable_wake Enable device to generate wake events from a low power
72 (Please see Documentation/power/pci.txt for descriptions
73 of PCI Power Management and the related functions)
75 The ID table is an array of struct pci_device_id ending with a all-zero entry.
76 Each entry consists of:
78 vendor, device Vendor and device ID to match (or PCI_ANY_ID)
79 subvendor, Subsystem vendor and device ID to match (or PCI_ANY_ID)
81 class, Device class to match. The class_mask tells which bits
82 class_mask of the class are honored during the comparison.
83 driver_data Data private to the driver.
85 Most drivers don't need to use the driver_data field. Best practice
86 for use of driver_data is to use it as an index into a static list of
87 equivalent device types, not to use it as a pointer.
89 Have a table entry {PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID}
90 to have probe() called for every PCI device known to the system.
92 New PCI IDs may be added to a device driver at runtime by writing
93 to the file /sys/bus/pci/drivers/{driver}/new_id. When added, the
94 driver will probe for all devices it can support.
96 echo "vendor device subvendor subdevice class class_mask driver_data" > \
97 /sys/bus/pci/drivers/{driver}/new_id
98 where all fields are passed in as hexadecimal values (no leading 0x).
99 Users need pass only as many fields as necessary; vendor, device,
100 subvendor, and subdevice fields default to PCI_ANY_ID (FFFFFFFF),
101 class and classmask fields default to 0, and driver_data defaults to
102 0UL. Device drivers must initialize use_driver_data in the dynids struct
103 in their pci_driver struct prior to calling pci_register_driver in order
104 for the driver_data field to get passed to the driver. Otherwise, only a
105 0 is passed in that field.
107 When the driver exits, it just calls pci_unregister_driver() and the PCI layer
108 automatically calls the remove hook for all devices handled by the driver.
110 Please mark the initialization and cleanup functions where appropriate
111 (the corresponding macros are defined in <linux/init.h>):
113 __init Initialization code. Thrown away after the driver
115 __exit Exit code. Ignored for non-modular drivers.
116 __devinit Device initialization code. Identical to __init if
117 the kernel is not compiled with CONFIG_HOTPLUG, normal
119 __devexit The same for __exit.
122 The module_init()/module_exit() functions (and all initialization
123 functions called only from these) should be marked __init/exit.
124 The struct pci_driver shouldn't be marked with any of these tags.
125 The ID table array should be marked __devinitdata.
126 The probe() and remove() functions (and all initialization
127 functions called only from these) should be marked __devinit/exit.
128 If you are sure the driver is not a hotplug driver then use only
129 __init/exit __initdata/exitdata.
131 Pointers to functions marked as __devexit must be created using
132 __devexit_p(function_name). That will generate the function
133 name or NULL if the __devexit function will be discarded.
136 2. How to find PCI devices manually (the old style)
137 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
138 PCI drivers not using the pci_register_driver() interface search
139 for PCI devices manually using the following constructs:
141 Searching by vendor and device ID:
143 struct pci_dev *dev = NULL;
144 while (dev = pci_get_device(VENDOR_ID, DEVICE_ID, dev))
145 configure_device(dev);
147 Searching by class ID (iterate in a similar way):
149 pci_get_class(CLASS_ID, dev)
151 Searching by both vendor/device and subsystem vendor/device ID:
153 pci_get_subsys(VENDOR_ID, DEVICE_ID, SUBSYS_VENDOR_ID, SUBSYS_DEVICE_ID, dev).
155 You can use the constant PCI_ANY_ID as a wildcard replacement for
156 VENDOR_ID or DEVICE_ID. This allows searching for any device from a
157 specific vendor, for example.
159 These functions are hotplug-safe. They increment the reference count on
160 the pci_dev that they return. You must eventually (possibly at module unload)
161 decrement the reference count on these devices by calling pci_dev_put().
164 3. Enabling and disabling devices
165 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
166 Before you do anything with the device you've found, you need to enable
167 it by calling pci_enable_device() which enables I/O and memory regions of
168 the device, allocates an IRQ if necessary, assigns missing resources if
169 needed and wakes up the device if it was in suspended state. Please note
170 that this function can fail.
172 If you want to use the device in bus mastering mode, call pci_set_master()
173 which enables the bus master bit in PCI_COMMAND register and also fixes
174 the latency timer value if it's set to something bogus by the BIOS.
176 If you want to use the PCI Memory-Write-Invalidate transaction,
177 call pci_set_mwi(). This enables the PCI_COMMAND bit for Mem-Wr-Inval
178 and also ensures that the cache line size register is set correctly.
179 Make sure to check the return value of pci_set_mwi(), not all architectures
180 may support Memory-Write-Invalidate.
182 If your driver decides to stop using the device (e.g., there was an
183 error while setting it up or the driver module is being unloaded), it
184 should call pci_disable_device() to deallocate any IRQ resources, disable
185 PCI bus-mastering, etc. You should not do anything with the device after
186 calling pci_disable_device().
188 4. How to access PCI config space
189 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
190 You can use pci_(read|write)_config_(byte|word|dword) to access the config
191 space of a device represented by struct pci_dev *. All these functions return 0
192 when successful or an error code (PCIBIOS_...) which can be translated to a text
193 string by pcibios_strerror. Most drivers expect that accesses to valid PCI
196 If you don't have a struct pci_dev available, you can call
197 pci_bus_(read|write)_config_(byte|word|dword) to access a given device
198 and function on that bus.
200 If you access fields in the standard portion of the config header, please
201 use symbolic names of locations and bits declared in <linux/pci.h>.
203 If you need to access Extended PCI Capability registers, just call
204 pci_find_capability() for the particular capability and it will find the
205 corresponding register block for you.
208 5. Addresses and interrupts
209 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
210 Memory and port addresses and interrupt numbers should NOT be read from the
211 config space. You should use the values in the pci_dev structure as they might
212 have been remapped by the kernel.
214 See Documentation/IO-mapping.txt for how to access device memory.
216 You still need to call request_region() for I/O regions and
217 request_mem_region() for memory regions to make sure nobody else is using the
220 All interrupt handlers should be registered with SA_SHIRQ and use the devid
221 to map IRQs to devices (remember that all PCI interrupts are shared).
224 6. Other interesting functions
225 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
226 pci_find_slot() Find pci_dev corresponding to given bus and
228 pci_set_power_state() Set PCI Power Management state (0=D0 ... 3=D3)
229 pci_find_capability() Find specified capability in device's capability
231 pci_module_init() Inline helper function for ensuring correct
232 pci_driver initialization and error handling.
233 pci_resource_start() Returns bus start address for a given PCI region
234 pci_resource_end() Returns bus end address for a given PCI region
235 pci_resource_len() Returns the byte length of a PCI region
236 pci_set_drvdata() Set private driver data pointer for a pci_dev
237 pci_get_drvdata() Return private driver data pointer for a pci_dev
238 pci_set_mwi() Enable Memory-Write-Invalidate transactions.
239 pci_clear_mwi() Disable Memory-Write-Invalidate transactions.
242 7. Miscellaneous hints
243 ~~~~~~~~~~~~~~~~~~~~~~
244 When displaying PCI slot names to the user (for example when a driver wants
245 to tell the user what card has it found), please use pci_name(pci_dev)
248 Always refer to the PCI devices by a pointer to the pci_dev structure.
249 All PCI layer functions use this identification and it's the only
250 reasonable one. Don't use bus/slot/function numbers except for very
251 special purposes -- on systems with multiple primary buses their semantics
252 can be pretty complex.
254 If you're going to use PCI bus mastering DMA, take a look at
255 Documentation/DMA-mapping.txt.
257 Don't try to turn on Fast Back to Back writes in your driver. All devices
258 on the bus need to be capable of doing it, so this is something which needs
259 to be handled by platform and generic code, not individual drivers.
262 8. Vendor and device identifications
263 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
264 For the future, let's avoid adding device ids to include/linux/pci_ids.h.
266 PCI_VENDOR_ID_xxx for vendors, and a hex constant for device ids.
268 Rationale: PCI_VENDOR_ID_xxx constants are re-used, but device ids are not.
269 Further, device ids are arbitrary hex numbers, normally used only in a
270 single location, the pci_device_id table.
272 9. Obsolete functions
273 ~~~~~~~~~~~~~~~~~~~~~
274 There are several functions which you might come across when trying to
275 port an old driver to the new PCI interface. They are no longer present
276 in the kernel as they aren't compatible with hotplug or PCI domains or
279 pci_find_device() Superseded by pci_get_device()
280 pci_find_subsys() Superseded by pci_get_subsys()
281 pci_find_slot() Superseded by pci_get_slot()