blob: cdf2f3c0ab14f8398bee52c1e4a39c41984b28b6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001
Grant Grundler74da15e2006-12-25 01:06:35 -07002 How To Write Linux PCI Drivers
3
4 by Martin Mares <mj@ucw.cz> on 07-Feb-2000
5 updated by Grant Grundler <grundler@parisc-linux.org> on 23-Dec-2006
Linus Torvalds1da177e2005-04-16 15:20:36 -07006
7~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Grant Grundler74da15e2006-12-25 01:06:35 -07008The world of PCI is vast and full of (mostly unpleasant) surprises.
9Since each CPU architecture implements different chip-sets and PCI devices
10have different requirements (erm, "features"), the result is the PCI support
11in the Linux kernel is not as trivial as one would wish. This short paper
12tries to introduce all potential driver authors to Linux APIs for
13PCI device drivers.
14
15A more complete resource is the third edition of "Linux Device Drivers"
16by Jonathan Corbet, Alessandro Rubini, and Greg Kroah-Hartman.
17LDD3 is available for free (under Creative Commons License) from:
18
19 http://lwn.net/Kernel/LDD3/
20
21However, keep in mind that all documents are subject to "bit rot".
22Refer to the source code if things are not working as described here.
23
24Please send questions/comments/patches about Linux PCI API to the
25"Linux PCI" <linux-pci@atrey.karlin.mff.cuni.cz> mailing list.
26
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
28
290. Structure of PCI drivers
30~~~~~~~~~~~~~~~~~~~~~~~~~~~
Grant Grundler74da15e2006-12-25 01:06:35 -070031PCI drivers "discover" PCI devices in a system via pci_register_driver().
32Actually, it's the other way around. When the PCI generic code discovers
33a new device, the driver with a matching "description" will be notified.
34Details on this below.
35
36pci_register_driver() leaves most of the probing for devices to
37the PCI layer and supports online insertion/removal of devices [thus
38supporting hot-pluggable PCI, CardBus, and Express-Card in a single driver].
39pci_register_driver() call requires passing in a table of function
40pointers and thus dictates the high level structure of a driver.
41
42Once the driver knows about a PCI device and takes ownership, the
43driver generally needs to perform the following initialization:
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45 Enable the device
Grant Grundler74da15e2006-12-25 01:06:35 -070046 Request MMIO/IOP resources
47 Set the DMA mask size (for both coherent and streaming DMA)
48 Allocate and initialize shared control data (pci_allocate_coherent())
49 Access device configuration space (if needed)
50 Register IRQ handler (request_irq())
51 Initialize non-PCI (i.e. LAN/SCSI/etc parts of the chip)
52 Enable DMA/processing engines
53
54When done using the device, and perhaps the module needs to be unloaded,
55the driver needs to take the follow steps:
56 Disable the device from generating IRQs
57 Release the IRQ (free_irq())
58 Stop all DMA activity
59 Release DMA buffers (both streaming and coherent)
60 Unregister from other subsystems (e.g. scsi or netdev)
61 Release MMIO/IOP resources
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 Disable the device
63
Grant Grundler74da15e2006-12-25 01:06:35 -070064Most of these topics are covered in the following sections.
65For the rest look at LDD3 or <linux/pci.h> .
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67If the PCI subsystem is not configured (CONFIG_PCI is not set), most of
Grant Grundler74da15e2006-12-25 01:06:35 -070068the PCI functions described below are defined as inline functions either
69completely empty or just returning an appropriate error codes to avoid
70lots of ifdefs in the drivers.
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
Grant Grundler74da15e2006-12-25 01:06:35 -0700741. pci_register_driver() call
75~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
76
77PCI device drivers call pci_register_driver() during their
78initialization with a pointer to a structure describing the driver
79(struct pci_driver):
80
81 field name Description
82 ---------- ------------------------------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 id_table Pointer to table of device ID's the driver is
84 interested in. Most drivers should export this
85 table using MODULE_DEVICE_TABLE(pci,...).
Grant Grundler74da15e2006-12-25 01:06:35 -070086
87 probe This probing function gets called (during execution
88 of pci_register_driver() for already existing
89 devices or later if a new device gets inserted) for
90 all PCI devices which match the ID table and are not
91 "owned" by the other drivers yet. This function gets
92 passed a "struct pci_dev *" for each device whose
93 entry in the ID table matches the device. The probe
94 function returns zero when the driver chooses to
95 take "ownership" of the device or an error code
96 (negative number) otherwise.
97 The probe function always gets called from process
98 context, so it can sleep.
99
100 remove The remove() function gets called whenever a device
101 being handled by this driver is removed (either during
102 deregistration of the driver or when it's manually
103 pulled out of a hot-pluggable slot).
104 The remove function always gets called from process
105 context, so it can sleep.
106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 suspend Put device into low power state.
Grant Grundler74da15e2006-12-25 01:06:35 -0700108 suspend_late Put device into low power state.
109
110 resume_early Wake device from low power state.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 resume Wake device from low power state.
Grant Grundler74da15e2006-12-25 01:06:35 -0700112
113 (Please see Documentation/power/pci.txt for descriptions
114 of PCI Power Management and the related functions.)
115
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 enable_wake Enable device to generate wake events from a low power
117 state.
118
Grant Grundler74da15e2006-12-25 01:06:35 -0700119 shutdown Hook into reboot_notifier_list (kernel/sys.c).
120 Intended to stop any idling DMA operations.
121 Useful for enabling wake-on-lan (NIC) or changing
122 the power state of a device before reboot.
123 e.g. drivers/net/e100.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Grant Grundler74da15e2006-12-25 01:06:35 -0700125 err_handler See Documentation/pci-error-recovery.txt
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Grant Grundler74da15e2006-12-25 01:06:35 -0700127 multithread_probe Enable multi-threaded probe/scan. Driver must
128 provide its own locking/syncronization for init
129 operations if this is enabled.
130
131
132The ID table is an array of struct pci_device_id entries ending with an
133all-zero entry. Each entry consists of:
134
135 vendor,device Vendor and device ID to match (or PCI_ANY_ID)
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 subvendor, Subsystem vendor and device ID to match (or PCI_ANY_ID)
Grant Grundler74da15e2006-12-25 01:06:35 -0700138 subdevice,
139
140 class Device class, subclass, and "interface" to match.
141 See Appendix D of the PCI Local Bus Spec or
142 include/linux/pci_ids.h for a full list of classes.
143 Most drivers do not need to specify class/class_mask
144 as vendor/device is normally sufficient.
145
146 class_mask limit which sub-fields of the class field are compared.
147 See drivers/scsi/sym53c8xx_2/ for example of usage.
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 driver_data Data private to the driver.
Grant Grundler74da15e2006-12-25 01:06:35 -0700150 Most drivers don't need to use driver_data field.
151 Best practice is to use driver_data as an index
152 into a static list of equivalent device types,
153 instead of using it as a pointer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
Grant Grundler74da15e2006-12-25 01:06:35 -0700156Most drivers only need PCI_DEVICE() or PCI_DEVICE_CLASS() to set up
157a pci_device_id table.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
Grant Grundler74da15e2006-12-25 01:06:35 -0700159New PCI IDs may be added to a device driver pci_ids table at runtime
160as shown below:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162echo "vendor device subvendor subdevice class class_mask driver_data" > \
Grant Grundler74da15e2006-12-25 01:06:35 -0700163/sys/bus/pci/drivers/{driver}/new_id
164
165All fields are passed in as hexadecimal values (no leading 0x).
166Users need pass only as many fields as necessary:
167 o vendor, device, subvendor, and subdevice fields default
168 to PCI_ANY_ID (FFFFFFFF),
169 o class and classmask fields default to 0
170 o driver_data defaults to 0UL.
171
172Once added, the driver probe routine will be invoked for any unclaimed
173PCI devices listed in its (newly updated) pci_ids list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175When the driver exits, it just calls pci_unregister_driver() and the PCI layer
176automatically calls the remove hook for all devices handled by the driver.
177
Grant Grundler74da15e2006-12-25 01:06:35 -0700178
1791.1 "Attributes" for driver functions/data
180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181Please mark the initialization and cleanup functions where appropriate
182(the corresponding macros are defined in <linux/init.h>):
183
184 __init Initialization code. Thrown away after the driver
185 initializes.
186 __exit Exit code. Ignored for non-modular drivers.
Grant Grundler74da15e2006-12-25 01:06:35 -0700187
188
189 __devinit Device initialization code.
190 Identical to __init if the kernel is not compiled
191 with CONFIG_HOTPLUG, normal function otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 __devexit The same for __exit.
193
Grant Grundler74da15e2006-12-25 01:06:35 -0700194Tips on when/where to use the above attributes:
195 o The module_init()/module_exit() functions (and all
196 initialization functions called _only_ from these)
197 should be marked __init/__exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Grant Grundler74da15e2006-12-25 01:06:35 -0700199 o Do not mark the struct pci_driver.
200
201 o The ID table array should be marked __devinitdata.
202
203 o The probe() and remove() functions should be marked __devinit
204 and __devexit respectively. All initialization functions
205 exclusively called by the probe() routine, can be marked __devinit.
206 Ditto for remove() and __devexit.
207
Grant Grundler26ba05e2007-02-11 00:04:04 -0700208 o If mydriver_remove() is marked with __devexit(), then all address
209 references to mydriver_remove must use __devexit_p(mydriver_remove)
Grant Grundler74da15e2006-12-25 01:06:35 -0700210 (in the struct pci_driver declaration for example).
211 __devexit_p() will generate the function name _or_ NULL if the
212 function will be discarded. For an example, see drivers/net/tg3.c.
213
214 o Do NOT mark a function if you are not sure which mark to use.
215 Better to not mark the function than mark the function wrong.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
217
Grant Grundler74da15e2006-12-25 01:06:35 -0700218
2192. How to find PCI devices manually
220~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
221
222PCI drivers should have a really good reason for not using the
223pci_register_driver() interface to search for PCI devices.
224The main reason PCI devices are controlled by multiple drivers
225is because one PCI device implements several different HW services.
226E.g. combined serial/parallel port/floppy controller.
227
228A manual search may be performed using the following constructs:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230Searching by vendor and device ID:
231
232 struct pci_dev *dev = NULL;
233 while (dev = pci_get_device(VENDOR_ID, DEVICE_ID, dev))
234 configure_device(dev);
235
236Searching by class ID (iterate in a similar way):
237
238 pci_get_class(CLASS_ID, dev)
239
240Searching by both vendor/device and subsystem vendor/device ID:
241
Grant Grundler74da15e2006-12-25 01:06:35 -0700242 pci_get_subsys(VENDOR_ID,DEVICE_ID, SUBSYS_VENDOR_ID, SUBSYS_DEVICE_ID, dev).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Grant Grundler74da15e2006-12-25 01:06:35 -0700244You can use the constant PCI_ANY_ID as a wildcard replacement for
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245VENDOR_ID or DEVICE_ID. This allows searching for any device from a
246specific vendor, for example.
247
Grant Grundler74da15e2006-12-25 01:06:35 -0700248These functions are hotplug-safe. They increment the reference count on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249the pci_dev that they return. You must eventually (possibly at module unload)
250decrement the reference count on these devices by calling pci_dev_put().
251
252
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Grant Grundler74da15e2006-12-25 01:06:35 -07002543. Device Initialization Steps
255~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Grant Grundler74da15e2006-12-25 01:06:35 -0700257As noted in the introduction, most PCI drivers need the following steps
258for device initialization:
259
260 Enable the device
261 Request MMIO/IOP resources
262 Set the DMA mask size (for both coherent and streaming DMA)
263 Allocate and initialize shared control data (pci_allocate_coherent())
264 Access device configuration space (if needed)
265 Register IRQ handler (request_irq())
266 Initialize non-PCI (i.e. LAN/SCSI/etc parts of the chip)
267 Enable DMA/processing engines.
268
269The driver can access PCI config space registers at any time.
270(Well, almost. When running BIST, config space can go away...but
271that will just result in a PCI Bus Master Abort and config reads
272will return garbage).
273
274
2753.1 Enable the PCI device
276~~~~~~~~~~~~~~~~~~~~~~~~~
277Before touching any device registers, the driver needs to enable
278the PCI device by calling pci_enable_device(). This will:
279 o wake up the device if it was in suspended state,
280 o allocate I/O and memory regions of the device (if BIOS did not),
281 o allocate an IRQ (if BIOS did not).
282
283NOTE: pci_enable_device() can fail! Check the return value.
284NOTE2: Also see pci_enable_device_bars() below. Drivers can
285 attempt to enable only a subset of BARs they need.
286
287[ OS BUG: we don't check resource allocations before enabling those
288 resources. The sequence would make more sense if we called
289 pci_request_resources() before calling pci_enable_device().
290 Currently, the device drivers can't detect the bug when when two
291 devices have been allocated the same range. This is not a common
292 problem and unlikely to get fixed soon.
293
294 This has been discussed before but not changed as of 2.6.19:
295 http://lkml.org/lkml/2006/3/2/194
296]
297
298pci_set_master() will enable DMA by setting the bus master bit
299in the PCI_COMMAND register. It also fixes the latency timer value if
300it's set to something bogus by the BIOS.
301
302If the PCI device can use the PCI Memory-Write-Invalidate transaction,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303call pci_set_mwi(). This enables the PCI_COMMAND bit for Mem-Wr-Inval
304and also ensures that the cache line size register is set correctly.
Grant Grundler74da15e2006-12-25 01:06:35 -0700305Check the return value of pci_set_mwi() as not all architectures
306or chip-sets may support Memory-Write-Invalidate.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
Grant Grundler74da15e2006-12-25 01:06:35 -0700308
3093.2 Request MMIO/IOP resources
310~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
311Memory (MMIO), and I/O port addresses should NOT be read directly
312from the PCI device config space. Use the values in the pci_dev structure
313as the PCI "bus address" might have been remapped to a "host physical"
314address by the arch/chip-set specific kernel support.
315
316See Documentation/IO-mapping.txt for how to access device registers
317or device memory.
318
319The device driver needs to call pci_request_region() to verify
320no other device is already using the same address resource.
321Conversely, drivers should call pci_release_region() AFTER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322calling pci_disable_device().
Grant Grundler74da15e2006-12-25 01:06:35 -0700323The idea is to prevent two devices colliding on the same address range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
Grant Grundler74da15e2006-12-25 01:06:35 -0700325[ See OS BUG comment above. Currently (2.6.19), The driver can only
326 determine MMIO and IO Port resource availability _after_ calling
327 pci_enable_device(). ]
328
329Generic flavors of pci_request_region() are request_mem_region()
330(for MMIO ranges) and request_region() (for IO Port ranges).
331Use these for address resources that are not described by "normal" PCI
332BARs.
333
334Also see pci_request_selected_regions() below.
335
336
3373.3 Set the DMA mask size
338~~~~~~~~~~~~~~~~~~~~~~~~~
339[ If anything below doesn't make sense, please refer to
340 Documentation/DMA-API.txt. This section is just a reminder that
341 drivers need to indicate DMA capabilities of the device and is not
342 an authoritative source for DMA interfaces. ]
343
344While all drivers should explicitly indicate the DMA capability
345(e.g. 32 or 64 bit) of the PCI bus master, devices with more than
34632-bit bus master capability for streaming data need the driver
347to "register" this capability by calling pci_set_dma_mask() with
348appropriate parameters. In general this allows more efficient DMA
349on systems where System RAM exists above 4G _physical_ address.
350
351Drivers for all PCI-X and PCIe compliant devices must call
352pci_set_dma_mask() as they are 64-bit DMA devices.
353
354Similarly, drivers must also "register" this capability if the device
355can directly address "consistent memory" in System RAM above 4G physical
356address by calling pci_set_consistent_dma_mask().
357Again, this includes drivers for all PCI-X and PCIe compliant devices.
358Many 64-bit "PCI" devices (before PCI-X) and some PCI-X devices are
35964-bit DMA capable for payload ("streaming") data but not control
360("consistent") data.
361
362
3633.4 Setup shared control data
364~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
365Once the DMA masks are set, the driver can allocate "consistent" (a.k.a. shared)
366memory. See Documentation/DMA-API.txt for a full description of
367the DMA APIs. This section is just a reminder that it needs to be done
368before enabling DMA on the device.
369
370
3713.5 Initialize device registers
372~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
373Some drivers will need specific "capability" fields programmed
374or other "vendor specific" register initialized or reset.
375E.g. clearing pending interrupts.
376
377
3783.6 Register IRQ handler
379~~~~~~~~~~~~~~~~~~~~~~~~
380While calling request_irq() is the the last step described here,
381this is often just another intermediate step to initialize a device.
382This step can often be deferred until the device is opened for use.
383
384All interrupt handlers for IRQ lines should be registered with IRQF_SHARED
385and use the devid to map IRQs to devices (remember that all PCI IRQ lines
386can be shared).
387
388request_irq() will associate an interrupt handler and device handle
389with an interrupt number. Historically interrupt numbers represent
390IRQ lines which run from the PCI device to the Interrupt controller.
391With MSI and MSI-X (more below) the interrupt number is a CPU "vector".
392
393request_irq() also enables the interrupt. Make sure the device is
394quiesced and does not have any interrupts pending before registering
395the interrupt handler.
396
397MSI and MSI-X are PCI capabilities. Both are "Message Signaled Interrupts"
398which deliver interrupts to the CPU via a DMA write to a Local APIC.
399The fundamental difference between MSI and MSI-X is how multiple
400"vectors" get allocated. MSI requires contiguous blocks of vectors
401while MSI-X can allocate several individual ones.
402
403MSI capability can be enabled by calling pci_enable_msi() or
404pci_enable_msix() before calling request_irq(). This causes
405the PCI support to program CPU vector data into the PCI device
406capability registers.
407
408If your PCI device supports both, try to enable MSI-X first.
409Only one can be enabled at a time. Many architectures, chip-sets,
410or BIOSes do NOT support MSI or MSI-X and the call to pci_enable_msi/msix
411will fail. This is important to note since many drivers have
412two (or more) interrupt handlers: one for MSI/MSI-X and another for IRQs.
413They choose which handler to register with request_irq() based on the
414return value from pci_enable_msi/msix().
415
416There are (at least) two really good reasons for using MSI:
4171) MSI is an exclusive interrupt vector by definition.
418 This means the interrupt handler doesn't have to verify
419 its device caused the interrupt.
420
4212) MSI avoids DMA/IRQ race conditions. DMA to host memory is guaranteed
422 to be visible to the host CPU(s) when the MSI is delivered. This
423 is important for both data coherency and avoiding stale control data.
424 This guarantee allows the driver to omit MMIO reads to flush
425 the DMA stream.
426
427See drivers/infiniband/hw/mthca/ or drivers/net/tg3.c for examples
428of MSI/MSI-X usage.
429
430
431
4324. PCI device shutdown
433~~~~~~~~~~~~~~~~~~~~~~~
434
435When a PCI device driver is being unloaded, most of the following
436steps need to be performed:
437
438 Disable the device from generating IRQs
439 Release the IRQ (free_irq())
440 Stop all DMA activity
441 Release DMA buffers (both streaming and consistent)
442 Unregister from other subsystems (e.g. scsi or netdev)
443 Disable device from responding to MMIO/IO Port addresses
444 Release MMIO/IO Port resource(s)
445
446
4474.1 Stop IRQs on the device
448~~~~~~~~~~~~~~~~~~~~~~~~~~~
449How to do this is chip/device specific. If it's not done, it opens
450the possibility of a "screaming interrupt" if (and only if)
451the IRQ is shared with another device.
452
453When the shared IRQ handler is "unhooked", the remaining devices
454using the same IRQ line will still need the IRQ enabled. Thus if the
455"unhooked" device asserts IRQ line, the system will respond assuming
456it was one of the remaining devices asserted the IRQ line. Since none
457of the other devices will handle the IRQ, the system will "hang" until
458it decides the IRQ isn't going to get handled and masks the IRQ (100,000
459iterations later). Once the shared IRQ is masked, the remaining devices
460will stop functioning properly. Not a nice situation.
461
462This is another reason to use MSI or MSI-X if it's available.
463MSI and MSI-X are defined to be exclusive interrupts and thus
464are not susceptible to the "screaming interrupt" problem.
465
466
4674.2 Release the IRQ
468~~~~~~~~~~~~~~~~~~~
469Once the device is quiesced (no more IRQs), one can call free_irq().
470This function will return control once any pending IRQs are handled,
471"unhook" the drivers IRQ handler from that IRQ, and finally release
472the IRQ if no one else is using it.
473
474
4754.3 Stop all DMA activity
476~~~~~~~~~~~~~~~~~~~~~~~~~
477It's extremely important to stop all DMA operations BEFORE attempting
478to deallocate DMA control data. Failure to do so can result in memory
479corruption, hangs, and on some chip-sets a hard crash.
480
481Stopping DMA after stopping the IRQs can avoid races where the
482IRQ handler might restart DMA engines.
483
484While this step sounds obvious and trivial, several "mature" drivers
485didn't get this step right in the past.
486
487
4884.4 Release DMA buffers
489~~~~~~~~~~~~~~~~~~~~~~~
490Once DMA is stopped, clean up streaming DMA first.
491I.e. unmap data buffers and return buffers to "upstream"
492owners if there is one.
493
494Then clean up "consistent" buffers which contain the control data.
495
496See Documentation/DMA-API.txt for details on unmapping interfaces.
497
498
4994.5 Unregister from other subsystems
500~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
501Most low level PCI device drivers support some other subsystem
502like USB, ALSA, SCSI, NetDev, Infiniband, etc. Make sure your
503driver isn't losing resources from that other subsystem.
504If this happens, typically the symptom is an Oops (panic) when
505the subsystem attempts to call into a driver that has been unloaded.
506
507
5084.6 Disable Device from responding to MMIO/IO Port addresses
509~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
510io_unmap() MMIO or IO Port resources and then call pci_disable_device().
511This is the symmetric opposite of pci_enable_device().
512Do not access device registers after calling pci_disable_device().
513
514
5154.7 Release MMIO/IO Port Resource(s)
516~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
517Call pci_release_region() to mark the MMIO or IO Port range as available.
518Failure to do so usually results in the inability to reload the driver.
519
520
521
5225. How to access PCI config space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Grant Grundler74da15e2006-12-25 01:06:35 -0700524
525You can use pci_(read|write)_config_(byte|word|dword) to access the config
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526space of a device represented by struct pci_dev *. All these functions return 0
527when successful or an error code (PCIBIOS_...) which can be translated to a text
528string by pcibios_strerror. Most drivers expect that accesses to valid PCI
529devices don't fail.
530
Grant Grundler74da15e2006-12-25 01:06:35 -0700531If you don't have a struct pci_dev available, you can call
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532pci_bus_(read|write)_config_(byte|word|dword) to access a given device
533and function on that bus.
534
Grant Grundler74da15e2006-12-25 01:06:35 -0700535If you access fields in the standard portion of the config header, please
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536use symbolic names of locations and bits declared in <linux/pci.h>.
537
Grant Grundler74da15e2006-12-25 01:06:35 -0700538If you need to access Extended PCI Capability registers, just call
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539pci_find_capability() for the particular capability and it will find the
540corresponding register block for you.
541
542
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
5446. Other interesting functions
545~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Grant Grundler74da15e2006-12-25 01:06:35 -0700546
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547pci_find_slot() Find pci_dev corresponding to given bus and
548 slot numbers.
549pci_set_power_state() Set PCI Power Management state (0=D0 ... 3=D3)
550pci_find_capability() Find specified capability in device's capability
551 list.
552pci_module_init() Inline helper function for ensuring correct
553 pci_driver initialization and error handling.
554pci_resource_start() Returns bus start address for a given PCI region
555pci_resource_end() Returns bus end address for a given PCI region
556pci_resource_len() Returns the byte length of a PCI region
557pci_set_drvdata() Set private driver data pointer for a pci_dev
558pci_get_drvdata() Return private driver data pointer for a pci_dev
559pci_set_mwi() Enable Memory-Write-Invalidate transactions.
560pci_clear_mwi() Disable Memory-Write-Invalidate transactions.
561
562
Grant Grundler74da15e2006-12-25 01:06:35 -0700563
Linus Torvalds1da177e2005-04-16 15:20:36 -07005647. Miscellaneous hints
565~~~~~~~~~~~~~~~~~~~~~~
Grant Grundler74da15e2006-12-25 01:06:35 -0700566
567When displaying PCI device names to the user (for example when a driver wants
568to tell the user what card has it found), please use pci_name(pci_dev).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
570Always refer to the PCI devices by a pointer to the pci_dev structure.
571All PCI layer functions use this identification and it's the only
572reasonable one. Don't use bus/slot/function numbers except for very
573special purposes -- on systems with multiple primary buses their semantics
574can be pretty complex.
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576Don't try to turn on Fast Back to Back writes in your driver. All devices
577on the bus need to be capable of doing it, so this is something which needs
578to be handled by platform and generic code, not individual drivers.
579
580
Grant Grundler74da15e2006-12-25 01:06:35 -0700581
Ingo Oeser9b860b82006-04-18 11:20:55 +02005828. Vendor and device identifications
583~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Ingo Oeser9b860b82006-04-18 11:20:55 +0200584
Grant Grundler74da15e2006-12-25 01:06:35 -0700585One is not not required to add new device ids to include/linux/pci_ids.h.
586Please add PCI_VENDOR_ID_xxx for vendors and a hex constant for device ids.
Ingo Oeser9b860b82006-04-18 11:20:55 +0200587
Grant Grundler74da15e2006-12-25 01:06:35 -0700588PCI_VENDOR_ID_xxx constants are re-used. The device ids are arbitrary
589hex numbers (vendor controlled) and normally used only in a single
590location, the pci_device_id table.
591
592Please DO submit new vendor/device ids to pciids.sourceforge.net project.
593
594
Ingo Oeser9b860b82006-04-18 11:20:55 +0200595
5969. Obsolete functions
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597~~~~~~~~~~~~~~~~~~~~~
Grant Grundler74da15e2006-12-25 01:06:35 -0700598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599There are several functions which you might come across when trying to
600port an old driver to the new PCI interface. They are no longer present
601in the kernel as they aren't compatible with hotplug or PCI domains or
602having sane locking.
603
Grant Grundler74da15e2006-12-25 01:06:35 -0700604pci_find_device() Superseded by pci_get_device()
605pci_find_subsys() Superseded by pci_get_subsys()
606pci_find_slot() Superseded by pci_get_slot()
607
608
609The alternative is the traditional PCI device driver that walks PCI
610device lists. This is still possible but discouraged.
611
612
613
61410. pci_enable_device_bars() and Legacy I/O Port space
615~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
616
617Large servers may not be able to provide I/O port resources to all PCI
618devices. I/O Port space is only 64KB on Intel Architecture[1] and is
619likely also fragmented since the I/O base register of PCI-to-PCI
620bridge will usually be aligned to a 4KB boundary[2]. On such systems,
621pci_enable_device() and pci_request_region() will fail when
622attempting to enable I/O Port regions that don't have I/O Port
623resources assigned.
624
625Fortunately, many PCI devices which request I/O Port resources also
626provide access to the same registers via MMIO BARs. These devices can
627be handled without using I/O port space and the drivers typically
628offer a CONFIG_ option to only use MMIO regions
629(e.g. CONFIG_TULIP_MMIO). PCI devices typically provide I/O port
630interface for legacy OSes and will work when I/O port resources are not
631assigned. The "PCI Local Bus Specification Revision 3.0" discusses
632this on p.44, "IMPLEMENTATION NOTE".
633
634If your PCI device driver doesn't need I/O port resources assigned to
635I/O Port BARs, you should use pci_enable_device_bars() instead of
636pci_enable_device() in order not to enable I/O port regions for the
637corresponding devices. In addition, you should use
638pci_request_selected_regions() and pci_release_selected_regions()
639instead of pci_request_regions()/pci_release_regions() in order not to
640request/release I/O port regions for the corresponding devices.
641
642[1] Some systems support 64KB I/O port space per PCI segment.
643[2] Some PCI-to-PCI bridges support optional 1KB aligned I/O base.
644
645
646
64711. MMIO Space and "Write Posting"
648~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
649
650Converting a driver from using I/O Port space to using MMIO space
651often requires some additional changes. Specifically, "write posting"
652needs to be handled. Many drivers (e.g. tg3, acenic, sym53c8xx_2)
653already do this. I/O Port space guarantees write transactions reach the PCI
654device before the CPU can continue. Writes to MMIO space allow the CPU
655to continue before the transaction reaches the PCI device. HW weenies
656call this "Write Posting" because the write completion is "posted" to
657the CPU before the transaction has reached its destination.
658
659Thus, timing sensitive code should add readl() where the CPU is
660expected to wait before doing other work. The classic "bit banging"
661sequence works fine for I/O Port space:
662
663 for (i = 8; --i; val >>= 1) {
664 outb(val & 1, ioport_reg); /* write bit */
665 udelay(10);
666 }
667
668The same sequence for MMIO space should be:
669
670 for (i = 8; --i; val >>= 1) {
671 writeb(val & 1, mmio_reg); /* write bit */
672 readb(safe_mmio_reg); /* flush posted write */
673 udelay(10);
674 }
675
676It is important that "safe_mmio_reg" not have any side effects that
677interferes with the correct operation of the device.
678
679Another case to watch out for is when resetting a PCI device. Use PCI
680Configuration space reads to flush the writel(). This will gracefully
681handle the PCI master abort on all platforms if the PCI device is
682expected to not respond to a readl(). Most x86 platforms will allow
683MMIO reads to master abort (a.k.a. "Soft Fail") and return garbage
684(e.g. ~0). But many RISC platforms will crash (a.k.a."Hard Fail").
685