Added BAR (Base Address Registers) as a parameter for PCI scan.
Some devices (as UART in Quark) use two set of BARs for different
purposes. A driver may require only one of them.
BARs are numbered from 0 to PCI_MAX_BARS.
PCI_BAR_ANY means ignore the BAR number. Constants are defined
in drivers/pci.h
If device class is not specified as a scanning parameter, and
set to 0, ignore it.
Change-Id: I6b7116c5c6cf9c470ab22bec9eb74842f15b5d99
Signed-off-by: Dmitriy Korovkin <dmitriy.korovkin@windriver.com>
This fixes many issues around PCI enumeration from old API:
- a static internal table was fed with scanning results, thus eating
memory, and worse: due to the limit of its size, scanning for new
classes was impossible unless growing statically the size of this
table --> more memory eaten! Not to mention PCI enumeration is done
once at boot time for driver initialization and that's all, so this
table is hanging around for nothing afterwards.
- one needs first to scan a class, then maybe he will be able to find
his device via pci_dev_find. Where all could be done at once.
- pci_dev_find was not trustworthy due again to the internal table. Now
if the device is not found, one will know it really went through all
the possbilities.
- still let the possibility for hard-coded BARs value on driver side
(thus no PCI scan required). However this is greatly advised not to do
so as BARs might change over a firmware/BIOS update.
Comparison:
old pci_dev_scan: could only filter out via class mask.
new pci_dev_scan: can filter out via a class, a vendor and device ID
(it could easily do the same for Function and BAR index as these are
usually fixed and informed through datasheet)
old pci_dev_scan: was limited in its findings by the size of the
internal result table.
new pci_dev_scan: can proceed through all the buses and devices every
time (there are optimizations to avoid useless work of course)
old results did not tell about the function or BAR index.
new one tells, and the structure has not bloated.
old internal code: was storing a big table of results
new internal code: is only storing a small lookup structure and an
array of Bus:Dev pairs for each PCI class for optimizations purpose.
(though, if needed, we could disable this through some #ifdef)
Usage:
- Have a local struct dev_info
- Fill it with what you want to look for, currently: only class and
vendor_id/device_id. Function and BAR index could be added if needed.
- Call pci_bus_scan_init(): this will reset the internal lookup
structure.
- Call pci_dev_scan(<a pointer to your dev_info>): at first call, the
internal lookup structure will pick up the informations from dev_info
and will try to find out what has been requested. It will return 1 on
success, or 0. On 1, your dev_info structure will be updated with the
found informations. If more devices can be found against the same
lookup informations, just call again pci_dev_scan(<a pointer to your
dev_info>) as long as it returns 1. When 0 is hit, it will mean you
found all.
Change-Id: Ibc2a16c4485ee3fed7ef4946af0ece032ae406e4
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Some checkpatch issues were solved by scripts leaving other problems
such as alignment and indentation issues. In order to comply with the
defined coding style the following fixes were made:
- Fixed the function declaration moving the parameters' comments above
the function in accordance to the doxygen format.
- Fixed functions' opening and closing brackets. These brackets should
not be indented.
- Fixed the 'if', 'for' and 'while' statements adding the brackets
around the sentence.
- Fixed comments' alignment.
- Fixed indentation.
The work was done manually and submitted as one commit. I didn't
separate these changes in different commits because they were fixed all
at once. Basically, all errors were fixed in every file at once.
Change-Id: Icc94a10bfd2cff82007ce60df23b2ccd4c30268d
Signed-off-by: Yonattan Louise <yonattan.a.louise.mendoza@intel.com>
Instead of drivers/pci/, the public API headers will be found in
include/drivers/pci.
Change-Id: I577036660383e6bd9c015d6bbbcbc14bf8fb67ec
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Dmitriy Korovkin