Friday, August 14, 2009

towards a integrated infrastructure for feature + errata handling in kernel

I have a confession to make: I am fanatic about clean code. I like it simple and I like it elegant.

Few days back, I had a chat on linux omap mailing list on what I'd dream to see instead of a half a dozen cpu_has_feature_XYZ() APIs in the system.. The thought is more of this form:
  • Some sort of infrastructure which is common to all silicons,platforms exist which provides method to register the tuple (silicon_rev_id | ip_module_rev_id, feature_errata_id)
  • Specific drivers such as DMA, I2C, McBSP, PRCM etc.. on registration provides to the system the an array of tuples for that device.
  • Platform on bootup provides the platform_data to the driver, which allows the driver to start, but now, it can also detect cpu_rev and can grap it's own IP Module rev from it's own register.. -> that is provided to the infrastructure

  • Every time there is a need to check to enable/disable feature/errata, all the code needs to do is:
if (is_enabled(dev, OMAP3_ISP) ) {
do my stuff;
}

is'nt that cleaner than
if (cpu_is_3430() && (cpu_rev >= OMAP3430_ES3P1) ) {
do my stuff;
}
or
if (cpu_is_3430() && (omap_features & OMAP_HAS_ISP) {
do my stuff;
}
  • Now what ever we do here can easily be displayed by the subsystem by exposing a cat /sys/devices/... /features or even club it with cat /proc/cpuinfo if that is what people want -> you can do it anytime anywhere if you retain just the data you need... ;)
  • No need to store the entire database of features in the infrastructure for ever, we can save memory by making the original array init_data and copying it into a list which is easier to search for the infrastructure..

Now, this infrastructure can also be readily used for external peripherals outside the chip - like camera sensors, RFBI displays etc.. come to think of this, it kinda looks the way how I2C board_info is handled in some ways.. hmmm...

Now, only thing I need is get some time to do this, as Kevin seems interested in the original patch.. couple of fortnights to go before i get some personal time to hack some *real* code.. ;) but what the heck.. if there is anyone out there who has the time and interest to do this.. go ahead and do it..

No comments: