Saturday, 28 September 2013

GSOC : A new dawn

The results are out for GSOC 2013. And thanks to Carol, I got an email congratulating me that I passed the midterms (once again !!). I'd like to thank my mentors and other prominent developers in the RTEMS community whose criticism & wisdom helped me in completing my task and making a worthwhile contribution to RTEMS.

Wanna know how I celebrated passing GSOC final evaluation ? I attended the GDG New Delhi DEVFEST at Google Office, Gurgaon & got a chance to relish this lovely cake. Have a bite  !!

And a couple of cupcakes with cool toppings...

Well, as they say that its not the end, but only the start of a new journey. So, I hope that my journey takes me on a memorable ride where I get to contribute greatly to open source, RTEMS & even more awesome projects.

Monday, 23 September 2013

GSOC : End of the Road ??

This final post written during the end of the coding period of GSOC 2013 summarizes the work done by me this summer on RTEMS. I've created a separate repository on github where all the code & work done by me can be found. I'll briefly take you through the work done & stuff left in the pipeline below :

Contributed Work

  • Submitted patches to remove legacy code from RTEMS bsps which have been accepted & are currently a part of the official RTEMS repository. These patches were submitted for bsps : csb337, csb336, edb7312, gumstix, gba, rtl22xx, & pc386.

  • Since more than about 100 bsps are a part of RTEMS, so guided instructions & rules for continuing this work has been compiled & is present in my github repository & publicly available on a google drive document.

  • Studied the PIC interrupt model based on the comments written in irq-generic.h & my own observations. Information about PIC Interrupt Model API  is added to the current RTEMS documentation.

  • For bsps that do not use the generic interrupt framework , rules to initialize & use it are written and present in the code repository. These rules specify how bsp specific code & the generic code works together & handles Interrupt Service Routines before, during and after an interrupt has occurred.

  • Based on my observations from model bsp lpc24xx , rules were compiled based on an ongoing discussion in the community on how bsp support files are included & compiled. These rules specify which set of files are a must to be used by a bsp & in which directory they all must lie. These rules will in future help in maintaining a unified pattern for BSP file organization.

  • Since the above rules are very difficult to check manually for each bsp, so a bash script was written merging the functionality of check_submission from rtems-testing. This script is supposed to be a final test for any bsp developer before his code is accepted into RTEMS, so as to check whether that the bsp code submitted follows this unified pattern of file organization & formatting or not.

To be Done

  • Continue working with the RTEMS community to strengthen the unified way rules and include more checks in it as suggested.

  • Increase code coverage for removing legacy support from bsps by converting them to PIC interrupt support model.

  • Submit regular patches so as to correct the warnings flagged by the new check_submission script in rtems-testing.

I sincerely hope that my work will benefit the RTEMS community as a whole in the long run & avoid incurring any more technical debt  so that someone else doesn't have to pay for it. I gladly look forward to working with RTEMS developers in future while continuing the work done in GSOC & maybe more !!

Saturday, 14 September 2013

GSOC : Automating rule checking for BSP Unified way

Alas, we've arrived at the last leg of this GSOC journey. At the start of the application phase of GSOC, my mentor Joel explained to me that the main objective of the Unified API project is to determine a unified pattern of file usage, directory structure & the methods that each file uses. And he wished that it would be great, if I could automate checking of these rules that I formulate by observing some model BSPs(lpc24xx).

I believe that I've accomplished that automated rule checking by writing a bash script that searches for discrepancies in a specific BSP, all BSPs in a family or all the RTEMS BSPs in libbsp directory according to the parameters passed. This script can be downloaded from here. Based on my observation, I compiled a list of critical files & methods that every BSP must include and use. These files & methods belonging to a component for a BSP can be deemed as critical or non-critical. The rule-checking is accomplished by the following way :
  1. Check if a file is being compiled by searching it's name in the for a BSP. 
  2. If file not found, then report it.
  3. Check if the file being compiled is present in the correct directory as defined by the rules.
  4. If location of file is not correct, report it.
  5. Check if the corresponding functions for a file are defined in there or not. This takes place by searching for the definition of a method in that file with the help of a simple regex [a-z|A-Z|0-9|_]+[ ]*$i[ ]*\(
  6. If the function is not located in it's specified file, then search whether it's defined in some other file maybe. Report it
  7. Check whether a required header is being installed by searching it's name in the
  8. Compile a list of all RTEMS internal functions(i.e starting with an '_' ) defined in cpukit & libcpu.
  9. Compile another list of functions starting with an '_' being executed in BSP by searching for a ';' to ensure that function is actually being called. Searching for a function being called from a file can be a bit intimidating if the function calling spans multiple lines, meaning that the function name is on one line & the ';' on another. Since grep only parses line by line, so it cannot do this job very easily. What I did is that I trimmed all newlines(\n) while searching a file. The whole file now being concatenated in a single line, so grep is ideal for searching the pattern [ |^][_]+[a-z|A-Z|0-9|_]*[ ]*[\n]*\([^;]*\)[ ]*;
  10. Find common function names in the above two lists in order to ensure that the function being called in a bsp is infact been defined in cpukit or libcpu. Report it.
In order to obtain the results of the script successfully, either call it from a libbsp, cpu family or a bsp directory or pass it a path for a single or multiple bsp directories so that it can validate the location of a bsp in the RTEMS tree. If no path is passed, the pwd is evaluated and searched for bsps.
  • Passing only the directory name for bsp, cpu family or libbsp displays the discrepancies found only for critical files & functions. Following parameters need to be passed to expand the horizon
  • --warnings : gives all the critical & non-critical discrepancies found in a bsp.
  • --verbose : gives the whole story of a BSP. Any discrepancy & a correct rule implementation is also displayed.
  • --tests : gives test specific checks for each given BSP.

Some of the interesting results I found out with my script were :
  • arm/gumstix : bspreset.c not compiled 
  • arm/gumstix : bsp_reset() present in file startup/bspstart.c
  • powerpc/beatnik : bspreset.c not compiled 
  • powerpc/beatnik : bsp_reset() present in file include/bsp.h startup/reboot.c
  • powerpc/mpc55xxevb : start.S not present in correct path 
  • powerpc/mpc55xxevb : bspreset.c not compiled 
  • powerpc/mpc55xxevb : bsp_reset() present in file startup/reset.c 
  • powerpc/mpc55xxevb : bspgetworkarea.c not compiled 
  • powerpc/mpc55xxevb : bsp_work_area_initialize() present in file startup/bspworkareainit.c
The above discrepancies would've been really hard to find if searched manually. I hope my work helps the RTEMS community in following the Unified way for BSP file organization & kill any discrepancies if found any.