driverloader with 2.6.24-rcX
[email protected] (Bob Tracy) Thu, 8 Nov 2007 20:37:54 -0600 (CST)
| Newsgroups | gmane.linux.kernel.driver-loader |
|---|---|
| Message-ID | <[email protected]> |
The driverloader-2.39 release worked great up until the first 2.6.24 kernel release candidates appeared. The biggest obstacle to building a valid driverloader module seems to be that do_softirq() is no longer being exported. There are other minor fixes to the driverloader code required, and as I don't have it in front of me right now, I'll have to go by memory :-(. The Kbuild process now complains bitterly about changes to CFLAGS, and wants out-of-tree builds to use EXTRA_CFLAGS instead. This means lots of trivial mods to the main driverloader module makefile, but the build seems to work fine once those mods are made. The access_ok macro is now defined in linux/uaccess_32.h and/or linux/uaccess_64.h, i.e., the Makefile check for whether that macro is defined is incorrectly looking in linux/uaccess.h, which simply points to the 32-bit and 64-bit files. SET_MODULE_OWNER has gone bye-bye, and all references to it can be removed for kernels greater than 2.6.23. I think this affects two files: dldrpci.c and dldrusb.c (if I'm remembering the names correctly). I don't know what an appropriate fix might be for the missing do_softirq(). Fixing everything else allows the build to succeed, but the resulting module won't load because of the undefined do_softirq symbol. If anyone can suggest an appropriate fix, I'll be happy to give it a try. Thanks! -- ------------------------------------------------------------------------ Bob Tracy | "They couldn't hit an elephant at this dist- " [email protected] | - Last words of Union General John Sedgwick, | Battle of Spotsylvania Court House, U.S. Civil War ------------------------------------------------------------------------