> -----Original Message----- > From: Stefan Reinauer [mailto:[EMAIL PROTECTED] > Sent: Thursday, November 06, 2008 9:17 AM > To: Vincent Legoll > Cc: Myles Watson; ron minnich; Carl-Daniel Hailfinger; > [email protected] > Subject: Re: [coreboot] r983 - in coreboot-v3: device include/device > > Vincent Legoll wrote: > > Isn't all that kind of things doable via function pointers and link-time > dead > > code elimination ? That would achieve the no ifdeffery goal, and may be > > cleaner code... > Link time optimization would suggest we look into compiling with LLVM > instead of gcc. > > Has anyone tried this, yet? >
As a start that wouldn't change the code much and would still work with gcc, we could have an enum or defines in the device code and a table where the device code that was compiled in could initialize a pointer to its default ops. In practice it only affects dynamically detected bridges, but it would be nice to support them. Something like #define DEVICE_PCI_DRIVER_NUM 1 #define DEVICE_HT_DRIVER_NUM 2 #define DEVICE_PCIE_DRIVER_NUM 1 #define DEVICE_PCIX_DRIVER_NUM 1 #define DEVICE_MAX_DRIVERS DEVICE_PCIX_DRIVER_NUM+1 static const device_operations * default_drivers[DEVICE_MAX_DRIVERS]; Then each device could set them: default_drivers[DEVICE_HT_DRIVER_NUM] = &default_ht_ops_bus; Or something similar. Note that this is just dry coding. Thanks, Myles Thanks, Myles -- coreboot mailing list: [email protected] http://www.coreboot.org/mailman/listinfo/coreboot
