I got past my tool build errors... it seems like maybe there are race conditions if you set the number of build jobs to 2? (I was trying to get things done faster on a dual-core host). It was odd errors about ld not being configured properly for options being passed to it (in the cross-compiler chain). Once I set the number of jobs back to 1 and did a distclean, things started building OK.
However, I built revision 7004 ar7-2.6 with wireless and adsl support and installed it, but never got to the point of DHCP or telnet responding. I do not have a serial adapter for my DSL-G604t, so I am pretty much stumped at this point. None of the squashfs or jffs2 images gave me any indication of whether they were booting or not, with the LEDs coming on for power and LAN link immediately and never changing (except to flicker if I sent DHCP/ARP requests on the LAN). I even tried setting the MAC_PORT adam2 variable as suggested in a writeup for 2.4 kernels. I have no way of knowing if the ethernet is non-functional, or the root filesystem, etc.
I followed the instructions for writing the firmware via FTP to my adam2 version 22.2 firmware that I found on a writeup for another model (DSL-504?)... I used hexdump to find the beginning of the squashfs (or jffs2) filesystems and tried to adjust the partition map so that the kernel/fs split was at the right spot, and I wrote the entire image to the kernel+fs partition so that the image byte 0 starts at the beginning of the kernel partition (mtd1). I found the jffs2 magic "0x1985" (stored as 85 19) starting after padding at offset 0x000d0000 in my build, whereas the 'hsqs' in the squashfs starts at an odd alignment of 10 bytes in from 0x000c67a0, i.e. 0x000c67aa. Is there some extra padding around this value that I didn't take into account? I used these values as the boundary between mtd1 and mtd0.
Is there a known-good revision I should start with to validate that I can get an image to boot with ethernet access?