The bad bridge mapping turned out to be an artifact of kernels that trusted the BIOS. Loading 2.2.19 and setting mem=509M in the lilo append made the system run. The BIOS reports the memory to kernel 2.4, and it also provides a map to the kernel where the I/O devices and bus bridges are located. These addresses overlapped. It was not sufficient to trim back memory in kernel 2.4. A kernel compiled for up to 64G RAM also gave in addition to the bad bridge mapping an inability to address ACPI. So one could call it a BIOS bug, or an oversight on the part of kernel developers to test every single notebook BIOS to see the reporting of memory maps was standardized, depending which side of the blame you are likely to stand on. Civileme
