Geert Uytterhoeven writes: > On Sat, 26 Jan 2002, Albert D. Cahalan wrote:
>> Due to the damn framebuffer, a ppc memory tester will need to >> carry along a good sized chunk of the kernel. With memtest86, >> only the Linux boot code is used. > > Think `serial console' for your minimal kernel... You and I are among the 42 people capable of doing this. As a general solution, this isn't going to work. As for being minimal... why bother? If a ppc memory tester needs to carry along most of the baggage anyway, it might as well function as your regular kernel. Then the people without a serial console won't need to write down the errors by hand. They just do "cat /proc/memtest > errors" to save, once the boot is allowed to proceed. This could even be in the Debian kernel. If somebody is having trouble with the install crashing, they might be instructed to put "memtest" on the kernel command line. The installer could even write the necessary command line to avoid using such memory in the future. > And what about putting the test in the very early boot code, > so you can launch it from OF and use OF text output routines? That might be OK. In that case, one may hit OF assumptions about memory and MMU usage.
