I'm trying to understand ARM VM hardware. A paragraph or two of
explanation may help explain why.
I've got a DNARD and a Mesa 4C81. The former works fine. The latter
doesn't. Years ago, doing a tarball extract on the 4C81 booted
diskless, I ended up with corruption in the extracted files. Recently,
I played with it a bit and created a kernel that did rudimentary memory
tests (mapping in each free page, writing to it, unmapping it, then do
the same only read-and-check instead of writing - trusting to the
amount of writing to provoke any necessary cache flushes). This found
two stuck-at bits in the 4C81's RAM. I then excluded the containing
pages from the memory map and it mostly works.
But it crashes every now and then. Attempts to build the world have
paniced twice, with different errors at different places. This leads
me to suspect there's at least one memory bit that's more sutbly flaky.
(Admittedly, it could be flakiness elsewhere instead, but with two
definitely-bad bits, more bad bits would be unsurprising.) So I want
to build a more stringent memory tester.
But I'd like to make it as small as feasible. Loading a whole kernel
just to test RAM is rather severe overkill, not to mention occupying a
lot more RAM than is needed for the task.
So, I want to understand the VM hardware. I'm going to have to fiddle
with it, and that's a lot harder if I don't know what sort of VM
hardware it's using. I've got the ARMv6 architecture manual from 2005,
but that isn't as clear as I'd like on the pre-v6 architectures. The
4C81 kernel config I've been using says CPU_ARM9, but as far as I can
tell ARM isn't up above v6, so I suspect there are multiple different
numbering schemes here.
Can anyone help me clear up the confusion?
/~\ The ASCII Mouse
\ / Ribbon Campaign
X Against HTML mouse%rodents-montreal.org@localhost
/ \ Email! 7D C8 61 52 5D E7 2D 39 4E F1 31 3E E8 B3 27 4B