harvey won't boot with initrd on real hardware

[gmacd]

Repro steps:

• build harvey as usual
• build sys/src/u-root.json
• edit your pxe file ('default' in my case) and append --- uroot.cpio.lzma to the APPEND line)
• run centre, boot and see #P/uroot.cpio.lzma exists

If you have the same problem I had, it'll pagefault on boot.

The initrd memory area is recognised by multiboot, and identified as PamMODULE as expected. It's placed after the kernel heap, but not immediately after - there's nearly 6Mi (5976064 bytes) of PamMEMORY in between the kernel heap and the module.

The problem seems to be in umeminit - when it's setting up the memory available for user space, it pagefaults when initialising the PamMEMORY area in between the kernel heap and the initrd module (in the plop function).

On real hardware, without the initrd module it boots fine. Qemu boots fine with or without the initrd.

[gmacd]

So I've got some ideas about the page fault with initrd on real hardware:

• In umeminit, setphysmembounds is called to set the user memory area to start from the start of the last PamMODULE section.
• umeminit then calls physinit to initialise every physical memory section of type PamMEMORY that's larger than 2Mi.
• The problem is that in my real hardware example, there's a PamMEMORY section larger than 2Mi between the kernel and the initrd module. This is considered outside the physical memory bound (in setphysmembounds), but we still try to call physinit for it.

A quick fix would be to not call physinit for any memory section before the module sections (if there is one). In this case you'd lose 4Mi of available ram (the actual size is just under 6Mi).