Support custom Linux kernel target for split/raw ramdump memory segments

[miniLQ]

Hi, thanks for maintaining drgn.

I am working on Android/Linux kernel ramdump analysis where the dump is not available as a single ELF vmcore. Instead, the platform provides a split/raw ramdump, usually several memory segment files plus physical base addresses, for example:

DDRCS0.BIN @ 0x80000000
DDRCS1.BIN @ 0x100000000
...

The normal drgn -c vmcore -s vmlinux flow works well for standard vmcore files, but it does not directly apply to this split dump format.

I noticed that drgn has APIs such as:

prog.add_memory_segment(..., physical=True)
prog.set_linux_kernel_custom(vmcoreinfo, is_live=False)

This looks like it may be possible to build a custom backend for split/raw ramdumps by registering each physical memory segment manually, then letting drgn perform Linux kernel virtual address translation using vmcoreinfo.

I would like to ask:

1. Is this the intended API path for supporting split/raw kernel ramdumps?
2. What is the minimal required vmcoreinfo content for arm64 Linux kernel virtual address translation?
3. Is there any example or test case showing set_linux_kernel_custom() with manually added physical memory segments?
4. Are there any known limitations compared with loading a standard ELF vmcore via -c?

A rough target use case would be:

from drgn import Program, Platform, Architecture

prog = Program(
    Platform(Architecture.AARCH64),
    vmcoreinfo=vmcoreinfo_text,
)

for segment in segments:
    prog.add_memory_segment(
        segment.phys_start,
        segment.size,
        segment.read_fn,
        physical=True,
    )

prog.set_linux_kernel_custom(vmcoreinfo_text, is_live=False)
prog.load_debug_info([vmlinux_path])

print(prog["init_task"].comm)

If this is expected to work, I would be happy to experiment and possibly contribute documentation or a small example for split/raw ramdump support.

Thanks!