Re: [PATCH 07/10] gdb: add qMachineId packet

[Thiago Jung Bauermann <thiago.bauermann@linaro.org>]

Hello Andrew,

This is a very interesting feature, thanks. For now at least, I have
just one comment on the machine-id for Linux:

Andrew Burgess via Gdb-patches <gdb-patches@sourceware.org> writes:

> Generating a suitable machine-id is, I think, always going to be
> target specific.  As such, I've structured the code in a way that
> allows different targets to provide their own implementations, but
> I've only implemented a solution for the Linux targets.
>
> The reply to a qMachineId packet looks like this:
>
>   predicate;key=value[;key=value]*
>
> the idea being that the reply consists of a number of key/value pairs,
> each of which must match in order for GDB to consider the machine-id a
> match.  I currently propose just two keys:
>
>   linux-boot-id - this returns the value from the file
>   /proc/sys/kernel/random/boot_id, which, if I understand correctly,
>   should be unique(ish) for each boot of each machine, and
>
>   cuserid - this returns the value of the cuserid call.
>
> My thinking is that if we know we are on the same machine (thanks to
> linux-boot-id), and we know we are the same effective user (thanks to
> cuserid) then there's a pretty good chance that GDB and the remote can
> access the same set of files.

Unfortunately the two keys above won't detect when GDB and gdbserver are
running on the same machine but in different containers. I suggest
adding a new key for that:

  mountinfo - The hash of the contents of the target's
  /proc/self/mountinfo file.

The hash algorithm could be either SHA1 or MD5, which have
implementations in libiberty.

Disclaimer: I'm no expert in Linux containers, so I wouldn't be
surprised if there were a better way to differentiate containers (or,
more specifically for this use case, filesystem namespaces).

-- 
Thiago