Re: [PATCH v2 1/2] contigmem: support including mapped buffers in core dump

[Lewis Donzis <lew@perftech.com>]

It seems unlikely that there are other users of contigmem on FreeBSD, especially since it's not necessary for other applications.

On FreeBSD, use of large and huge pages is automatic; you just call mmap() with a large size and it automatically tries to use the largest physical pages possible.  So the only thing contigmem is doing for us on FreeBSD is providing the physical address and, of course, making it consistent with Linux.

At least, that's my understanding.


----- On Oct 28, 2024, at 8:26 AM, Dmitry Kozlyuk dmitry.kozliuk@gmail.com wrote:

> 2024-10-26 06:43 (UTC-0500), Lewis Donzis:
>> Is the extra control necessary, i.e., why not just always do this and let
>> the EAL option control whether the pages get dumped?
> 
> I've been evaluating your suggestion and see no downsides,
> except contigmem default behavior change, but does it have non-DPDK users?
> If no one objects, I'll prepare v3 doing the following:
> 1) everything from v2,
> 2) except always mark contigmem buffers as dumpable,
> 3) add --dump-mapped back and make DPDK disable dumping by default.
> As a result, in order to include mapped memory in coredump:
> * FreeBSD will require only "--dump-mapped";
> * Linux will require both "coredump_filter" setup and "--dump-mapped".