Re: [PATCH makedumpfile 0/9] Improvements to makedumpfile extensions, plus userspace stack tracing extension

[Stephen Brennan <stephen.s.brennan@oracle.com>]

Hello all,

Just a gentle ping on this series.

Thanks,
Stephen

Stephen Brennan <stephen.s.brennan@oracle.com> writes:
> Hello all,
>
> Building on Tao Liu's excellent work for makedumpfile extensions, I'd like to
> share some incremental improvements to the extension system, as well as my own
> two extensions which enable userspace stack tracing. There are three topics
> related to the core makedumpfile extension system which are addressed here:
>
> 1. Tail pages handling. As patch 1 explains, extensions currently get called for
> tail pages, but their decisions are not respected. I believe the best approach
> is to only call extensions for the head page, but allow extensions to return a
> decision related to just a subset of the pages. To this end, patch 7 adds
> PG_INCLUDE_HEAD to include just the head page (leaving the rest up to the
> default policy), and patch 9 shows the elfheader extension which uses that
> functionality.
>
> 2. Counting pages included by extensions. This is done by patch 4. It alters the
> logic because it's really helpful to know whether a page would have been
> included by makedumpfile's dump-level filtering anyway, so we only count the
> additional pages.
>
> 3. Sharing makedumpfile's logic and knowledge about page metadata. Currently,
> extensions just get a "pcache" pointer, which means they must redo whatever
> logic makedumpfile has already done, in order to get information like
> compound_order. Patches 3, 5, and 6 factor out the page metadata into a new
> struct pginfo, which gets passed to extensions, and it also makes more of the
> helpers like "isSlab()" usable with this pointer.
>
> With those improvements, the final two patches (8 & 9) implement a pair of
> extensions that are useful for including a minimal amount of information which
> can be used to do stack traces of the userspace processes in a vmcore. The
> "userstack" extension includes the stack memory itself for each thread, and the
> "elfheader" extension includes the first page of mapped ELF files, so that ELF
> build IDs can be identified from the vmcore.
>
> My testing so far:
> - Live testing (/proc/kcore) x86_64 on Oracle UEK 8 (6.12.x based)
> - Refiltering on vmcores generated from UEK8, 7 (5.15.x based), 6 (5.4.x based)
> - Using the resulting filtered vmcores with drgn's contrib/pstack.py to generate
>   userspace stack traces.
>
> I still have significantly more testing to do and then share out:
> - Testing in kexec environment
> - Testing with upstream kernels, also verifying in a Fedora userspace
> - Sharing runtime and size comparisons with different workloads
>
> Despite the limited testing, I wanted to share these patches as they are,
> because I think the improvements for the core extensions framework are useful
> already, and don't require as much verification.
>
> For the extensions themselves, I would like to consider merging them as well, if
> there is interest. I think it would be useful for makedumpfile to contain shared
> extensions, including those for the GPU buffers which Tao Liu was working on, as
> well as the userspace stack tracing ones which I'm sharing.
>
> Thanks,
> Stephen
>
> Stephen Brennan (9):
>   Do not call extensions for tail pages
>   Honor CFLAGS in extension/Makefile
>   Share page information with extension callbacks
>   Introduce a stat for pages retained by extension
>   Move page checks into makedumpfile.h
>   Simplify arguments for page checks
>   Add PG_INCLUDE_HEAD extension return status
>   Add userstack extension
>   Add elfheader extension
>
>  extension.c             |  10 +-
>  extension.h             |  10 +-
>  extensions/Makefile     |   8 +-
>  extensions/elfheader.c  |  93 ++++++++++++
>  extensions/userstack.c  | 325 ++++++++++++++++++++++++++++++++++++++++
>  extensions/vma_mtree.c  | 140 +++++++++++++++++
>  extensions/vma_mtree.h  |   7 +
>  extensions/vma_rbtree.c |  56 +++++++
>  extensions/vma_rbtree.h |  12 ++
>  makedumpfile.c          | 188 +++++++++--------------
>  makedumpfile.h          |  78 +++++++++-
>  11 files changed, 798 insertions(+), 129 deletions(-)
>  create mode 100644 extensions/elfheader.c
>  create mode 100644 extensions/userstack.c
>  create mode 100644 extensions/vma_mtree.c
>  create mode 100644 extensions/vma_mtree.h
>  create mode 100644 extensions/vma_rbtree.c
>  create mode 100644 extensions/vma_rbtree.h
>
> -- 
> 2.47.3