From: Mark Wielaard <mark@klomp.org>
To: Jonathon Anderson <jma14@rice.edu>, elfutils-devel@sourceware.org
Cc: Srdan Milakovic <sm108@rice.edu>
Subject: Re: [PATCH] libdw: add thread-safety to dwarf_getabbrev()
Date: Wed, 21 Aug 2019 11:16:00 -0000 [thread overview]
Message-ID: <410becd068420cbd3b93c161be2084c6a5d2f362.camel@klomp.org> (raw)
In-Reply-To: <1565983469.1826.0@smtp.mail.rice.edu>
Hi Jonathon and Srđan,
On Fri, 2019-08-16 at 14:24 -0500, Jonathon Anderson wrote:
> For parallel applications that need the information in the DIEs, the
> Dwarf_Abbrev hash table et al. become a massive data race. This fixes
> that by:
>
> 1. Adding atomics & locks to the hash table to manage concurrency
> (lib/dynamicsizehash_concurrent.{c,h})
> 2. Adding a lock & array structure to the memory manager (pseudo-TLS)
> (libdwP.h, libdw_alloc.c)
> 3. Adding extra configure options for Helgrind/DRD annotations
> (configure.ac)
> 4. Including "stdatomic.h" from FreeBSD, to support C11-style atomics.
> (lib/stdatomic.h)
This looks like really nice work. Thanks!
I am splitting review in some smaller parts if you don't mind.
Simply because it is large and I cannot keep everything in my head at
once :) But here some initial overall comments.
> Notes:
> - GCC >= 4.9 provides <stdatomic.h> natively; for those versions
> lib/stdatomic.h could be removed or disabled. We can also rewrite the
> file if the copyright becomes an issue.
If the compiler provides stdatomic.h then I think it would be good to
use that instead of our own implementation. The copyright isn't a
problem. But do you have a reference/URL to the upstream version? I
like to add that somewhere, so we can sync with it in the future. I see
various commented out parts. Was that already upstream? Should we just
remove those parts?
> - Currently the concurrent hash table is always enabled,
> performance-wise there is no known difference between it
> and the non-concurrent version.
> This can be changed to toggle with --enable-thread-safety
> if preferred.
I would prefer it always enabled, unless there is a massive slowdown of
the single-threaded case. The problem with --enable-thread-safety is
that it is a) known broken (sigh) and b) it basically introduces two
separate libraries that behave subtly differently. I would very much
like to get rid of --enable-thread-safety by fixing the broken locking
and simply making it the default.
> - Another implementation of #2 above might use dynamic TLS
> (pthread_key_*),
> we chose this implementation to reduce the overall complexity.
Are there any other trade-offs?
Thanks,
Mark
next prev parent reply other threads:[~2019-08-21 11:16 UTC|newest]
Thread overview: 55+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-08-16 19:24 Jonathon Anderson
2019-08-21 11:16 ` Mark Wielaard [this message]
2019-08-21 14:21 ` Jonathon Anderson
2019-08-23 21:22 ` Mark Wielaard
[not found] ` <1566396518.5389.0@smtp.mail.rice.edu>
2019-08-23 18:25 ` Mark Wielaard
2019-08-23 22:36 ` Jonathon Anderson
2019-08-21 21:50 ` Mark Wielaard
2019-08-21 22:01 ` Mark Wielaard
2019-08-21 22:21 ` Jonathon Anderson
2019-08-23 21:26 ` Mark Wielaard
2019-08-24 23:24 ` Mark Wielaard
2019-08-25 1:11 ` Jonathon Anderson
2019-08-25 10:05 ` Mark Wielaard
2019-08-26 1:25 ` Jonathon Anderson
2019-08-26 13:18 ` Mark Wielaard
2019-08-26 13:37 ` Jonathon Anderson
2019-08-27 3:52 ` Jonathon Anderson
2019-08-29 13:16 ` Mark Wielaard
2019-08-29 13:16 ` [PATCH 2/3] libdw: Rewrite the memory handler to be thread-safe Mark Wielaard
2019-10-21 16:13 ` Mark Wielaard
2019-10-21 16:28 ` Jonathon Anderson
2019-10-21 18:00 ` Mark Wielaard
2019-10-24 16:47 ` Mark Wielaard
2019-08-29 13:16 ` [PATCH 3/3] lib + libdw: Add and use a concurrent version of the dynamic-size hash table Mark Wielaard
2019-10-25 23:50 ` Mark Wielaard
2019-10-26 4:11 ` Jonathon Anderson
2019-10-27 16:13 ` Mark Wielaard
2019-10-27 17:49 ` Jonathon Anderson
2019-10-28 14:08 ` Mark Wielaard
2019-10-28 20:12 ` Mark Wielaard
2019-11-04 16:21 ` Mark Wielaard
2019-11-04 16:19 ` Mark Wielaard
2019-11-04 17:03 ` [PATCH] " Jonathon Anderson
2019-11-07 11:07 ` Mark Wielaard
2019-11-07 15:25 ` Jonathon Anderson
2019-11-08 14:07 ` Mark Wielaard
2019-11-08 15:29 ` Jonathon Anderson
2019-11-10 23:24 ` Mark Wielaard
2019-11-11 23:38 ` Jonathon Anderson
2019-11-12 21:45 ` Mark Wielaard
2019-08-29 13:16 ` [PATCH 1/3] Add some supporting framework for C11-style atomics Mark Wielaard
2019-10-22 16:31 ` Mark Wielaard
2019-10-26 10:54 ` [PATCH] libdw: add thread-safety to dwarf_getabbrev() Florian Weimer
2019-10-26 12:06 ` Mark Wielaard
2019-10-26 16:14 ` Florian Weimer
2019-10-26 16:45 ` Jonathon Anderson
2019-10-26 16:50 ` Florian Weimer
2019-10-26 22:53 ` Mark Wielaard
2019-10-27 8:59 ` Florian Weimer
2019-10-27 18:11 ` Jonathon Anderson
2019-10-27 18:44 ` Florian Weimer
2019-10-26 22:50 ` Mark Wielaard
2019-10-27 0:56 ` Jonathon Anderson
2019-10-28 13:26 ` Mark Wielaard
2019-10-28 15:32 ` Jonathon Anderson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=410becd068420cbd3b93c161be2084c6a5d2f362.camel@klomp.org \
--to=mark@klomp.org \
--cc=elfutils-devel@sourceware.org \
--cc=jma14@rice.edu \
--cc=sm108@rice.edu \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).