mirrors/emacs
1
Fork 0
mirror of git://git.sv.gnu.org/emacs.git synced 2026-04-27 16:51:06 -07:00
Code Issues Projects Releases Wiki Activity

Mps wiki article on gc: brief notes on protection and incremental collection

Browse source 
Copied from Perforce
 Change: 159264
 ServerID: perforce.ravenbrook.com
This commit is contained in:
Richard Kistruck 2006-06-16 12:17:15 +01:00
parent 54aa9ab652
commit ed07a5e2ae
1 changed files with 54 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

57
mps/manual/wiki/gc.html
View file

@ -46,7 +46,7 @@
<h2>Introduction</h2>
<p>The essential MPS concepts for GC are:</p>
<h3>The essential MPS concepts for GC are:</h3>
<dl>
<dt> <a href="#Format">Format&darr;</a></dt>
<dd> lets the MPS ask the client a question about an object;</dd>
@ -55,8 +55,15 @@
<dd> tell the MPS which things (eg. stack, globals) are always alive.</dd>
</dl>
<p>The advanced MPS concepts for GC are:</p>
<h3>The advanced MPS concepts for GC are:</h3>
<dl>
<dt> <a href="#Protection">Protection and incremental collection&darr;</a></dt>
<dd> MPS sets operating-system read- or write-barriers on certain
pages. When client tries to use this protected memory, the MPS
interrupts and does just enough garbage-collection to allow the
client to see this page.
</dd>
<dt> Allocation Point
<a href="#Allocation-Point-User">(user's guide&darr;)</a></dt>
<dd> reserve memory, build a new object, and commit
@ -220,10 +227,53 @@ AMCFix() (and now AMCHeaderFix() too).</p>
<h2><a id="Protection">Protection and incremental collection</a></h2>
<p>These notes on Protection and incremental collection were written
by RHSK between 2006-06-15 and 2006-06-15, following research and
discussion with RB and DRJ.</p>
<p>The MPS can set operating-system read- or write-barriers on certain
pages. When client tries to use this protected memory, the MPS
interrupts and does just enough garbage-collection to allow the
client to see this page.</p>
<p>On Windows, currently, these barriers are implemented using SEH -- Structured Exception Handling.</p>
<blockquote><pre>
<a href="http://info.ravenbrook.com/mail/2006/05/17/11-03-54/0.txt">//info.ravenbrook.com/mail/2006/05/17/11-03-54/0.txt</a>
"chat with DRJ RB RHSK about work for Configura, and other MPS issues"
...
MPS should be part of runtime for a language.
One necessary part is SEH:
- required once per thread;
- AND required on foreign function calls inward.
Otherwise,
1. some foreign C++ might steal our exception;
2. this might be a thread we've never seen before
(In fact, client need to have a bit standard of code that works out
if the MPS has seen this thread before. It must call this code on
every inbound call from foreign code.) DRJ: "Dylan does this".
There's a bunch of assumes about compiled code, ABI etc, play nice
with sighandler etc.
Might be poss for GR to geneerate code that doesn't use SEH if he
wasn't using MPS.
MPS has a requirement that you use this extremely poorly documented
SEH interface.
2 improvements:
1. better documentation for this
2. try to make the MPS *not* require SEH
job001360 -- SEH
</pre></blockquote>
<h2><a id="Allocation-Point-User">Allocation Points -- User's Guide</a></h2>
<p>These notes on Allocation Points were written by RHSK between
2006-06-07 and 2006-06-12, following research and discussion with
2006-06-07 and 2006-06-15, following research and discussion with
RB and NB.</p>
<p><strong>Note:</strong> the text in this "Allocation Point"
@ -987,6 +1037,7 @@ RankEXACT now, come across another ref to child, and move it.
2006-06-15 RHSK Allocation Points: clarifications and further notes
into User Guide and Internals, including initializing
as-yet unused parts of new objects to reduce spoofing.
2006-06-15 RHSK Brief notes on Protection and incremental collection.
</pre>