aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation/RCU
diff options
context:
space:
mode:
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>2014-02-15 11:52:32 -0500
committerPaul E. McKenney <paulmck@linux.vnet.ibm.com>2014-02-17 17:56:10 -0500
commite4696a1d3b1125d427de685531a44258ea6263df (patch)
tree1cf8e3104b8ee9205d782bd24f3b31c66c3a540c /Documentation/RCU
parent9b2b3bf53124dca4ac815bd2fca53a31e5e262bd (diff)
documentation: Fix some inconsistencies in RTFP.txt
Some of the early history leaves out some citations and vice versa. This commit fixes these up. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Diffstat (limited to 'Documentation/RCU')
-rw-r--r--Documentation/RCU/RTFP.txt149
1 files changed, 125 insertions, 24 deletions
diff --git a/Documentation/RCU/RTFP.txt b/Documentation/RCU/RTFP.txt
index 273e654d7d08..2f0fcb2112d2 100644
--- a/Documentation/RCU/RTFP.txt
+++ b/Documentation/RCU/RTFP.txt
@@ -31,6 +31,14 @@ has lapsed, so this approach may be used in non-GPL software, if desired.
31(In contrast, implementation of RCU is permitted only in software licensed 31(In contrast, implementation of RCU is permitted only in software licensed
32under either GPL or LGPL. Sorry!!!) 32under either GPL or LGPL. Sorry!!!)
33 33
34In 1987, Rashid et al. described lazy TLB-flush [RichardRashid87a].
35At first glance, this has nothing to do with RCU, but nevertheless
36this paper helped inspire the update-side batching used in the later
37RCU implementation in DYNIX/ptx. In 1988, Barbara Liskov published
38a description of Argus that noted that use of out-of-date values can
39be tolerated in some situations. Thus, this paper provides some early
40theoretical justification for use of stale data.
41
34In 1990, Pugh [Pugh90] noted that explicitly tracking which threads 42In 1990, Pugh [Pugh90] noted that explicitly tracking which threads
35were reading a given data structure permitted deferred free to operate 43were reading a given data structure permitted deferred free to operate
36in the presence of non-terminating threads. However, this explicit 44in the presence of non-terminating threads. However, this explicit
@@ -41,11 +49,11 @@ providing a fine-grained locking design, however, it would be interesting
41to see how much of the performance advantage reported in 1990 remains 49to see how much of the performance advantage reported in 1990 remains
42today. 50today.
43 51
44At about this same time, Adams [Adams91] described ``chaotic relaxation'', 52At about this same time, Andrews [Andrews91textbook] described ``chaotic
45where the normal barriers between successive iterations of convergent 53relaxation'', where the normal barriers between successive iterations
46numerical algorithms are relaxed, so that iteration $n$ might use 54of convergent numerical algorithms are relaxed, so that iteration $n$
47data from iteration $n-1$ or even $n-2$. This introduces error, 55might use data from iteration $n-1$ or even $n-2$. This introduces
48which typically slows convergence and thus increases the number of 56error, which typically slows convergence and thus increases the number of
49iterations required. However, this increase is sometimes more than made 57iterations required. However, this increase is sometimes more than made
50up for by a reduction in the number of expensive barrier operations, 58up for by a reduction in the number of expensive barrier operations,
51which are otherwise required to synchronize the threads at the end 59which are otherwise required to synchronize the threads at the end
@@ -55,7 +63,8 @@ is thus inapplicable to most data structures in operating-system kernels.
55 63
56In 1992, Henry (now Alexia) Massalin completed a dissertation advising 64In 1992, Henry (now Alexia) Massalin completed a dissertation advising
57parallel programmers to defer processing when feasible to simplify 65parallel programmers to defer processing when feasible to simplify
58synchronization. RCU makes extremely heavy use of this advice. 66synchronization [HMassalinPhD]. RCU makes extremely heavy use of
67this advice.
59 68
60In 1993, Jacobson [Jacobson93] verbally described what is perhaps the 69In 1993, Jacobson [Jacobson93] verbally described what is perhaps the
61simplest deferred-free technique: simply waiting a fixed amount of time 70simplest deferred-free technique: simply waiting a fixed amount of time
@@ -90,27 +99,29 @@ mechanism, which is quite similar to RCU [Gamsa99]. These operating
90systems made pervasive use of RCU in place of "existence locks", which 99systems made pervasive use of RCU in place of "existence locks", which
91greatly simplifies locking hierarchies and helps avoid deadlocks. 100greatly simplifies locking hierarchies and helps avoid deadlocks.
92 101
932001 saw the first RCU presentation involving Linux [McKenney01a] 102The year 2000 saw an email exchange that would likely have
94at OLS. The resulting abundance of RCU patches was presented the 103led to yet another independent invention of something like RCU
95following year [McKenney02a], and use of RCU in dcache was first 104[RustyRussell2000a,RustyRussell2000b]. Instead, 2001 saw the first
96described that same year [Linder02a]. 105RCU presentation involving Linux [McKenney01a] at OLS. The resulting
106abundance of RCU patches was presented the following year [McKenney02a],
107and use of RCU in dcache was first described that same year [Linder02a].
97 108
98Also in 2002, Michael [Michael02b,Michael02a] presented "hazard-pointer" 109Also in 2002, Michael [Michael02b,Michael02a] presented "hazard-pointer"
99techniques that defer the destruction of data structures to simplify 110techniques that defer the destruction of data structures to simplify
100non-blocking synchronization (wait-free synchronization, lock-free 111non-blocking synchronization (wait-free synchronization, lock-free
101synchronization, and obstruction-free synchronization are all examples of 112synchronization, and obstruction-free synchronization are all examples of
102non-blocking synchronization). In particular, this technique eliminates 113non-blocking synchronization). The corresponding journal article appeared
103locking, reduces contention, reduces memory latency for readers, and 114in 2004 [MagedMichael04a]. This technique eliminates locking, reduces
104parallelizes pipeline stalls and memory latency for writers. However, 115contention, reduces memory latency for readers, and parallelizes pipeline
105these techniques still impose significant read-side overhead in the 116stalls and memory latency for writers. However, these techniques still
106form of memory barriers. Researchers at Sun worked along similar lines 117impose significant read-side overhead in the form of memory barriers.
107in the same timeframe [HerlihyLM02]. These techniques can be thought 118Researchers at Sun worked along similar lines in the same timeframe
108of as inside-out reference counts, where the count is represented by the 119[HerlihyLM02]. These techniques can be thought of as inside-out reference
109number of hazard pointers referencing a given data structure rather than 120counts, where the count is represented by the number of hazard pointers
110the more conventional counter field within the data structure itself. 121referencing a given data structure rather than the more conventional
111The key advantage of inside-out reference counts is that they can be 122counter field within the data structure itself. The key advantage
112stored in immortal variables, thus allowing races between access and 123of inside-out reference counts is that they can be stored in immortal
113deletion to be avoided. 124variables, thus allowing races between access and deletion to be avoided.
114 125
115By the same token, RCU can be thought of as a "bulk reference count", 126By the same token, RCU can be thought of as a "bulk reference count",
116where some form of reference counter covers all reference by a given CPU 127where some form of reference counter covers all reference by a given CPU
@@ -123,8 +134,10 @@ can be thought of in other terms as well.
123 134
124In 2003, the K42 group described how RCU could be used to create 135In 2003, the K42 group described how RCU could be used to create
125hot-pluggable implementations of operating-system functions [Appavoo03a]. 136hot-pluggable implementations of operating-system functions [Appavoo03a].
126Later that year saw a paper describing an RCU implementation of System 137Later that year saw a paper describing an RCU implementation
127V IPC [Arcangeli03], and an introduction to RCU in Linux Journal 138of System V IPC [Arcangeli03] (following up on a suggestion by
139Hugh Dickins [Dickins02a] and an implementation by Mingming Cao
140[MingmingCao2002IPCRCU]), and an introduction to RCU in Linux Journal
128[McKenney03a]. 141[McKenney03a].
129 142
1302004 has seen a Linux-Journal article on use of RCU in dcache 1432004 has seen a Linux-Journal article on use of RCU in dcache
@@ -383,6 +396,21 @@ for Programming Languages and Operating Systems}"
383} 396}
384} 397}
385 398
399@phdthesis{HMassalinPhD
400,author="H. Massalin"
401,title="Synthesis: An Efficient Implementation of Fundamental Operating
402System Services"
403,school="Columbia University"
404,address="New York, NY"
405,year="1992"
406,annotation={
407 Mondo optimizing compiler.
408 Wait-free stuff.
409 Good advice: defer work to avoid synchronization. See page 90
410 (PDF page 106), Section 5.4, fourth bullet point.
411}
412}
413
386@unpublished{Jacobson93 414@unpublished{Jacobson93
387,author="Van Jacobson" 415,author="Van Jacobson"
388,title="Avoid Read-Side Locking Via Delayed Free" 416,title="Avoid Read-Side Locking Via Delayed Free"
@@ -671,6 +699,20 @@ Orran Krieger and Rusty Russell and Dipankar Sarma and Maneesh Soni"
671[Viewed October 18, 2004]" 699[Viewed October 18, 2004]"
672} 700}
673 701
702@conference{Michael02b
703,author="Maged M. Michael"
704,title="High Performance Dynamic Lock-Free Hash Tables and List-Based Sets"
705,Year="2002"
706,Month="August"
707,booktitle="{Proceedings of the 14\textsuperscript{th} Annual ACM
708Symposium on Parallel
709Algorithms and Architecture}"
710,pages="73-82"
711,annotation={
712Like the title says...
713}
714}
715
674@Conference{Linder02a 716@Conference{Linder02a
675,Author="Hanna Linder and Dipankar Sarma and Maneesh Soni" 717,Author="Hanna Linder and Dipankar Sarma and Maneesh Soni"
676,Title="Scalability of the Directory Entry Cache" 718,Title="Scalability of the Directory Entry Cache"
@@ -727,6 +769,24 @@ Andrea Arcangeli and Andi Kleen and Orran Krieger and Rusty Russell"
727} 769}
728} 770}
729 771
772@conference{Michael02a
773,author="Maged M. Michael"
774,title="Safe Memory Reclamation for Dynamic Lock-Free Objects Using Atomic
775Reads and Writes"
776,Year="2002"
777,Month="August"
778,booktitle="{Proceedings of the 21\textsuperscript{st} Annual ACM
779Symposium on Principles of Distributed Computing}"
780,pages="21-30"
781,annotation={
782 Each thread keeps an array of pointers to items that it is
783 currently referencing. Sort of an inside-out garbage collection
784 mechanism, but one that requires the accessing code to explicitly
785 state its needs. Also requires read-side memory barriers on
786 most architectures.
787}
788}
789
730@unpublished{Dickins02a 790@unpublished{Dickins02a
731,author="Hugh Dickins" 791,author="Hugh Dickins"
732,title="Use RCU for System-V IPC" 792,title="Use RCU for System-V IPC"
@@ -735,6 +795,17 @@ Andrea Arcangeli and Andi Kleen and Orran Krieger and Rusty Russell"
735,note="private communication" 795,note="private communication"
736} 796}
737 797
798@InProceedings{HerlihyLM02
799,author={Maurice Herlihy and Victor Luchangco and Mark Moir}
800,title="The Repeat Offender Problem: A Mechanism for Supporting Dynamic-Sized,
801Lock-Free Data Structures"
802,booktitle={Proceedings of 16\textsuperscript{th} International
803Symposium on Distributed Computing}
804,year=2002
805,month="October"
806,pages="339-353"
807}
808
738@unpublished{Sarma02b 809@unpublished{Sarma02b
739,Author="Dipankar Sarma" 810,Author="Dipankar Sarma"
740,Title="Some dcache\_rcu benchmark numbers" 811,Title="Some dcache\_rcu benchmark numbers"
@@ -749,6 +820,19 @@ Andrea Arcangeli and Andi Kleen and Orran Krieger and Rusty Russell"
749} 820}
750} 821}
751 822
823@unpublished{MingmingCao2002IPCRCU
824,Author="Mingming Cao"
825,Title="[PATCH]updated ipc lock patch"
826,month="October"
827,year="2002"
828,note="Available:
829\url{https://lkml.org/lkml/2002/10/24/262}
830[Viewed February 15, 2014]"
831,annotation={
832 Mingming Cao's patch to introduce RCU to SysV IPC.
833}
834}
835
752@unpublished{LinusTorvalds2003a 836@unpublished{LinusTorvalds2003a
753,Author="Linus Torvalds" 837,Author="Linus Torvalds"
754,Title="Re: {[PATCH]} small fixes in brlock.h" 838,Title="Re: {[PATCH]} small fixes in brlock.h"
@@ -982,6 +1066,23 @@ Realtime Applications"
982} 1066}
983} 1067}
984 1068
1069@article{MagedMichael04a
1070,author="Maged M. Michael"
1071,title="Hazard Pointers: Safe Memory Reclamation for Lock-Free Objects"
1072,Year="2004"
1073,Month="June"
1074,journal="IEEE Transactions on Parallel and Distributed Systems"
1075,volume="15"
1076,number="6"
1077,pages="491-504"
1078,url="Available:
1079\url{http://www.research.ibm.com/people/m/michael/ieeetpds-2004.pdf}
1080[Viewed March 1, 2005]"
1081,annotation={
1082 New canonical hazard-pointer citation.
1083}
1084}
1085
985@phdthesis{PaulEdwardMcKenneyPhD 1086@phdthesis{PaulEdwardMcKenneyPhD
986,author="Paul E. McKenney" 1087,author="Paul E. McKenney"
987,title="Exploiting Deferred Destruction: 1088,title="Exploiting Deferred Destruction: