Select Git revision
-
Davidlohr Bueso authoredFor a custom microbenchmark on a 3.30GHz Xeon SandyBridge, which calls IPC_STAT over and over, it was calculated that, on avg the cost of ipc_get_maxid() for increasing amounts of keys was: 10 keys: ~900 cycles 100 keys: ~15000 cycles 1000 keys: ~150000 cycles 10000 keys: ~2100000 cycles This is unsurprising as maxid is currently O(n). By having the max_id available in O(1) we save all those cycles for each semctl(_STAT) command, the idr_find can be expensive -- which some real (customer) workloads actually poll on. Note that this used to be the case, until commit 7ca7e564 ("ipc: store ipcs into IDRs"). The cost is the extra idr_find when doing RMIDs, but we simply go backwards, and should not take too many iterations to find the new value. [akpm@linux-foundation.org: coding-style fixes] Link: http://lkml.kernel.org/r/20170831172049.14576-5-dave@stgolabs.net Signed-off-by:
Davidlohr Bueso <dbueso@suse.de>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>Davidlohr Bueso authoredFor a custom microbenchmark on a 3.30GHz Xeon SandyBridge, which calls IPC_STAT over and over, it was calculated that, on avg the cost of ipc_get_maxid() for increasing amounts of keys was: 10 keys: ~900 cycles 100 keys: ~15000 cycles 1000 keys: ~150000 cycles 10000 keys: ~2100000 cycles This is unsurprising as maxid is currently O(n). By having the max_id available in O(1) we save all those cycles for each semctl(_STAT) command, the idr_find can be expensive -- which some real (customer) workloads actually poll on. Note that this used to be the case, until commit 7ca7e564 ("ipc: store ipcs into IDRs"). The cost is the extra idr_find when doing RMIDs, but we simply go backwards, and should not take too many iterations to find the new value. [akpm@linux-foundation.org: coding-style fixes] Link: http://lkml.kernel.org/r/20170831172049.14576-5-dave@stgolabs.net Signed-off-by:
Davidlohr Bueso <dbueso@suse.de>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
af_unix.c 52.56 KiB
/*
* NET4: Implementation of BSD Unix domain sockets.
*
* Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*
* Fixes:
* Linus Torvalds : Assorted bug cures.
* Niibe Yutaka : async I/O support.
* Carsten Paeth : PF_UNIX check, address fixes.
* Alan Cox : Limit size of allocated blocks.
* Alan Cox : Fixed the stupid socketpair bug.
* Alan Cox : BSD compatibility fine tuning.
* Alan Cox : Fixed a bug in connect when interrupted.
* Alan Cox : Sorted out a proper draft version of
* file descriptor passing hacked up from
* Mike Shaver's work.
* Marty Leisner : Fixes to fd passing
* Nick Nevin : recvmsg bugfix.
* Alan Cox : Started proper garbage collector
* Heiko EiBfeldt : Missing verify_area check
* Alan Cox : Started POSIXisms
* Andreas Schwab : Replace inode by dentry for proper
* reference counting
* Kirk Petersen : Made this a module
* Christoph Rohland : Elegant non-blocking accept/connect algorithm.
* Lots of bug fixes.
* Alexey Kuznetosv : Repaired (I hope) bugs introduces
* by above two patches.
* Andrea Arcangeli : If possible we block in connect(2)
* if the max backlog of the listen socket
* is been reached. This won't break
* old apps and it will avoid huge amount
* of socks hashed (this for unix_gc()
* performances reasons).
* Security fix that limits the max
* number of socks to 2*max_files and
* the number of skb queueable in the
* dgram receiver.
* Artur Skawina : Hash function optimizations
* Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
* Malcolm Beattie : Set peercred for socketpair
* Michal Ostrowski : Module initialization cleanup.
* Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
* the core infrastructure is doing that
* for all net proto families now (2.5.69+)
*
*
* Known differences from reference BSD that was tested:
*
* [TO FIX]
* ECONNREFUSED is not returned from one end of a connected() socket to the
* other the moment one end closes.
* fstat() doesn't return st_dev=0, and give the blksize as high water mark
* and a fake inode identifier (nor the BSD first socket fstat twice bug).
* [NOT TO FIX]
* accept() returns a path name even if the connecting socket has closed
* in the meantime (BSD loses the path and gives up).
* accept() returns 0 length path for an unbound connector. BSD returns 16
* and a null first byte in the path (but not for gethost/peername - BSD bug ??)
* socketpair(...SOCK_RAW..) doesn't panic the kernel.
* BSD af_unix apparently has connect forgetting to block properly.
* (need to check this with the POSIX spec in detail)
*
* Differences from 2.0.0-11-... (ANK)
* Bug fixes and improvements.