Miklos Tirpak [Tue, 19 Jun 2007 13:09:35 +0000 (13:09 +0000)]
- Local UAC related functions do not register the TMCB_LOCAL_COMPLETED and TMCB_LOCAL_RESPONSE_OUT callbacks anymore,
they accept a flag indicating the requested callback types instead.
- The function parameters are grouped together in uac_req_t structure, use set_uac_req() macro to fill it (see uac.h)
- It is possible now to free the memory allocated for the callback parameter if necessary, register the TMCB_DESTROY
callback, and free the memory when:
- TMCB_DESTROY callback is called
- or tm module fails to create the UAC, the function returns error.
Andrei Pelinescu-Onciul [Mon, 18 Jun 2007 21:26:34 +0000 (21:26 +0000)]
- when failing to send on all branches (due to send error, onsend_router or
blacklist), always send back a send error reply (ignore possible dns
errors which might have happened during dns failover)
Andrei Pelinescu-Onciul [Mon, 18 Jun 2007 21:23:54 +0000 (21:23 +0000)]
- updated to the modified dns api
Andrei Pelinescu-Onciul [Mon, 18 Jun 2007 21:20:58 +0000 (21:20 +0000)]
- dns naptr support (off by default)
- dns naptr related config options: dns_try_naptr (off by default),
dns_udp_pref, dns_tcp_pref. dns_tls_pref (protocol preferences for naptr
record selection)
- dns srv load balancing config options: dns_srv_lb (off by default)
- dns resolver & cache api change (to support getting the protocol via naptr)
- fix: dns iteration through A & AAAA records was not correct
For more info see doc/dns.txt.
Maxim Sobolev [Mon, 18 Jun 2007 21:00:13 +0000 (21:00 +0000)]
Regent to accomodate latest schema XML changes and postgresql XLT fix.
Maxim Sobolev [Mon, 18 Jun 2007 20:58:51 +0000 (20:58 +0000)]
Don't forget to insert rows into the version table, like other DB
backends do.
Andrei Pelinescu-Onciul [Thu, 14 Jun 2007 23:29:31 +0000 (23:29 +0000)]
- fix: incorrect iteration through A & AAAA records
- support for SRV weight based load balancing as described in rfc2782
(as opposed to simple failover)
Andrei Pelinescu-Onciul [Thu, 14 Jun 2007 23:23:47 +0000 (23:23 +0000)]
- added ISAAC, Bob Jenkins's fast pseudo-random generator and some
wrapper arround it (see http://www.burtleburtle.net/bob/rand/isaacafa.html)
Andrei Pelinescu-Onciul [Thu, 14 Jun 2007 23:17:57 +0000 (23:17 +0000)]
- uses the new run_actions interface
Andrei Pelinescu-Onciul [Thu, 14 Jun 2007 23:14:29 +0000 (23:14 +0000)]
- updated all the users of run_actions() or do_action() to the new
interface
Andrei Pelinescu-Onciul [Thu, 14 Jun 2007 23:12:26 +0000 (23:12 +0000)]
- fix: run_actions() & friends used some static variables to save some
internal state (run_flags, last_ret a.s.o). Problems would appear when
different routes (e.g. main route and an onsend route) would execute in
the same process (e.g. an onsend route return would exit also the
main route). Now run_actions(), do_action() and eval_expr() take an
extra "context" parameter (which must be initialized before use).
Andrei Pelinescu-Onciul [Thu, 14 Jun 2007 16:59:21 +0000 (16:59 +0000)]
-faster page pre-faulting (when shm_force_alloc = yes): touch only one word
per page
Vaclav Kubart [Thu, 14 Jun 2007 15:08:02 +0000 (15:08 +0000)]
added incompatibility description into presence handbook and XCAP module documentation as requested by SER-283
Andrei Pelinescu-Onciul [Tue, 12 Jun 2007 22:43:19 +0000 (22:43 +0000)]
- try first to spin a little and only if this fails go through the futex
syscalls (uses ADAPTIVE_WAIT and ADAPTIVE_WAIT_LOOPS like FAST_LOCK). This
gives a 5% performance increase compared to the "pure" futex version on my
machine (for tm).
Andrei Pelinescu-Onciul [Tue, 12 Jun 2007 18:22:45 +0000 (18:22 +0000)]
- attempt to properly compute the used w/o overhead info
Andrei Pelinescu-Onciul [Tue, 12 Jun 2007 18:21:42 +0000 (18:21 +0000)]
- more mem info: used and real_used (real_used includes the malloc
overheads)
- extensive debuggin command for sf_malloc and ll_malloc: core.sfmalloc
(enabled only if ser compiled with -DLL_MALLOC or -DSF_MALLOC).
Andrei Pelinescu-Onciul [Tue, 12 Jun 2007 18:07:12 +0000 (18:07 +0000)]
- added sf_malloc: a multi-process safe (internal locking), multi-pool
f_malloc version. Should perform better on lots of CPU once properly
tunned (experimental for now)
- added ll_malloc: same as above but most operation are lockless
(except for "big" fragment handling). For now needs tunning and
it still keeps too many debugging statistics. (experimental for now)
Greger V. Teigre [Tue, 12 Jun 2007 12:57:56 +0000 (12:57 +0000)]
Refs SER-274, ported from rel_2_0_0
Deleted README.cfg and updated README and added a warning about obsoleted
file in README-MODULES.
Greger V. Teigre [Tue, 12 Jun 2007 12:52:17 +0000 (12:52 +0000)]
Ref SER-271, ported from rel_2_0_0
Used cvs2cl.pl (http://www.red-bean.com/cvs2cl/) to generate GNU-style changelog from
cvs log. Replaced very old ChangeLog with the newly generated one.
The ChangeLog only goes back to the rel_0_9_0 release branch and thus contains changes
for release 2.0 release candidate 1.
Andrei Pelinescu-Onciul [Sat, 9 Jun 2007 17:48:52 +0000 (17:48 +0000)]
- execute wait timer handlers & retr. handlers in the fast timer
(=> faster response), but switch to the slow timer for the FR (which
might take some time: dns failover, failure routes a.s.o). This change
brought me 300 extra cps (~3%).
- added missing include
Andrei Pelinescu-Onciul [Sat, 9 Jun 2007 17:44:18 +0000 (17:44 +0000)]
- use lr from the parsed uri (no need to re-parse the uri params)
Andrei Pelinescu-Onciul [Thu, 7 Jun 2007 21:52:30 +0000 (21:52 +0000)]
- ver. no ++
- -DHAVE_SCHED_SETSCHEDULER added on linux
- excluded from normal compilation the modules that throw compile errors:
avpops bdb db_ops dbtext flatstore lcr msilo pdt permissions speeddial
Andrei Pelinescu-Onciul [Thu, 7 Jun 2007 21:49:53 +0000 (21:49 +0000)]
- support for pre-faulting the shm mem. pages (memset(0))
Andrei Pelinescu-Onciul [Thu, 7 Jun 2007 21:45:56 +0000 (21:45 +0000)]
- tm stats changes: cleaner & faster
- updated to the new PROC_INIT stuff (fixes a tm bug)
Andrei Pelinescu-Onciul [Thu, 7 Jun 2007 21:44:00 +0000 (21:44 +0000)]
- new script config vars: mlock_pages, shm_force_alloc, real_time,
rt_prio, rt_policy, rt_timer1_prio, rt_timer1_policy, rt_timer2_prio,
rt_timer2_policy (for more info see NEWS)
Andrei Pelinescu-Onciul [Thu, 7 Jun 2007 21:40:14 +0000 (21:40 +0000)]
- added the new PROC_INIT rank and the init_child(PROC_INIT) calls
(see doc/modules_init.txt for more info)
- introduced new options for locking pages in memory, pre-faulting
the shared memory pages and real time mode
Andrei Pelinescu-Onciul [Thu, 7 Jun 2007 21:34:03 +0000 (21:34 +0000)]
- short module initialization functions description: what you should and
shouldn't do in mod_init, child_init(PROC_MAIN), child_init(PROC_INIT)
and child_int(other_values)
Andrei Pelinescu-Onciul [Thu, 7 Jun 2007 21:04:32 +0000 (21:04 +0000)]
- attempt to fix the replied_locally statistics
Andrei Pelinescu-Onciul [Thu, 7 Jun 2007 21:02:40 +0000 (21:02 +0000)]
- t_unref bug check relaxed (false positives)
Andrei Pelinescu-Onciul [Wed, 6 Jun 2007 21:57:13 +0000 (21:57 +0000)]
- partial update (new tm stuff a.s.o.)
Andrei Pelinescu-Onciul [Wed, 6 Jun 2007 21:56:53 +0000 (21:56 +0000)]
- ver. no increased
Andrei Pelinescu-Onciul [Wed, 6 Jun 2007 21:56:27 +0000 (21:56 +0000)]
- attempt to unroll one of the hash loops
Andrei Pelinescu-Onciul [Wed, 6 Jun 2007 21:54:04 +0000 (21:54 +0000)]
- changed tm hash bucket lists (simplified & a little bit faster)
- inlined some used-a-lot functions
- don't allocate space for md5 if syn_branches is on (default) =>
32 bytes saved/transaction
Andrei Pelinescu-Onciul [Wed, 6 Jun 2007 17:36:52 +0000 (17:36 +0000)]
- re-arranged tmcb structure members so that even if a module compiled
without TMCB_ONSEND uses a callback from a TMCB_ONSEND compiled tm, it
will still work
Andrei Pelinescu-Onciul [Wed, 6 Jun 2007 14:54:04 +0000 (14:54 +0000)]
- paste error fixed ig TMCB_ONSEND is defined (default off)
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 22:13:35 +0000 (22:13 +0000)]
- reply code when no new branch was added fix
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 22:12:41 +0000 (22:12 +0000)]
- removed EXTRA_DEBUG abort() from bug check in t_check and replaced
the error message w/ a warning: as Gergely and Miklos pointed out this is a
perfectly valid case and not a bug
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 21:29:26 +0000 (21:29 +0000)]
- blackisted message switched to L_DBG (less noisy)
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 21:27:48 +0000 (21:27 +0000)]
- extra BUG checks (just to be sure we catch early kr related bugs)
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 21:26:44 +0000 (21:26 +0000)]
- warnings & DBG cleanup
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 21:25:35 +0000 (21:25 +0000)]
- new stats support
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 21:24:12 +0000 (21:24 +0000)]
- fix: kr was not properly reset (set_kr(0) doesn't have the desired effect)
- new statistics support
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 21:21:06 +0000 (21:21 +0000)]
- even more tm statistics (usefull for debugging)
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 16:52:53 +0000 (16:52 +0000)]
- compiled only if -DDL_MALLOC is defined (code surrounded by #ifdef)
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 16:35:27 +0000 (16:35 +0000)]
- warning fixes (force casts)
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 15:16:44 +0000 (15:16 +0000)]
- e2ecancel is sent hop by hop (closes SER-68)
- delayed err reply: delay error replies from t_relay() until end of script: if
script hasn't replied or dropped the trans., send reply(tm_errno).
- automatic 100 reply generation for invites can be turned on/off via
a tm global parameter or on a per transaction basis (auto_inv_100,
t_set_auto_inv_100()) (closes SER-199)
- different error replies for too many branches, resolve or send error
- if error adding branches, reply with the lowest error
- membar_write before increasing nr_of_outgoings (branches) and membar_depends
in which_cancel (should allow lockless which_cancel, with the possible race
of not canceling a branch that is being created in the same time, OTOH
this happens even if locks are used)
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 14:35:17 +0000 (14:35 +0000)]
- fix: relay_reply/local_reply on error cancel/cancel_bitmap: avoid attempts
to cancel twice
- fix: wrong silent discarding condition
- noisy_ctimer is now 1 (on) by default (due to the above fix you'll really
see silent discards without it which is not probably what you want/expect)
- cancel updates:
- cancel not sent on timeout, for branches on which no reply was received
(closes SER-254).
- cancels for silently canceled branches (no replies ever received) are
automatically generated if later a provisional reply arrives on such a
branch (close SER-232)
- if branch canceled (non-silently) and provis. reply on it => immediately
retransmit cancel
- should_cancel_branch() can now be called w/o any lock held
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 14:12:36 +0000 (14:12 +0000)]
- tm per transaction retr timers: support for changing the retransmission
intervals on the fly, from the script, on a per transaction basis
(it is enabled if tm is compiled with -DTM_DIFF_RT_TIMEOUT -- default):
t_set_retr(t1, t2)
- tm transaction lifetime for inv/noninv (config. per transaction): added
maximum transaction lifetime - a transaction is not allowed to be active
longer then this interval. See t_set_max_lifetime() or the new
script params. max_inv_lifetime and max_noninv_lifetime.
- fix: retr_interval was not disabled on error in retr_buf_hanlder
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 13:53:41 +0000 (13:53 +0000)]
- added MAX_UVAR_VALUE(), MAX_int(a,b) MIN_int(a,b)
Andrei Pelinescu-Onciul [Tue, 5 Jun 2007 13:52:13 +0000 (13:52 +0000)]
- fix: blacklist timer would could enter and infinite loop, after ~1 h, due
to an overflowing unsigned short var
Hendrik Scholz [Fri, 1 Jun 2007 09:20:34 +0000 (09:20 +0000)]
Add a new memory allocator (-DDL_MALLOC) which is based on Doug Lea's
dl_malloc implementation (http://g.oswego.edu/dl/html/malloc.html).
Patch has been provided by Jan Andres <jan.andres@freenet-ag.de>
Notes:
- very fast, esp. on shm_malloc() in comparison to other algorithms
- has been in production use at freenet(.de) for 6+ months
- is picky about buffer overruns (aka detects and crashes on them)
- disabled by default (set -DDL_MALLOC in Makefile.defs)
- closes SER-278
Andrei Pelinescu-Onciul [Thu, 31 May 2007 16:17:20 +0000 (16:17 +0000)]
- possible infinite loop fix
Andrei Pelinescu-Onciul [Wed, 30 May 2007 22:22:42 +0000 (22:22 +0000)]
- added callbacks/hooks for dst_blacklist_add: one can register such
a callback and deny adding a destination to the blacklist list, or
it can change the destinations flags.
- autodetect if locks or lock_set are not limited and if so use
one lock per hash bucket or a lock_set of hash table size (should work
better on multiple cpus, less cacheline invalidations)
Andrei Pelinescu-Onciul [Wed, 30 May 2007 21:50:39 +0000 (21:50 +0000)]
- minor typos
Andrei Pelinescu-Onciul [Wed, 30 May 2007 18:35:54 +0000 (18:35 +0000)]
- TMCB_E2ECANCEL_IN reverted again to pass the invite transaction as param
- comments update (registering callbacks should now be safe from everywhere)
Hendrik Scholz [Wed, 30 May 2007 12:48:24 +0000 (12:48 +0000)]
Add two new config variables to set source IP on outbound TCP connections.
As Jan stated this is useful for HA setups with virtual IPs.
Patch provided by Jan Andres <jan.andres@freenet-ag.de>
Closes SER-277
Miklos Tirpak [Wed, 30 May 2007 12:28:17 +0000 (12:28 +0000)]
t_relay_cancel() is added to the sample configuration
Miklos Tirpak [Wed, 30 May 2007 11:19:01 +0000 (11:19 +0000)]
documentation is updated with t_relay_cancel(), reparse_invite, and ac_extra_hdrs
Andrei Pelinescu-Onciul [Tue, 29 May 2007 23:59:34 +0000 (23:59 +0000)]
- transactions are now deleted when their reference counter reaches 0
(this removes the need for the delete timer hack and some extra locking and
it should reduce memory usage in very busy situations). For now it's still
possible to revert to the old behaviour by removing the TM_DEL_UNREF define
in h_table.h.
Andrei Pelinescu-Onciul [Tue, 29 May 2007 21:26:20 +0000 (21:26 +0000)]
- fixed usage of unalloc. mem.
Andrei Pelinescu-Onciul [Tue, 29 May 2007 21:25:53 +0000 (21:25 +0000)]
- fixed usage of unalloc. mem.
Miklos Tirpak [Tue, 29 May 2007 15:52:37 +0000 (15:52 +0000)]
CANCELs and negative ACKs are constructed from the INVITE which was sent out
instead of building them from the received one. The disadvantage is
that the outgoing INVITE has to be partially reparsed, the advantage is that
the CANCEL/ACK is always RFC-compliant, it always contains the same route-set
as the INVITE message. (closes SER-212)
build_local_reparse() function can be used to construct a CANCEL or ACK, it
reparses the outgoing INVITE and applies the modifications specified by the RFC.
new module parameters:
- reparse_invite: set it to 0 in order to revert to the old behaviour)
- ac_extra_hdrs: prefix for additional headers which are kept by build_local_reparse()
new script function:
- t_relay_cancel() - can be used to catch CANCEL requests at the beginning of the script
and bypass the rest of the script processing:
if (!t_relay_cancel()) { # implicit drop if relaying was successful,
# nothing to do
# corresponding INVITE transaction found but error occurred
sl_reply("500", "Internal Server Error");
drop;
}
# bad luck, corresponding INVITE transaction is missing,
# do the same as for INVITEs
...
Andrei Pelinescu-Onciul [Tue, 29 May 2007 13:44:19 +0000 (13:44 +0000)]
- added membar_depends() for run_callbacks and unmatched_totag
- callbacks ids are no longer set: nobody uses them and when doing
a lockeless insert they cause extra expensive writes
Andrei Pelinescu-Onciul [Tue, 29 May 2007 11:31:29 +0000 (11:31 +0000)]
- added membar_depends(), needed on smp archs. with separate cache banks
where it's possible to get a new pointer value, but the old pointer content
(e.g. if the two are in different cache banks and the "content" bank is very
busy processing a long invalidations queue). For now only Alpha SMP
needs it, on all other archs is a no-op (for more info see atomic_ops.h
, http://lse.sourceforge.net/locking/wmbdd.html,
http://www.linuxjournal.com/article/8212 or Alpha Architecture Reference
Manual Chapter 5.6.
- added membar_atomic_op(), membar_atomic_setget(), membar_read_atomic_op(),
membar_read_atomic_setget(), membar_write_atomic_op(),
membar_write_atomic_setget() -- special case memory barriers that can be
optimized if the atomic ops already force some kind of barrier (e.g. x86),
see the description in atomic_ops.h for more info.
Andrei Pelinescu-Onciul [Tue, 29 May 2007 09:36:51 +0000 (09:36 +0000)]
- more callback fixes (TMCB_E2E_CANCEL_IN never called after the latest
changes)
- some minor optimizations attempts (predict all the callback branches/ifs as
not taken since in the normal case we don't have any callbacks)
Vaclav Kubart [Mon, 28 May 2007 11:33:01 +0000 (11:33 +0000)]
cosmetical changes in presence handbook
Andrei Pelinescu-Onciul [Sat, 26 May 2007 11:12:29 +0000 (11:12 +0000)]
- darwin sigwait() bug workaround: on darwin sigwait is buggy: it will cause
extreme slow down on signal delivery for the signals it doesn't wait on (I've
measured a 36s delay on a 1.5Ghz powerbook). The workaround is: sigwait() on
all the signals we are interested in ser and manually call the master signal
handler (sig_usr(sig_no)) if the signal!= slow timer signal.
Andrei Pelinescu-Onciul [Sat, 26 May 2007 10:55:39 +0000 (10:55 +0000)]
- changed darwin module link flags (instead of -bundle_loader ser -flat_namespace -undefined suppress) -- in this way the build module dependecy on ser is
removed
Andrei Pelinescu-Onciul [Fri, 25 May 2007 16:15:11 +0000 (16:15 +0000)]
- fixed TMCB_E2ECANCEL_IN wrong value
- fixed wrong buffer length in some TMCB_RESPONSE_SENT callback calls
- switched TMCB_E2ECANCEL to path the cancel transaction as param. and not
the invite one
(bugs reported by Gergely Kovacs <gergo@iptel.org>)
Andrei Pelinescu-Onciul [Fri, 25 May 2007 15:50:11 +0000 (15:50 +0000)]
- more darwin build with libs fixes (works now)
- sercmd is no longer force-rebuilt every time
Andrei Pelinescu-Onciul [Thu, 24 May 2007 18:01:19 +0000 (18:01 +0000)]
- tmcb onsend callbacks update (they get also the request or the reply if they
are known when the callback is called)
- added unlikely() when checking if an onsend callback has to be called
(so that the branch will be predicted as untaken)
Andrei Pelinescu-Onciul [Thu, 24 May 2007 09:59:39 +0000 (09:59 +0000)]
- added TMCB_E2E_CANCEL_IN callback, called each a CANCEL mathcing an exisitng invite transaction arrives
- fixed has_tran_tmcb typo(?)
Vaclav Kubart [Thu, 24 May 2007 07:49:15 +0000 (07:49 +0000)]
needed to call Makefile.nonser in subdirectories
Vaclav Kubart [Thu, 24 May 2007 07:47:25 +0000 (07:47 +0000)]
creating links with relative path instead of absolute - helps when installed into other directory than used for run
Andrei Pelinescu-Onciul [Mon, 21 May 2007 13:55:51 +0000 (13:55 +0000)]
- generic build libs fixes
- untested darwin support for automatically setting the library path /
rebuilding the lib on changed install path (on darwin
the library install path is stored in the library itself and from there it
is copied in the linked object => if the library path is changed we have to
re-link the library instead of re-linking the object that links it)
Andrei Pelinescu-Onciul [Fri, 18 May 2007 10:15:21 +0000 (10:15 +0000)]
- darwin: if no locking method found revert to sys v sems (neither pthread
mutexes or unnamed posix sems work in darwin for inter process synchronization)
Andrei Pelinescu-Onciul [Fri, 18 May 2007 10:08:47 +0000 (10:08 +0000)]
- reverted to addic instead of addi - sometimes gcc uses r0 as the second
operand in addi and addi rD,r0, val is a special case, equivalent with
rD=0+val and not rD=r0+val. addic does not have this problem (but sets
the carry flag in XER).
- some macro fixes
Andrei Pelinescu-Onciul [Thu, 17 May 2007 11:04:56 +0000 (11:04 +0000)]
- experimental lockless callback insert (callback registration should now be
safe from everywhere)
Vaclav Kubart [Thu, 17 May 2007 09:58:05 +0000 (09:58 +0000)]
Added functions for storing/loading flagged attributes according to user-defined key.
Documentation will be updated soon (documentation of this module is out of date and
needs to be completely rewritten).
solves SER-252
Andrei Pelinescu-Onciul [Wed, 16 May 2007 18:56:42 +0000 (18:56 +0000)]
- fixed compile problems when DIALOG_CALLBACKS is defined
Andrei Pelinescu-Onciul [Wed, 16 May 2007 18:32:17 +0000 (18:32 +0000)]
- callbacks can register to be also called when the transaction or dialog is
destroyed (so that they can free their parameters if needed)
Andrei Pelinescu-Onciul [Mon, 14 May 2007 21:29:46 +0000 (21:29 +0000)]
- added futex based locking, used by default on linux 2.5.70+
(to turn it off compile with make use_futex=no)
- moved get_sys_ver from io_wait.c into ut.c
Andrei Pelinescu-Onciul [Mon, 14 May 2007 17:29:31 +0000 (17:29 +0000)]
- parts of atomic_ops.h moved into atomic/atomic_common.h and
atomic/atomic_native.h
- added membar_eneter_lock() and membar_leave_lock() (to be used only if
creating locks using the atomic ops functions, for more info see atomic_ops.h)
Andrei Pelinescu-Onciul [Mon, 14 May 2007 17:09:31 +0000 (17:09 +0000)]
*** empty log message ***
Maxim Sobolev [Mon, 14 May 2007 09:20:04 +0000 (09:20 +0000)]
Make buildable with the latest version of the curl (CURLOPT_MUTE has
been depreciated).
Andrei Pelinescu-Onciul [Fri, 11 May 2007 20:44:15 +0000 (20:44 +0000)]
- atomic_add & atomic_cmpxchg added to ppc
- atomic_unkown (used when the procesor does not suport atomic ops or
is not among the supported ones), tries now to use a "hash" of locks
if GEN_LOCK_SET_T_UNLIMITED is defined => less contention on multi-cpus
- atomic_ops.h defines *_UNLIMITED macros when the number of locks or set
size are limited only by the available memory (everything exept SYSV sems)
- license changes: all the atomic* stuff and the locks are now under a
BSD (OpenBSD) style license
Vaclav Kubart [Fri, 11 May 2007 13:09:32 +0000 (13:09 +0000)]
removed 'module dependencies on libs' from main Makefile because Andrei already
updated these modules and libs to use new making style
Vaclav Kubart [Fri, 11 May 2007 11:27:06 +0000 (11:27 +0000)]
examples updated according to anounced auth_db change
(see http://lists.iptel.org/pipermail/serdev/2007-May/010034.html)
Andrei Pelinescu-Onciul [Thu, 10 May 2007 18:27:07 +0000 (18:27 +0000)]
- alpha, armv6 and mip isa2+ atomic_add and atomic_cmpxchg (armv6 & alpha not
tested at all due to lacking hardware or emulators)
Vaclav Kubart [Thu, 10 May 2007 16:46:08 +0000 (16:46 +0000)]
next piece of documentation
Vaclav Kubart [Thu, 10 May 2007 11:29:26 +0000 (11:29 +0000)]
Yet another doxygen configuration, but it might be useful to generate
documentation for libraries only.
Libraries were partialy documented in docbook but it seems to be very hard to
manage development documentation outside of code and thus it is changing to
documentation generated from code in this case.
Vaclav Kubart [Thu, 10 May 2007 11:25:07 +0000 (11:25 +0000)]
updated to change in reference counters and CDS initialization
Vaclav Kubart [Thu, 10 May 2007 11:23:49 +0000 (11:23 +0000)]
started doxygen documentation of CDS library and little changes which seemed
useful when creating documentation (reference counters, removed library init
and cleanup function because it is not needed more)
Vaclav Kubart [Thu, 10 May 2007 11:21:22 +0000 (11:21 +0000)]
fixed documentation comment to avoid doxygen warnings
Andrei Pelinescu-Onciul [Tue, 8 May 2007 19:41:59 +0000 (19:41 +0000)]
- x86: atomic_add
- sparc64: atomic_cmpxchg, atomic_add
Andrei Pelinescu-Onciul [Tue, 8 May 2007 09:03:11 +0000 (09:03 +0000)]
- fix: s/-/_/ in exported Makefile vars (some shells don't support '-' in
variables)
Andrei Pelinescu-Onciul [Tue, 8 May 2007 08:49:22 +0000 (08:49 +0000)]
- removed nested function (made it static inline)
Andrei Pelinescu-Onciul [Tue, 8 May 2007 08:35:24 +0000 (08:35 +0000)]
- updated the ser lib using modules to the new lib building system
(libs are built automatically if needed by a module and when a module that uses a ser lib is installed it will be re-linked with the correct rpath => no more
LD_LIBRARY_PATH)
Andrei Pelinescu-Onciul [Tue, 8 May 2007 08:30:40 +0000 (08:30 +0000)]
- switched all the ser libraries to the new ser lib build system
Andrei Pelinescu-Onciul [Tue, 8 May 2007 08:25:37 +0000 (08:25 +0000)]
- added cmpxchg