commit 383d2e6325ab9f00259f54a84c50f0956c7bee3d
Author: Niels Roest <niels@directfb.org>
Date:   Tue Mar 31 12:22:39 2009 +0200

    [shm] 32-bit changes; changed address from ulong to pointer

commit 83b2354860febdcb470267045faebbf1c4b2cdb5
Author: Niels Roest <niels@directfb.org>
Date:   Thu Mar 12 20:53:08 2009 +0100

    [release] linux-fusion 8.1.0
    
    hybrid fusion: support for API 3.x, 4.x and 8.x simultaneously.
    Primarily, FusionEnter now accepts more inputs.
    FUSION_API_MAJOR/MINOR is now fixed on API 3.3,
    so user space apps that require 4.x or 8.x should specify the
    required version instead of using the FUSION_API_... defines.
    /proc filesystem: improve safety by using seq_file instead of sprintf.
    formatting: now uses Linux kernel formatting rules.
    documentation: add API/ioctl interface docs.

commit 6bbcaec59842c010034e6931a4c7f0e584079168
Author: Niels Roest <niels@directfb.org>
Date:   Thu Mar 12 20:42:28 2009 +0100

    [doc] Add API doc fusion.pdf. Rename fusion.texi to libfusion.texi.

commit 1261554d83945fc69c8e2701d3da35563f082e63
Author: Niels Roest <niels@directfb.org>
Date:   Thu Mar 12 20:00:44 2009 +0100

    [api] upgrade provided API to 8.1.0

commit 85770dfb2368d4450e05348d4e6a3e011c47c75f
Author: Niels Roest <niels@directfb.org>
Date:   Mon Mar 9 14:59:33 2009 +0100

    [API] call_return: check serial only for API >= 4.0
    (and: remove /proc/fusion/0/call duplicates)

commit d4e69697d57beea233267db7f9976e34039c1acc
Author: Niels Roest <niels@directfb.org>
Date:   Tue Mar 3 19:58:19 2009 +0100

    [32/64bit] make macros less arch-dependent

commit b12d9159b94ec45b28857e39b011822cab2eb9f8
Author: Niels Roest <niels@directfb.org>
Date:   Tue Mar 3 19:03:37 2009 +0100

    [formatting] Lindent to match Linux kernel style

commit e1b9ae9723753c621c54ce94b2e06bab2d3d1fe5
Author: Niels Roest <niels@directfb.org>
Date:   Mon Mar 2 21:21:10 2009 +0100

    rewrite towards mainlining
    
    all: use seq_file for safe /proc output
    call: use FusionEntry to align with other fusion objects
    call: fix for bug in DirectFB 1.0.x when using one-way calls

commit 45c17fe5daa80900b0b3eafe8a1e6ab561dfe2a2
Author: Niels Roest <niels@directfb.org>
Date:   Fri Feb 27 16:37:35 2009 +0100

    [API] preliminary support for API 3.x, 4.x and 8.x
    
    3.x for DirectFB 1.0
    4.x for DirectFB 1.1
    8.x for DirectFB 1.2 and later
    
    In Fusion Enter the master passes the requested APIs,
    the slaves have to match (allowed is smaller minor).
    
    For 1.0 compatibility, the macros FUSION_API_MAJOR
    and _MINOR default to API 3.3.

commit 5f69dd97bc1c73935f8bd5da6ce1800c7a8e6be6
Author: Niels Roest <niels@directfb.org>
Date:   Fri Feb 6 15:56:32 2009 +0100

    [release] linux-fusion 8.0.3
    
    fixes for 2.6.27 kernel
    call: fixed possible deadlock at shutdown
    call: extra checks for call with/without return:
      Return EOPNOTSUP if FUSION_CALL_RETURN is attempted with a zero serial
      Return ENOMSG if serial number has not been found
    fusionee: Flush previous messages before waiting for new ones which only applies to the unused blocking mode

commit 9ef27337f4ab671b6d863beb5f3ea493eb5332fa
Author: Niels Roest <niels@directfb.org>
Date:   Mon Jan 19 16:30:58 2009 +0100

    [call] small change on advise of dok, fixes other cases too.

commit 6ca69c003cff627c04c04012baaaa0b92628fc86
Author: Niels Roest <niels@directfb.org>
Date:   Tue Jan 13 18:57:02 2009 +0100

    [call] fixes forever-wait in ioctl shutdown + small fix for interrupted shutdown

commit 28550540dfc76601288b8ec06403a036e0e6f46e
Merge: 4319618... db9bfae...
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Sat Oct 18 03:14:45 2008 +0200

    Merge branch 'master' of git+ssh://git.directfb.org/git/directfb/core/linux-fusion

commit 43196188dcbab64e3c3adefd1feb60d48821b9c7
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Sat Oct 18 03:11:48 2008 +0200

    [call] Use zero serial number to indicate call without return. Fail upon invalid return again.
    
    Return EOPNOTSUP if FUSION_CALL_RETURN is attempted with a zero serial.
    
    Return ENOMSG if serial number has not been found.

commit 33a29f995614bd68db7802c57580f876d8609ae6
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Sat Oct 18 02:48:06 2008 +0200

    [fusionee] Flush previous messages before waiting for new ones which only applies to the unused blocking mode.

commit ca246664364600b422d3e5550f64240604be06d8
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Sat Oct 18 02:47:01 2008 +0200

    [fifo] Remove obsolete proto type.

commit db9bfaea729d0013d0631da03badd98249bced71
Author: Niels Roest <niels@directfb.org>
Date:   Fri Oct 10 15:45:29 2008 +0200

    [fusion] changes for 2.6.27-rc8 kernel

commit 675c45e0d957518c5ae2321893cccf50c18f170e
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Mon Sep 29 12:20:22 2008 +0200

    Updated ChangeLog

commit 174daaf1df8a693101be64a941188c5f23e14f54
Merge: 566a807... d61ea53...
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Mon Sep 29 12:20:06 2008 +0200

    Merge branch 'master' of git+ssh://git.directfb.org/git/directfb/core/linux-fusion

commit 566a8070584aefe7aa058b8d8a5cadaab1b78a32
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Mon Sep 29 12:19:47 2008 +0200

    Updated ChangeLog

commit d61ea53415720bd07838aaba40dffca5c5878e81
Author: Niels Roest <niels@directfb.org>
Date:   Thu Sep 18 10:52:53 2008 +0200

    [fusion] new feature: automatically unlock transferred skirmishes
    If you expect that a callee will lock a skirmish, call you and thereby
    transferring that skirmish to you, you may already lock it; it will
    be unlocked automatically when the call is due.

commit 65d1b30f089509d98dd00fcc5519644da3a513c1
Author: Niels Roest <niels@directfb.org>
Date:   Mon Sep 15 15:22:46 2008 +0200

    [fusion] made include order consistent. Was needed for new 2.6.26.5 kernel.

commit fa3c697fc3d31958c27bf4dbaaa2a31b00d8b22f
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Sun Sep 14 02:52:21 2008 +0200

    [calls] Fixed dead lock in fusion_call_destroy() leading to unkillable process.
    
    There was a typo that caused the call to be unlocked instead of the list.
    
    Furthermore, there were two places without checking for a pending signal after
    calling fusion_sleep_on().

commit 0353c456668028cd658b5a95f8a44e1d5e89ff4e
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Wed Sep 10 18:25:15 2008 +0200

    [README] Some initial cross build notes...

commit 597949f214df557d3c5b77c7ba2ec7ffe90cffd7
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Wed Sep 10 18:21:59 2008 +0200

    [shm] Align pool base on a 64K boundary.

commit 0323ca485081e86000bba37e5ffc17b6a3d9c3de
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Wed Sep 10 18:16:30 2008 +0200

    [Makefile] Allow SYSROOT variable to resolve absolute symbolic links relative to it.

commit ecb399e30402798059f4378a66d22221512261cd
Author: Denis Oliver Kropp <dok@debian.(none)>
Date:   Wed Sep 3 00:45:43 2008 +0900

    [types] Changed maximum version for bool definition to 2.6.18 to make it build on such kernels.

commit f8a74066663849a94ee967efc1130d49b76ce730
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Tue Aug 12 14:50:54 2008 +0200

    updated

commit b26593de0f879ee57ee5bf5c5ad64fd543306c34
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Tue Aug 12 14:49:54 2008 +0200

    [skirmish] Replaced bad debugging printk by new FUSION_SKIRMISH_LOG macro (no-op).

commit 2889e1738d5e8b39987a40de39eac994d2c15389
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Sat Jul 19 18:15:25 2008 +0200

    [calls] Fixed FUSION_CALL_RETURN errors with destroyed calls during shutdown.
    
    When a FusionCall is destroyed via FUSION_CALL_DESTROY, it waits for all pending
    execution requests to be processed (synchronous or asynchronous) and waits for
    the results to be sent back for synchronous calls in case they get defered...
    
    Added fusionee_wait_processing() to wait for processing of all pending messages
    with a certain type and ID.

commit 86c1cadd7d62e4d4c759c94941a41172044187b1
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Sat Jul 19 18:13:20 2008 +0200

    [tests] Change calls test to use asynchronous (FCEF_ONEWAY) calls, more output.

commit 23605c744abcccbd911ab7cf3185d2bf86c1e355
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Sat Jul 19 18:12:23 2008 +0200

    [entries] Include <linux/mutex.h> instead of <linux/semaphore.h>, probably not working with all kernels...

commit a7aa6226c4f2c67d093a92e09118f86efb6d7254
Author: Denis Oliver Kropp <dok@cyto.(none)>
Date:   Mon Jun 2 16:40:35 2008 +0200

    updated ChangeLog

commit 392ec5afbc51af64505321922d3283ecfa82d328
Author: Denis Oliver Kropp <dok@shark.directfb.home>
Date:   Sat May 31 17:01:51 2008 +0200

    Fixes for 2.6.26.

commit c53d14b217900928f13269ec1e11b009cc8a5913
Author: Denis Oliver Kropp <dok@shizo.directfb.home>
Date:   Wed Mar 26 11:37:10 2008 +0100

    REACTOR: Fixed call to memset() in reactor_attach(). Thanks to Peter Deckmyn!

commit 40f4adbcaf0bf9cb016d478a89e156a06cb04466
Author: Denis Oliver Kropp <dok@shizo.directfb.home>
Date:   Tue Feb 19 15:29:05 2008 +0100

    [skirmish] Support resume of interrupted wait and regain lock after timeout.
    
    This is compatible with mutex/condition, except for expected errors.
    
    Requires user space changes.

commit d57e856bad14cf2fc84eff0a62375ceba5ae5c13
Author: Denis Oliver Kropp <dok@shizo.directfb.home>
Date:   Mon Feb 18 22:36:44 2008 +0100

    [Makefile] Don't override KERNEL_VERSION and other variables if they're set in the environment.

commit 196b44e17fa2bb73c284284758c23675cb6c5c7d
Author: Ville Syrjala <syrjala@sci.fi>
Date:   Fri Feb 22 00:00:26 2008 +0200

    Add a Kconfig file.

commit 5c7f24e59db5709ef95c5db0c20054ee12352253
Author: Denis Oliver Kropp <dok@shizo.directfb.home>
Date:   Sat Jan 26 07:47:09 2008 +0100

    ignore

commit 0f3a13ced72bd467ae68de06e8d74e6926a2c7d0
Merge: 0ed6741... a3de178...
Author: Denis Oliver Kropp <dok@shizo.directfb.home>
Date:   Sat Jan 26 05:15:51 2008 +0100

    Merge branch 'master' of git+ssh://git.directfb.org/git/directfb/core/linux-fusion

commit 0ed6741a2a8e1a366c813d6243248373f66c2c61
Author: Denis Oliver Kropp <dok@shizo.directfb.home>
Date:   Sat Jan 26 03:04:38 2008 +0100

    REACTOR: Added missing memset() and kfree() in reactor_attach(). Thanks to Amit Mahajan and Niels Roest!
    
    Without memset() the new counters for the channels are not initialized.
    
    The kfree() was missing in the code that reallocated the array for more channels.

commit ceabe82a1024e4cea518b10d1301a3d0fe41e96a
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Thu Oct 11 11:59:22 2007 +0200

    Don't build with yield() by default, but require FUSION_SKIRMISH_YIELD to be defined.

commit a3de17819d39a23199d7b1c6f606d70c0d7580db
Author: Ville Syrjala <syrjala@sci.fi>
Date:   Sat Dec 1 21:18:59 2007 +0200

    Use current_kernel_time() instead of xtime
    
    xtime is no longer exported in 2.6.24 so use current_kernel_time() instead.

commit a31196f97f2bc93db11b3ce435d7cfb105490ec9
Author: Ville Syrjala <syrjala@sci.fi>
Date:   Sat Dec 1 21:15:18 2007 +0200

    Fix Makefile for 2.6.24
    
    2.6.24 top-level Makefile no longer accepts CPPFLAGS. KCPPFLAGS must be
    used instead and it is combined with the normal kernel CPPFLAGS.

commit 767e7e2e7c49b4e9981704254bcc813ace4f927e
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Tue Oct 9 04:06:08 2007 +0200

    Implemented lock transfer (halfway) during execution of synchronous calls.
    
    When process A calls process B, the locks held by A are unlocked
    and reserved for B's dispatcher thread to lock it. Other threads
    would block as if it's still locked.
    
    When B returns the call, the locks are reclaimed. A FUSION_ASSUME will
    fail if B did not unlock properly and the lock is hauled to A again.

commit 349b9c20fcfd76f842c648a45afa07050948aa56
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Tue Oct 9 03:55:07 2007 +0200

    Added fusionee_dispatcher_pid( fusion_id ) to look up the pid of the (last) thread reading from the device.
    
    There should always be only one thread reading, but for safety there's a FUSION_ASSUME failing if the pid changes.

commit 346fe2b0d6c395adc384d938237d98beba0ef0db
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Tue Oct 9 03:51:41 2007 +0200

    Fix missing cleanup of execution entry if caller receives a signal while waiting for the execution.

commit db58a75995528bb1afb049822e7a6385028eb580
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Tue Oct 9 03:47:31 2007 +0200

    Added missing do {} while (0) around FUSION_ASSERT and FUSION_ASSUME.

commit 071679df51b424eed8bd9f276fb45287396e52c9
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Tue Oct 9 00:48:43 2007 +0200

    Changed version check for bool from <= 2.6.18 to < 2.6.18.

commit 43a0afce0e61c8548e4045186b15b945a1f0368b
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Tue Aug 14 03:30:12 2007 +0200

    Align each on a 4 byte boundary if multiple messages are read at once.

commit 960ea01f754d7ac49fb543685e0c8ff6be2dcb71
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Mon Aug 13 20:59:07 2007 +0200

    Kill all threads when killing Fusionees.

commit 4eda904239946f6728694123e42b243f034e9a22
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Sun Aug 12 23:25:50 2007 +0200

    Change maximum number of pre-acquisitions to remember from 32 to 256.

commit 8316621e8657423fc98ef82df123a89be462b701
Merge: 0d36d9a... 559d701...
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Sun Aug 12 16:46:46 2007 +0200

    Merge branch 'master' of git+ssh://git.directfb.org/git/directfb/core/linux-fusion

commit 0d36d9a8b7654597d465cc6c5ef73da116f92e6a
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Sun Aug 12 16:46:30 2007 +0200

    Changed major id to 250. Dynamic majors please.

commit 45746f501feb31d11d57d5bd1095782f9208deeb
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Sun Aug 12 16:45:40 2007 +0200

    Show total number of dispatches in stat.

commit 559d7011ce70b8f1b5a897e3da2244a4a46c8408
Author: Denis Oliver Kropp <dok@halblang.zion.home>
Date:   Sat Aug 11 10:50:46 2007 -0700

    Build without signal blocking unless FUSION_BLOCK_SIGNALS is defined.
    
    Define PID_MAX_DEFAULT to PID_MAX if undefined.

commit d1530c06e206d39eabd2cf3cc24bde21c1eda8c7
Author: Denis Oliver Kropp <dok@shizo.(none)>
Date:   Sun Jun 3 19:11:53 2007 +0200

    Send proper channel number in reactor message.

commit 00bda99ecce8784820404f59418401bffb33b8fe
Author: Denis Oliver Kropp <dok@shizo.(none)>
Date:   Sun Jun 3 16:27:14 2007 +0200

    Signal blocking #2
    
    Only block SIGSTOP, SIGTTIN, SIGTERM and SIGINT.
    
    Better cleanup of killed threads.

commit f636045b5cef8556b7bc256244feb624f202ef58
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Fri Jun 1 09:54:18 2007 +0200

    Added optional timeout to FUSION_SKIRMISH_WAIT.

commit 568e9042a78e2596304d3a2d60d06a78e8b12603
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Wed May 30 16:53:12 2007 +0200

    Well, ~16k GB was a bit too much. Changed size to ~64 GB starting at 0x523000010000.

commit 65161b89c918ae092debc69ea9fb9d0b49e15f79
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Wed May 30 16:44:34 2007 +0200

    Start using architecture dependent SHM virtual address area.
    User space still uses 0x20000000 for a few root pages though.
    
    For x86_64 (address space!) use 0x500000010000 with length 0x0FFFFFFEF000.
    
    For mips use 0x50010000 and others 0x20010000 as before, length 0x1FFEF000.
    
    If there are conflicts on other platforms, please add an elsif :)

commit e6a752730968e94487c6b0a2070793fa200fcf1d
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Wed May 30 16:27:55 2007 +0200

    Fix typo in address range check.

commit e7898b414eee6009f71175231df921a01ea2553f
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Tue May 29 20:01:01 2007 +0200

    Revert "Defer signals until all locks are dismissed."
    
    Causing problems on exit...
    
    This reverts commit eeea7577d29255beb9b31f26ae0e5ce6e322c1fb.

commit 2e27256be550b18dbe96cf5bcc4f1bbdc245704c
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Tue May 29 17:05:01 2007 +0200

    FusionReactorDispatchCallback: If call_ptr is NULL and the dispatched message is of (void*) size and its content points to shared memory, then pass that into the dispatch callback's call_ptr rather than NULL.

commit eeea7577d29255beb9b31f26ae0e5ce6e322c1fb
Author: Denis Oliver Kropp <dok@shizo.(none)>
Date:   Tue May 29 13:55:36 2007 +0200

    Defer signals until all locks are dismissed.

commit ce344a8d5b5dd70913faf546f1054e5509a13fa8
Author: Denis Oliver Kropp <dok@shizo.(none)>
Date:   Mon May 28 19:48:23 2007 +0200

    Added FUSION_SKIRMISH_WAIT and FUSION_SKIRMISH_NOTIFY.
    
    When you hold the lock you can wait until you get notified,
    where others can get the lock while you're waiting.
    
    At the moment it requires the notifier to acquire the lock.

commit bfba2c7a09a52c7d7ddc0b8f451547cc13f35046
Author: Denis Oliver Kropp <dok@shizo.(none)>
Date:   Mon May 28 17:03:19 2007 +0200

    API break!
    
    Added channel parameter to low level messages and reactors.
    
    Attach, detach, dispatch of reactor require
    additional channel number specification now!

commit 51b9cfafcc587b1099bc14c949301d4ad3c075aa
Author: Denis Oliver Kropp <dok@shizo.(none)>
Date:   Mon May 28 14:36:55 2007 +0200

    Removed call_arg option from dispatch callback. It will be used for something else.
    
    Fixed missing dispatch callback if there were no recipients.

commit a8eebd851409f7faf5e25a42f281d605f7d13fc0
Author: Denis Oliver Kropp <dok@shizo.(none)>
Date:   Mon May 28 10:50:44 2007 +0200

    API break!
    
    Added serial value to call messages.
    
    Need to specify that serial when returning calls!
    
    +
    +     unsigned int        serial;
     } FusionCallReturn;
    
    +
    +     unsigned int        serial;
     } FusionCallMessage;

commit ee4c73c060fc15d0f6ec5fe1f822ba63b4ccfdaf
Author: Denis Oliver Kropp <dok@shizo.zion.home>
Date:   Sat May 26 20:39:30 2007 +0200

    Implemented a callback mechanism to get notified when all
    recipients have processed the dispatched message.
    
    Added FUSION_REACTOR_SET_DISPATCH_CALLBACK ioctl to bind a
    FusionCall to a reactor.

commit 487286a21a1055340249577a812b0c4a13ea40cd
Author: Claudio Ciccani <klan@directfb.org>
Date:   Tue May 15 21:21:50 2007 +0200

    Workaround to make forked processes work with fusion:
    reference local refs on fork() instead of duplicating them.

commit af37c4a7772d26711efecac0c8be8c977fbbbc04
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Mar 8 14:30:37 2007 +0000

    updated

commit adac43cbf3cc7dd88d941541f70c01171e83fa80
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Mar 8 13:10:10 2007 +0000

    another update

commit f286bdd8a7f99858367793b15b9fbde06f1182a4
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Mar 8 13:02:47 2007 +0000

    Fix fusion_sleep_on() for 2.6 kernels, using prepare_to_wait() and finish_wait() etc.

commit 49909476fc22f54d433577227c4214b25c405df3
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Mar 8 13:02:04 2007 +0000

    In fusion_entry_wait() after calling fusion_sleep_on() first check for
    signals, then check for timeout.
    
    In fusion_entry_notify() don't rely on valid "waiters" counter, but always
    do the wake up on the queue.

commit 9fe70f78fc378dc66ad2a4dcd43c1035a30f6af1
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Mar 3 23:40:22 2007 +0000

    updated

commit f0946d0028d3bf025396c5dffd4e26a18ee67845
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Sat Feb 17 14:17:36 2007 +0000

    Pass kernel version to depmod. Get the kernel patchlevel from the build kernel, not the currently running kernel.

commit ada4168a91e0569083bf49a962800f1a3d5b0e93
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Jan 29 00:43:50 2007 +0000

    Don't include <asm/types.h>.

commit a61bc37a5320522c79e2f84695871cae9ed03e5c
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Sun Jan 28 23:31:00 2007 +0000

    Make fusionee_poll() return POLLERR on error. The return value is a
    bitmask so returning the negative return value from lock_fusionee()
    doesn't seem like good idea.

commit e23eb0cd8d2c5e367916c38b6795f2923eea673b
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Jan 26 22:21:24 2007 +0000

    Added "-I$(KERNEL_BUILD)/include".
    
    Had to do this for building on a freshly installed Ubuntu Feisty (AMD64).

commit 60761ebd6ff10db95c326ac11e4762e05b638da7
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Jan 23 21:19:25 2007 +0000

    Don't use interruptible locking in fusionee_destroy() called by fusiondev_release(),
    which must complete.
    
    This should fix left over ghost fusionees which required reloading the module,
    causing some badness messages in the proc file system.

commit bb2eb4968758b58fb8b504c102f8e331d790f134
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Jan 20 04:16:22 2007 +0000

    Fixed O_EXCL (master) with slaves already waiting.

commit 4b67e870218621c628653383e8564014a62c422e
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Jan 20 04:03:01 2007 +0000

    Implemented waiting for master when entering a world as a slave
    by setting the O_APPEND flag during open().
    
    Some restructuring was necessary to delay allocation of FusionIDs
    until the world is really entered.

commit 3175a8ec99931abcddd015273c29a6a20fe403e5
Author: Marcel Siegert <mws@directfb.org>
Date:   Thu Jan 11 18:08:22 2007 +0000

    fix mentioned compile errors for kernel 2.6.20
    i took 2.6.20 as the minimum kernel version for the additional includes
    maybe that appeared already at 2.6.19
    
    linux-fusion/linux/drivers/char/fusion/fusiondev.c: In function 'fusiondev_deinit':
    linux-fusion/linux/drivers/char/fusion/fusiondev.c:253: warning: implicit declaration of function 'ClearPageReserved'
    linux-fusion/linux/drivers/char/fusion/fusiondev.c:253: error: invalid use of undefined type 'struct page'
    linux-fusion/linux/drivers/char/fusion/fusiondev.c: In function 'fusion_mmap':
    linux-fusion/linux/drivers/char/fusion/fusiondev.c:1011: error: dereferencing pointer to incomplete type

commit 5a3281b423576cd9122721e0ea62a1ada5dca6f9
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Sat Jan 6 13:36:58 2007 +0000

    Don't allow fusion_release() to fail. I'm not sure if it's a good idea
    but at least keeps my console working when XDirectFB crashes.

commit 43ae8f628bc5051c888b19081f3de41d506ea75e
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Sat Jan 6 13:26:36 2007 +0000

    Fix a potential double up() on dev->fusionee.lock.

commit 691b48aae9b6082589e8e28288fa2216a21b4c4d
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Sat Jan 6 13:19:37 2007 +0000

    lookup_fusionee() can't return a NULL fusionee on success.

commit ddf29fa6f36bc73a24c4d23ba5c9723c32d21882
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Sat Jan 6 13:09:45 2007 +0000

    linux/config.h is no more.

commit 13135b6f123ec049254c2acc906396b5cd897d3b
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Sat Jan 6 13:03:48 2007 +0000

    Fix bool redefinition. Patch from Daniel Laird.

commit c17e0d0a85fdac33104f5c98a26b38879e47465c
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Sep 26 07:12:36 2006 +0000

    updated

commit def00ab31b41098063d87fcf294b2f239029e45e
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Sep 26 07:10:44 2006 +0000

    Guess we can really remove it now.

commit a5defb1b50c69c6fe264b4e6f13cbf9fbbf8a4ac
Author: Marcel Siegert <mws@directfb.org>
Date:   Mon Sep 25 20:55:27 2006 +0000

    fix 64-bit compile warnings "casting from pointer to integer of different size"

commit 99a74d5f79d4b3dea55620397e41abb261dd2b2f
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Sep 15 08:25:11 2006 +0000

    Specify full path to include.

commit a90b486793ba9cd8f0a9814dbe8c6f39c214e5e1
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Sep 14 20:48:01 2006 +0000

    Just add include/linux/autoconf.h if it exists, no version checking or other magic :)

commit 2bf2a0643895acd4c034325d3003d3944df0945b
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Sep 14 19:20:31 2006 +0000

    Only add the include of autoconf.h if building for 2.6.

commit 4017fdf8c1dcba3b55a7c4497f70f29cc545714b
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Sep 13 13:02:23 2006 +0000

    Unlock entry before freeing it, thanks to Maxim Syrchin <syrchin@dev.rtsoft.ru>!

commit 63e33f4ba2b9c6b7d9127a410cbf286649ffa88f
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Sun Sep 3 13:40:34 2006 +0000

    Fix compilation with linux-2.6.18.

commit 8a0107b86f914ddd785965db6127bac85515becf
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sun Sep 3 11:50:55 2006 +0000

    Only mark reactors as being destroyed, but actually destroy/remove them
    when the last node detached. No more attach or dispatch is allowed though.

commit 6a5f2fdaedbfba3be26d032da45cec71353ba69c
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Aug 14 09:16:54 2006 +0000

    Added new ioctl FUSION_FORK to be used whenever a Fusionee has fork()ed and
    wants to stay in the world, i.e. without doing exec or similar...
    
    It doubles all local references, reactor and shmpool attach's and gives
    the forked Fusionee its own Fusion ID.

commit 43f5e224307316228661abfada9596233fb7d3b6
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Aug 4 16:28:01 2006 +0000

    FusionRef is also based on FusionEntry as most of the others are already.

commit 67942e80dd14257b9dbc18abaf7c01f7e752dc43
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Jun 30 08:54:55 2006 +0000

    New ioctl FUSION_SKIRMISH_LOCK_COUNT to query the lock counter of the
    calling thread. If it's locked by a different thread, it simply returns
    a zero count.

commit e8eab5976f31b866827aee25a477d0a0dce92e83
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Sun Apr 30 11:33:39 2006 +0000

    Made depmod respect $(DESTDIR).
    Thanks to Mark Pustjens <pustjens@dds.nl> for the idea.

commit 37a6ec7126729e3f29e11aee42c3625ecd3979e0
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Apr 26 11:48:24 2006 +0000

    Some fixes to the Makefile.
    
    Thanks to Alex Lau <alex@avengergear.com>!

commit 387dff127a6d69225cab5187a61c2f180d9218fe
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Apr 20 09:42:56 2006 +0000

    Fix memory leak in remove_node().
    
    Thanks to caizhihua@radio-tel.com!

commit c1ca520ccbccaf8874811875f8d88e1d234bda21
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sun Mar 26 18:44:19 2006 +0000

    Implemented detection of *potential* dead locks by remembering which locks
    are already acquired when another one is being acquired. As soon as a twist
    is detected, a warning will be printed.
    
    The procfs entry for each skirmish shows a list of skirmishs that had ever
    been acquired while itself was being acquired (the pre-acquisition list).
    
    Needs to be enabled by setting DEBUG=yes, e.g. "DEBUG=yes sudo make install".
    
    This does not detect dead locks right before they occur. It finds any
    theoretically possible future dead lock situation due to lock order switches!

commit bf6e6b3d24228093281f561a24e13adc43e84cc8
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Mar 1 04:16:37 2006 +0000

    Added FusionCallExecFlags featuring FCEF_ONEWAY which makes the caller
    return immediately after queueing the messages, not waiting for any response,
    making the call asynchronous, like a dispatch via a reactor.

commit 2de5da7c70441553632495389b7e440f65bf9714
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Mar 1 00:47:21 2006 +0000

    Show callers and number of waiters in proc.

commit 36dc6e07af70a02ce3a1af538e7f46f070bef405
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Feb 1 12:03:54 2006 +0000

    Applied patch from Tilman Sauerbeck <tilman@code-monkey.de>
    that adds DESTDIR support and updates the README, thanks!

commit 33cc70ec4dbe9c36499ea20f6b7bc87063ff7bac
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Dec 29 22:57:58 2005 +0000

    Made compilation happen before installation again :)

commit d8bd8db20004c7efd1855e3624595d6701f47912
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Dec 29 22:57:08 2005 +0000

    Builds with 2.6.15 now.

commit e64c3cd492a76e073cc21c4d07fbedce69e4dbb1
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Dec 27 07:21:13 2005 +0000

    Fixed deadlock in fusionee_send_message() by replacing the
    "listlock-item1lock-listunlock-listlock-item2lock-listunlock" sequence by
    "listlock-item1lock-item2lock-listunlock".
    
    Deadlock occured with concurrent fusionee_get_message() or fusionee_poll()
    doing a "listlock-item1lock-listunlock".
    
    Thanks to Hyperthreading for digging this one up :)

commit 0219da55c109ab2f1d5099b8dbbd5be04708f058
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Fri Dec 9 14:26:37 2005 +0000

    Fixed a typo.

commit 41c2ee1cbf0022de3a61066a447e2b7475b4f8bf
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Wed Nov 9 14:17:20 2005 +0000

    Made all targets phony.

commit 5bbf7f95f572ea96f3895bd33da9c6a341a07ab5
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Wed Nov 9 14:14:06 2005 +0000

    Remove fusion.[k]o from the old install location.

commit 8889c26a2fd63c03ca61be2241e724c0a1bf8829
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Tue Nov 8 15:34:41 2005 +0000

    Install fusion.h. Changed module install path.

commit fc5e1a7760a8a73f0c1eb66f038fe1b68c44fb4c
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Wed Nov 2 11:30:56 2005 +0000

    Updated udev instructions.

commit 1f66f6f4d564079f5b99b0a04d5d27a3d21dd22d
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Wed Nov 2 11:23:28 2005 +0000

    Made fusion_entry_wait() clear entry->lock_pid before calling
    fusion_sleep_on() thus making FUSION_ASSUME() happy.

commit f1c27f42b5344a932a202370bd2ecf782ab7b71d
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Wed Nov 2 10:31:42 2005 +0000

    Clear the reserved bit from shared_area before freeing it. Otherwise the
    kernel will complain and become tainted.

commit ad897de4b09dc56396e3a3a8cb6c36b7ac0d54bb
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Oct 29 05:11:10 2005 +0000

    Added .tmp_versions.

commit ec73cac144a6d8f2c78a2959f497cdb416d508e3
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Oct 29 05:10:43 2005 +0000

    Make it build on 2.4, let's see if it works, too :)

commit 5d1199a92d98d1793bcc38b4670a7c18a19f257d
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Oct 29 00:38:05 2005 +0000

    Fusion Kernel API 2.0
    
    
    Added management of shared memory pools
    - register pools with a specified maximum possible size
    - manage virtual address space so that all pools system wide can be mapped in parallel
    - dispatch messages when the shared memory heap has been resized (for remapping)
    - new ioctls are FUSION_SHMPOOL_NEW, ATTACH, DETACH, DISPATCH and DESTROY
    
    Added support for mmap() on the Fusion device itself, providing one page of shared
    memory serving as the root structure for all other shared information/memory in a world.
    
    Block all slaves at FUSION_ENTER until the master issues the new FUSION_UNBLOCK ioctl.

commit 12e7a9f05943571d84efbdfe85c3b0a075697bdc
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Oct 18 23:34:55 2005 +0000

    No more patches.

commit 08f33a0e5b252e8327e2ed1c42cf650e1514beee
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Oct 18 09:06:47 2005 +0000

    Removed some commented out stuff.

commit d66a4a1875a5e70dbd5bb0f41371779ac1aa7ceb
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Sep 13 14:32:25 2005 +0000

    Thanks to "Rob Shortt <rob@tvcentric.com>" for this patch fixing build with
    Linux 2.6.13 that introduced some changes to the device class API.

commit a6f2b6887459b998761e4cb5ee3bbc6a39bc07d0
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sun Jun 19 12:45:43 2005 +0000

    Fixed Makefiles for built-in mode.

commit 44b9ee7760c8f5899f1444954ecb774e60311d8a
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Mar 29 13:51:27 2005 +0000

    Ported from FUSION_GET_ID to FUSION_ENTER.

commit 7da47ead017cea2695af0436ac109035593a7d3e
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Dec 20 07:57:35 2004 +0000

    updated

commit 3289e14837b6e129ebb0974008ccac4c0c2a3855
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sun Dec 12 13:09:39 2004 +0000

    Increase performance by ~10% by not calling wake_up*() if nobody is waiting.

commit 78b08e7628630a21406f4a8e671bd042800bd625
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Fri Nov 26 10:23:15 2004 +0000

    xtime is struct timespec on some kernels.

commit e8efa5414fc1fccf1e282224a17f64173deb1128
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Fri Nov 26 10:16:28 2004 +0000

    MINOR() -> iminor().

commit e6f9992cd87315516703cf9fe72f69629a9b9e51
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Nov 16 19:22:37 2004 +0000

    Changed FusionReactor to use the generic FusionEntry code. It shows each
    reactor's name, number of total dispatch and current nodes in /proc/fusion.
    
    Show total number of locks for each Skirmish in /proc/fusion.

commit 33d07b44c43796b3937e795631e8beb09ce7f0d4
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sun Nov 14 09:13:57 2004 +0000

    Updated patches.

commit 8c96863abf1ddda0a929b987ba94bbefadfe26f8
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sun Nov 14 06:10:38 2004 +0000

    API version is 1.1 now.
    Older binaries compiled with the 1.0 API should work with a new 1.1 module.
    
    The generic fusion_entries_read_proc() prints the generic entry information,
    while the specific implementation can add further information like lock status.
    
    Take a timestamp whenever an entry is locked (each access) and show the
    difference between that and now in the proc view.
    his doesn't use the still too expensive do_gettimeofday() which produces
    10% overhead here. Instead, just copy the global "struct timeval xtime".
    
    Added FUSION_ENTRY_SET_INFO and FUSION_ENTRY_GET_INFO to store/retrieve
    meta information about an entry. Right now there's just a name to set.
    
    This is the full output while running XDirectFB (and some clients)
    along with 2x DFBTerm, df_cpuload, df_netload and df_dioload:
    
    :  dok@skunk[~] cat /proc/fusion/0/skirmishs
    :    3  ms  ( 5544) 0x00000014  Layer Context
    :   13  ms  ( 5544) 0x0000000b  Surface Manager
    :   13  ms  ( 5544) 0x00000002  Fusion Reactor Globals
    :   24  ms  ( 5544) 0x00000015  Layer Region
    :  321.2 s  ( 5544) 0x00000009  Clipboard Core
    :  326.3 s  ( 5544) 0x00000000  Shared Memory Pool
    :  681.0 s  ( 5544) 0x00000007  Surface Pool
    :  681.5 s  ( 5544) 0x00000013  UniQuE Window Pool
    :  681.5 s  ( 5544) 0x00000008  Window Pool
    :    0.2 h  ( 5544) 0x0000000e  Display Layer 0
    :    0.2 h  ( 5544) 0x00000003  Arena 'DirectFB/Core'
    :    0.2 h  ( 5544) 0x00000001  Fusion Arenas
    :    0.3 h  ( 5544) 0x00000012  UniQuE Context Pool
    :    0.3 h  ( 5544) 0x00000005  Layer Region Pool
    :    0.3 h  ( 5544) 0x00000004  Layer Context Pool
    :    0.3 h  ( 5544) 0x00000011  Display Layer 3
    :    0.3 h  ( 5544) 0x00000010  Display Layer 2
    :    0.3 h  ( 5544) 0x0000000f  Display Layer 1
    :    0.3 h  ( 5544) 0x0000000d  Screen 1
    :    0.3 h  ( 5544) 0x0000000c  Screen 0
    :    0.3 h  ( 5544) 0x0000000a  Colorhash Core
    :    0.3 h  ( 5544) 0x00000006  Palette Pool
    
    Added ":" to avoid white space stripping at the front of each line by logmail.

commit 39104a537c14ed87a738b059b4c24e635a6e3ea0
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Nov 12 17:39:23 2004 +0000

    New patches.

commit 8319d3d8e05d0d04ea147d78360fa5894a045028
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Nov 11 00:15:57 2004 +0000

    Updated.

commit 16901c508cb04cb0abf4face9ca6215d96e5cfe1
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sun Nov 7 12:20:53 2004 +0000

    Fixed duplicate watch notification sent during cleanup.

commit 3937c791f1ba24639561825a22859137ce4a03cb
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sun Nov 7 09:15:11 2004 +0000

    Added version numbering to the Fusion Kernel API.
    The major number is increased if binary compatibility breaks,
    while the minor number is increased for new features.
    
    Added new ioctl FUSION_REF_INHERIT to have a reference inherit
    the local reference count from another reference, i.e. adding
    the other's local count to its own. If the inherited reference
    is changed, the inheritor is also updated.
    
    There's just one reference that can be inherited from by each
    reference, but a reference can have multiple inheritors.
    
    Replaced FUSION_GET_ID by FUSION_ENTER. The new ioctl gets a
    struct containing the API version the applicaiton was compiled
    with. If the versions are incompatible, an error is returned,
    otherwise the fusion id passed back to the application via the
    struct.
    
    Use the 'type' field of the ioctl code to encode one of the
    following command group identifiers: FT_LOUNGE, FT_MESSAGING,
    FT_CALL, FT_REF, FT_SKIRMISH, FT_PROPERTY and FT_REACTOR.
    
    Now, the 'nr' field value is starting at zero within each group.
    
    The huge switch in fusion_ioctl() for all commands is replaced
    by a switch for just the different types (see above). Each type
    has its own function with a switch for the command numbers.
    It's not faster on my machine, but at least it looks much better.
    
    Fixed a memory leak when clearing local references upon process exit.
    
    Added some TODOs.

commit 5ab52a21af673f5433463619326491285629d8c1
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Aug 23 12:54:32 2004 +0000

    Added Print() method to FusionEntryClass.
    
    Added generic fusion_entries_read_proc() which uses Print() to
    write each entry into the buffer.

commit be86af04373d713282ca1bc09d21c47813557d2a
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Aug 18 14:09:42 2004 +0000

    Added FUSION_ASSUME(exp) to use instead of FUSION_ASSERT(exp) at some places.

commit b7c8bfcc99bcfec885578fb352775a3a7ce67d23
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Aug 18 14:00:31 2004 +0000

    Use new code for skirmish.

commit 87a17f9fdcd9613c85ff51a05e37f7fbab4ccc49
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Aug 17 20:00:32 2004 +0000

    Simplified timeout code.
    
    Added recursive purchase and some more error checking.

commit 274628deb2a61cc57c9a793a26144821950e4778
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Aug 17 17:24:36 2004 +0000

    Added base code for all primitives.
    
    Property already uses the new clean code.

commit 3257789127d71214d09a5854a63df73585ed5580
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Jul 30 15:01:43 2004 +0000

    Implemented recursive leasing, fixes dead lock during emergency shutdown.

commit 6cbb9c375e2238e67e18421a46b74058dbe3c101
Author: Ville Syrjala <syrjala@directfb.org>
Date:   Tue Jul 13 15:42:30 2004 +0000

    Added sysfs/udev support.

commit 0970bf1a21325abf0a4177f2edb96e90673ee32c
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Feb 23 23:13:42 2004 +0000

    ;->

commit f16799dec22b6fe59cddc379af3ff72cc1360f2b
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Feb 16 16:41:21 2004 +0000

    Added patch for 2.6.3-rc3.

commit 577012ae801c2bd4146660f0fe9e6ee13b6eaeae
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Feb 16 16:33:31 2004 +0000

    Removed obsolete patches.
    
    Added patch for 2.4.24.

commit 9b98cd85e8110c42a5543b4d4487402726492284
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Feb 16 12:27:37 2004 +0000

    Use major 252.

commit 21c5f0ba16585497d5b52048c344a8873aa20fe2
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Feb 6 12:43:58 2004 +0000

    Remove non-working patches.

commit 2b92a537d3fdce50cea753d6f76e1ebf62f17633
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Feb 3 04:35:23 2004 +0000

    Special version without semaphores for testing.

commit 9343decb1fa089269910f3e46ed8ece82d9b94e0
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Jan 30 15:13:14 2004 +0000

    Updated patch and changelog.

commit bb3aa48a9282721a9d863192e200f819e89956f6
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Jan 7 11:41:02 2004 +0000

    Fixed version check for "make install".

commit fa419aaafc9b90cfd0101821ea7d1fbdebcd88cb
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Dec 16 06:05:40 2003 +0000

    Automatically dismiss skirmishs upon death of a single thread.
    Previously only the death of all threads (of one fusionee) did that.
    
    This means that killing or canceling a thread releases its held locks now.

commit e341dbf1e78f578873de84e85fa6befd4bf997a7
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Dec 15 17:11:48 2003 +0000

    'make' and 'make install' works with 2.4 and 2.6 now.

commit 606500d365a573ea4206d892d5a3406f3b439c5b
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Dec 12 18:46:10 2003 +0000

    Use semaphores instead of spinlocks.

commit 4b38ef09ae6388ffcd16719ecebf432aec606e91
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Dec 4 12:55:49 2003 +0000

    Show the owner of a property in /proc.

commit 1a2ac5a30ced77ab3aa9ccf987e0af2319de5b3a
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Oct 29 12:11:05 2003 +0000

    Corrected locks for 2.6.0-test9. Also use non-irq versions now in both version.
    
    iminor() fix by Johannes Stezenbach <js@convergence.de>.

commit d320e017c1cb14956ab4673867ebc4b88d1ec975
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Oct 24 09:50:48 2003 +0000

    More padding in /proc/fusion/*/fusionees.

commit bb97deb79add6252468dba574ec3938ec943ef37
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Oct 22 17:21:12 2003 +0000

    Prepend local include directory.

commit 505b480e934176ecd082a2e8a1d7fcfedb342fdc
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Oct 17 05:34:21 2003 +0000

    Updated patches.

commit 7feb333f2bfe7166b12dd1cb70d53361b99a58db
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Oct 17 05:14:33 2003 +0000

    Removed "sizeof" from ioctl definitions.
    Use "MINOR" instead of "minor".
    
    Made ioctl numbers sequential.
    Use "switch(_IOC_NR(cmd))" to help gcc optimizing.

commit 68ce2a93d30fcd8dbfab7b9c791571e9f9fd2cf5
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Oct 14 07:52:01 2003 +0000

    Updated patches for 2.6, too.

commit 0a81376b765a4615763b0bb76316fb34ea5b9c87
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Oct 14 07:41:51 2003 +0000

    Updated patches, 2.6 will follow.

commit 0689abee339b7d6becabd14a8a48036a9252ab2d
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Oct 14 07:15:03 2003 +0000

    Added "/proc/fusion/*/stat":
    
    'lease/purchase   cede      attach     detach      ref up   ref down  prevail/swoop dismiss'
    '    139068     139068         969        841        1315       1161      186576     186296'
    
    These are total number of calls.

commit 70ea3ad45455a8ff9fc2de590aa1d4b8bb9f0214
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Oct 4 15:39:42 2003 +0000

    Fixed dead lock for SMP enabled kernels. I thought spinlocks were recursive.

commit 9bce62a3d7f08aad99d7726d0618db45d794f2dd
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Jul 25 12:46:10 2003 +0000

    Remove old and broken patches.

commit e3396149f6859cc44c2b5d716393eecbf48e5106
Author: Andreas Hundt <andi@directfb.org>
Date:   Thu Jul 24 11:42:43 2003 +0000

    updated patch

commit 448b86e52cf20656948ff48999fef5123206df25
Author: Andreas Hundt <andi@directfb.org>
Date:   Thu Jul 24 11:38:44 2003 +0000

    fix oops during initialization on 2.6.0-test1

commit 42db5856afcb14a5fd40d4d6ae07e1ca45010d2b
Author: Andreas Hundt <andi@directfb.org>
Date:   Thu Jul 24 01:12:58 2003 +0000

    build patches for 2.6.0-test1

commit 8e72e18bd071a5db83e4a71bcdb1dfb1d843f9ee
Author: Andreas Hundt <andi@directfb.org>
Date:   Thu Jul 24 00:13:33 2003 +0000

    fix typo ;-)

commit 85e8c281710d59ac395cf9d84dcc74686dd347c4
Author: Andreas Hundt <andi@directfb.org>
Date:   Thu Jul 24 00:10:17 2003 +0000

    support for linux-2.6-test1 (new internal devfs api) should now work with 2.4, 2.5 and 2.6

commit 61da68633c1c69323c9147a8cdce78237dd92c84
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Jul 21 13:05:31 2003 +0000

    New ChangeLog.

commit c4c44be9bd98ebbbafc02bf0f19517e8101175a6
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Jul 10 06:57:38 2003 +0000

    Improved documentation, wrote about shared memory and arenas.

commit 1c4deb5018f447ef18eeec11397be7f951d18a75
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Jul 9 23:31:28 2003 +0000

    Added FUSION_REF_WATCH which can be used to install a reference watcher with
    an optional argument that will be passed to it.
    
    The reference watcher is a call that is executed when the reference counter
    gets zero. The optional argument could, for example, be the id of a user space
    resource associated with that reference.
    
    This is a much better way than polling all references regularly.

commit 74296130c78de7f9e8e2434fd235953ae788644e
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Jul 8 13:47:48 2003 +0000

    This time right.

commit 1f31c8e8a9a9c3c9ce25ea8912e60ec848997a01
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Jul 8 13:42:16 2003 +0000

    Forgot to convert a FusionCallHandler to void*.

commit d8b658be9bcdb5d1e0dd2c2621c910dba2bd1c05
Author: mm <mm>
Date:   Tue Jul 8 11:48:25 2003 +0000

    - revert dok's 1.10 -> 1.11 patch to make the build-sys work again

commit 8115eb657be2991bc58141d297d9429b093cccec
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Jul 7 18:03:31 2003 +0000

    Removed call handler type definition.

commit 668874d50547c91c8fc49f11f4d1c5e37542e5e1
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Jun 30 16:46:00 2003 +0000

    Fixed patches, didn't compile with SMP or preemption.

commit 442cfacdf9dd84b77cc182787ee029d6c32799e3
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Jun 30 16:41:12 2003 +0000

    Fixed undeclared fusionees_lock, which didn't cause an error if spinlocks
    are not used.

commit e1c7434c1e58cdc29061b8175be58145035a54c1
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Jun 24 16:03:43 2003 +0000

    Show lock-holder (pid and fid) in /proc/fusion.

commit 3a2df192edcba5f6260222ac56ab12a738c1ef43
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Jun 17 12:38:40 2003 +0000

    Added version of throughput that uses a pipe.

commit e8fecf843ee355f5efc38e67f53e5bab137cc066
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Jun 16 17:47:03 2003 +0000

    Added support for multiple Fusion "worlds" via device minor.
    
    New directory layout is
    	mkdir /dev/fusion
    	mknod /dev/fusion/0 c 253 0
    	mknod /dev/fusion/1 c 253 1
    	...currently limited to eight minors
    
    The /proc directory layout also changed this way.
    
    Updated all patches.

commit 9eea0ad8c74a3e7f01689ab85798c9bd03c1ab8b
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Jun 7 01:30:49 2003 +0000

    Added Makefile for 2.5 kernels to the repository.

commit 5e325982975382e11012481055a6b125d4b713e3
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Jun 5 14:32:08 2003 +0000

    Fixed Fusion for 2.5 kernels! No more tons of kernel debug messages,
    not even a single note about a process exitting with an active preempt counter.
    
    1)
    Call kmalloc() with GFP_ATOMIC instead of GFP_KERNEL which isn't allowed while a
    spinlock is held. I'm not sure if we should use GFP_ATOMIC including __GFP_HIGH,
    but it's the only GFP_* which doesn't include the offending __GFP_WAIT.
    
    2)
    Added missing spin_unlock() calls. I didn't forget them but dropped the
    unlock in case the spinlock is thrown away and never locked again anyways.
    Unfortunately these are always required for preemptible kernels because
    spin_lock/unlock() increase/decrease the preempt counter.

commit 3b97815226dbaacc91ef417c78ecd6dbf96dbbf2
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Jun 4 18:58:10 2003 +0000

    Added new ioctl FUSION_KILL which sends a signal to one or more fusionees
    with an optional timeout for waiting for their termination.

commit b4d1afca28326ec4d50d0d6bc736fb2e30e27613
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Jun 2 17:11:32 2003 +0000

    Updated patches, include recent improvements and fixes.

commit a2965ee815a1dc90bab0bcfe2ec048311b155bbb
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sun Jun 1 21:02:08 2003 +0000

    Extended TOC.

commit 4e247f38eb31d1307b0c49cb491d3362968fe08d
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sun Jun 1 19:51:31 2003 +0000

    Initial version of conceptual documentation.

commit 3719429be995c338f44aa677415a4dcbb6f5ad33
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue May 13 13:17:16 2003 +0000

    Do yield() in fusion_property_cede() if property was purchased.

commit 8f295629246a4b7530eb266aa4620a41bbd6330f
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon May 12 17:13:36 2003 +0000

    Fixed missing arguments.

commit 7110ab53e4fa19d3439afb8a5562a9df9efe94a5
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon May 12 17:08:54 2003 +0000

    Unlock all refs locked by a process during its death.
    Added error checking (lock owner verification) to ref locking.

commit ee10497ca7edd6cf2ea3033df3325137e5a5cfed
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon May 12 16:48:18 2003 +0000

    Increased the purchase timeout to one second. Don't wait if property is
    purchased for more than one second.
    
    Timeout of leasing stays at 0.1 sec, not waiting at all if property is
    purchased for more than 0.1 sec.

commit 7ceb6f1082ea74cf703d1f2bf440607500d1672f
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu May 8 18:52:35 2003 +0000

    Don't do a lease/purchase with timeout if property has been purchased more
    than 100 ms ago. Otherwise timeout after another 100 ms. Better suitable
    for long time purchasing which fullscreen apps do.

commit 611cf2452979bf4085999af94a73eb350cbca614
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu May 8 17:58:47 2003 +0000

    Added timeout option to fusion_sleep_on() that is used by properties.
    Leasing/purchasing doesn't fail immediately now if a property is purchased.
    Instead it fails after a timeout of 20 ms.
    
    This improves DirectFB DRI interoperation significantly as DirectFB doesn't
    give up early to access the hardware if an OpenGL application does the rendering
    which is most of the time, but depending on the framerate very short periods.
    If this period is short enough (say having 40 fps) the timeout will never
    expire and no software fallback is used.
    
    Running OpenGL with ~150 fps while scrolling in an IDE which is half transp.
    above runs quite smooth.

commit 70bc5f305cc8b49a44d0eedcf538d95298373dbd
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Apr 15 11:22:15 2003 +0000

    Fixed typo.

commit 0d9ee559e38e51f527d70930a080bdff03d48ea6
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Apr 15 09:10:03 2003 +0000

    Some more items.

commit a1e653642b8315b7f39c548dbf9ce3b5ee5ee5cc
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Apr 11 11:18:43 2003 +0000

    Updated patch.

commit e09094a9d3b662d657c1aec51004a91fd3ec7e12
Author: Andreas Hundt <andi@directfb.org>
Date:   Tue Apr 8 10:00:39 2003 +0000

    forgot to include <linux/version.h>

commit b53d718728ce5c06e26b8b75444200b0f000eca4
Author: Andreas Hundt <andi@directfb.org>
Date:   Tue Apr 8 09:39:23 2003 +0000

    fix compilation for 2.5.xx

commit c22ce9b9ec7055b43dd2788477c69d20d31b5478
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Apr 7 13:10:03 2003 +0000

    Updated 2.4 patches.

commit e6ba11ad8ca7ed646c2a75c4a243745ee2a1c0a3
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Apr 7 12:10:59 2003 +0000

    Added an RPC benchmark and a Makefile.

commit 661cdd11250d30ecd7558b2245f203c27f300a05
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Apr 7 12:04:20 2003 +0000

    Added utility function fusion_sleep_on() which does the same as
    interruptible_sleep_on() but takes an additional argument specifying
    a spinlock that is unlocked after the wait queue has been modified.
    
    Replaced everything like:
    
    foo_unlock (foo);
    /* unsafe gap */
    interruptible_sleep_on (&foo->wait);
    
    by:
    
    fusion_sleep_on (&foo->wait, &foo->lock);
    
    Closes the unsafe gap (resulting in blocking) on SMP or preemptible kernels.
    
    
    Forgot to call fusion_call_destroy_all() in fusionee_destroy().

commit 874ed124c32313a207caecc68faa3d01a75ed8df
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Apr 5 02:45:42 2003 +0000

    Updated patches.

commit c1d1f057f6de03d8459618903bd21f5138724e5c
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Apr 5 02:42:59 2003 +0000

    Added synchronous RPC mechanism (with return value).

commit 6247a1cc81c2993048be89ccc1ef081852041346
Author: Andreas Hundt <andi@directfb.org>
Date:   Mon Mar 17 23:04:54 2003 +0000

    built patches for linux 2.5.64.
    
    it works somehow, but I get a kernel oops when applications using fusion quit. (even the latency.c application).

commit b8ddadfbeb3903ee0a308409eb4616973285a128
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Mar 3 22:19:15 2003 +0000

    Remove copy of fusion.o during "mkae clean".

commit fd58464b1147e576838d80413b20cb3aad5b10b5
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Mar 3 22:17:42 2003 +0000

    Added Makefile for compiling the fusion module outside the kernel tree.
    Simply type "make" in the linux-fusion directory. You may need to adjust
    the kernel path in the Makefile or use "make KERNEL_SOURCE=/foo".

commit e5fa4848161fa5691bd227b47d5c83abbc74e750
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Mar 3 16:18:28 2003 +0000

    Don't free already freed message data during fusionee_destroy().

commit 8db10ffd7702c80e15339ec5fae1d30c6826ce13
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Mar 3 15:23:32 2003 +0000

    Increased message throughput by about 15% by doing one kmalloc per message
    (header/data) instead of two.

commit 5b96a8fdd82a60fe8fde306dcf6692f0ee229610
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Feb 28 19:56:03 2003 +0000

    Updated patches, added ChangeLog.

commit 81846682c4104f9a3a27dfcbfae717ee24050545
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Feb 28 19:45:34 2003 +0000

    Ouput total messages sent and received.

commit 09bc80e2d65a98028033672e53f2060cdb7a8bf4
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Feb 13 19:14:15 2003 +0000

    Output total number of messages received by each fusionee in /proc.

commit 209f9131a23f05c5b5e2dde04b7bde7d96bd1d01
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Feb 4 20:03:14 2003 +0000

    Plugged a very small mem leak.

commit 64d960d5cc3985fd3898fec22275d9c471469b6b
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Jan 31 17:48:10 2003 +0000

    Added #if 0'ed code that gives the receiver thread realtime priority.
    This results in 100% synchronized message dispatching and processing
    independent of the nice value of the sender thread. Unfortunately it
    limits the throughput at 280 thousand messages per second on my machine.

commit 43b91a1bc4cf1b7a27efb32890179d6a5d07a45c
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Jan 31 16:04:45 2003 +0000

    Measure how many messages are dispatched synchronized.
    I got 0% with nice 20 and 25% with nice -20. What if the
    receiver has a higher priority?

commit 89b7e43c7a795bc456e41fbb40ae38d55f7f43f8
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Jan 31 15:33:54 2003 +0000

    Small program measuring the messages per second sent/received.
    Interestingly it depends on the nice value. I got about 370 thousand
    messages per second with a nice value of zero, but got about 530 thousand
    messages per second with a nice value of 20.
    
    Doing "cat /proc/fusion/fusionees" during the test shows that no messages
    queue up when using a nice value of -20. I guess the wakeup_interruptible()
    within the kernel code leads to synchronized message dispatching.
    I will add another test that measures the percentage of messages that got
    received and processed "virtually during" the ioctl to send them.

commit 935ab96d3fc9fdd7a2403cf0e7cda47ffe5d340f
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Jan 30 22:24:35 2003 +0000

    Wrote small test program that stops time of travel of a message.
    It creates a new thread for receiving messages (generic receiver code
    doing a blocking read and decoding the message headers). The main
    thread will send ten messages with a random pause between.
    One time stamp is taken just before the message gets sent. The other
    time stamp is taken after the receiver decoded the message header.
    
    This piece of code is also the first stand alone example of using the device.
    
    
    The output (PIII 800, 2.4.20-rc2 preemptive):
    
    Got message with 0.019 ms latency.
    Got message with 0.008 ms latency.
    Got message with 0.009 ms latency.
    Got message with 0.009 ms latency.
    Got message with 0.008 ms latency.
    Got message with 0.008 ms latency.
    Got message with 0.011 ms latency.
    Got message with 0.007 ms latency.
    Got message with 0.007 ms latency.
    Got message with 0.011 ms latency.
    
    
    These values would be the same if the sender and receiver wouldn't be
    two threads but two distinct processes. This also doesn't say anything
    sure about the throughput yet, though 1000/0.007 gives a quite sane
    value of about 143 thousand messages per second.
    
    I'm not sure if the preemptive kernel patch influences this latency,
    but I guess the throughput would be affected.

commit 079e978ed9b59cd55483d0ff3e44932135fbbdf4
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Jan 14 15:21:35 2003 +0000

    Updated all patches.

commit 532522d097f2bca9c7e0f8378c57cba4054994f4
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Jan 14 15:09:30 2003 +0000

    "Move-to-front" optimization added to other primitives, too.
    Reset all id pools during cleanup (last fusionee closes device).

commit 7e085b807cabab0966168bca59cdaaeb1b650162
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Jan 14 13:50:58 2003 +0000

    Needed to fix failing hunk in miscdevice.h

commit 99495db6850ccd25e1d7ed621c44db9e28218b46
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Jan 13 18:56:53 2003 +0000

    Updated patches.

commit 2e6610cff4c1a553f15fc2e574a4313f33d50c75
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Jan 13 18:46:09 2003 +0000

    Move skirmish to the front of the doubly linked list each time it is locked.
    This nearly doubled the frame rate of ClanBomber2 on multi app core
    on my system.

commit 5f7f625fdb4e4c7b15efa2315e83ef2cf1eca764
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Dec 13 17:13:18 2002 +0000

    Updated patches.

commit 197ebda8a2c0b473421ac4c3787ce3b4aff1f7bc
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Dec 13 17:05:01 2002 +0000

    Added FUSION_PROPERTY_HOLDUP that sends a SIGKILL to the process that
    purchased the property (e.g. a hanging fullscreen application).

commit 4c70abbcaa01800833b83a729e8faad0c2a64a23
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Dec 5 17:46:50 2002 +0000

    New patches.

commit 8c2125cd4084277f6df32349aaa2ec86bc482533
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Thu Dec 5 17:36:08 2002 +0000

    Implemented poll(). Making select() work, too.

commit 77d8395a2593cfd50ea5922e83638c2c26d51867
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Nov 12 21:30:39 2002 +0000

    Updated patches.
    Fixed copyright info.

commit a5efa672acff6b777e92f4f5ae54e5e54af04086
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Nov 12 21:21:12 2002 +0000

    Implemented auto-cleanup if last fusionee closes the device.
    
    ipc_cleanup script is obsolete now. Simply kill all processes having
    /dev/fusion open and everything gets cleaned up.
    
    If it will ever crash... ;)

commit 65642328b9fbbfddf1ad9b5dd9bcf67227fe08bf
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Nov 11 04:10:44 2002 +0000

    Updated patches and TODO file.

commit b3f88da7c4c7164231accdc8ce06432223c31b3e
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Nov 11 01:58:58 2002 +0000

    Added messaging to the Fusion Kernel Device. There are two ways of sending
    messages. The simple way is to use FUSION_SEND_MESSAGE specifying the
    recipient's fusion id and an optional message id. The other way is to
    create a reactor with FUSION_REACTOR_NEW which is the process shared
    part of the old multi app reactor. FUSION_REACTOR_ATTACH adds the callers
    fusion id to the list of attached fusionees. FUSION_REACTOR_DISPATCH
    is used to send a message to all attached fusionees. So reactor_dispatch()
    only needs one system call (instead of two plus two times the number
    of attached fusionees).
    
    Messages are received by doing a read() on the device. This way multiple
    messages (depending on the read buffer size) are received by one system
    call.
    
    There's only one thread now for all reactors.
    
    FusionSkirmish is a recursive lock now.

commit c10fb7ab0d1611e99a54eadc26d7107ec9c64f80
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Tue Nov 5 21:42:00 2002 +0000

    No need to patch misc.c, fusion_init() got called twice if compiled in.

commit f09f81571c24098102d915696bf947eb90c7e507
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Nov 4 16:41:26 2002 +0000

    Updated.

commit 932374095a6672ea3ec9b9841dd1cd838c0aa0ee
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Nov 4 16:28:36 2002 +0000

    Implemented FusionProperty in kernel space (200k -> 700k lease/cede per sec).

commit c5dca1868608329a9c21e07156d068ee0085867e
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Nov 2 06:09:54 2002 +0000

    Fixed.

commit 8da48b1129509fdb130f0a47cc5053f14dc7a87c
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Nov 2 03:26:59 2002 +0000

    fusion_init/exit non-static now.

commit ddac134bb27ca074b7a5006caf67200dd993c6c9
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Nov 2 03:25:00 2002 +0000

    Should really work now if compiled in.
    Added 2.4.16 patch.

commit 37feb82a0ffc00ba7331881134e25523d146f13e
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Nov 2 02:19:58 2002 +0000

    New patches.

commit fa3d384f61b7665db73cb6a5d3641128c82a2d15
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Sat Nov 2 01:28:29 2002 +0000

    Added skirmish implementation,
    no benefit yet except speed (600k -> 700k prevail/dismiss per second).
    
    SIGSTOP defering/blocking could be added.

commit 84f322339abc2ea5cb1bb5809e28a26c75b313bb
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Nov 1 15:13:02 2002 +0000

    Use ANSI C member initialization.

commit 753c0a11b5a8bf60281bb8af88ec6ac56ca26704
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Nov 1 14:55:58 2002 +0000

    Added /proc/fusion/fusionees. Minor cleanup. Added locking to fusionee code.

commit 4e0bcb5b40c6df9e2332c16b7c910dd2bc483f56
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Fri Nov 1 13:59:12 2002 +0000

    Remember pid of process that created the reference, shown in /proc/fusion/refs.
    Added spin_lock() to fusion_ref_destroy() before destroying the reference.

commit ed2b0cf08869cbbe4bfa17f44c17b67d79c61b9c
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Oct 28 17:31:32 2002 +0000

    Added TODO file with three items.

commit c3e12dda48467db2bc7faec7b598170592cdf92c
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Oct 28 17:17:03 2002 +0000

    Should work now if built in (not using as a module).
    Updated patches.

commit 9c85229ac56321c9472a0ba6d686a463e6896c40
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Mon Oct 28 17:04:55 2002 +0000

    Added proc entry ("/proc/fusion/refs" for now).

commit 2b77e705af5883f504e0ed13ba870d84635e60ba
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Oct 16 16:03:39 2002 +0000

    Implemented fusion_ref_zero_lock() via wait queue.

commit b3ff6a21f01288371e7cae477e00a5937b237966
Author: Denis Oliver Kropp <dok@directfb.org>
Date:   Wed Oct 16 13:05:47 2002 +0000

    New CVS module for the (still optional) Fusion Kernel Device.
