summaryrefslogtreecommitdiff
path: root/src/os_posix
Commit message (Collapse)AuthorAgeFilesLines
* Boolean conversion pass over include.Keith Bostic2015-09-271-6/+6
|
* Boolean conversion pass over os_posix.Keith Bostic2015-09-246-20/+21
|
* Boolean conversion pass through block, conn and evict.Keith Bostic2015-09-233-7/+7
|
* WT-2093 Use the C99 bool type to clarify when functions return true/false.Michael Cahill2015-09-081-2/+2
|
* Inline __wt_cond_wait so we don't need different POSIX/Windows versions,Keith Bostic2015-09-041-16/+4
| | | | minor restructuring for fewer instructions/branches.
* WT-2085 Run some of the log_server threads operations more frequentlySusan LoVerso2015-09-041-4/+21
| | | | than others. In particular, flush the log slot 20x per second.
* Merge branch 'develop' into wt-1967-evict-anyKeith Bostic2015-08-211-3/+9
|\
| * Tweak line wrapping.Michael Cahill2015-08-211-7/+7
| |
| * WT-2050 - switch PRIu64 with WT_SIZET_FMTDavid Hows2015-08-211-7/+9
| |
| * WT-2050 - when memory allocation fails, say how much memory we were trying ↵David Hows2015-08-211-3/+7
| | | | | | | | to allocate
* | Merge branch 'develop' into wt-1967-evict-anyKeith Bostic2015-08-193-10/+10
|\ \ | |/
| * Rename atomic functions from number of bytes to number of bits.Keith Bostic2015-08-133-10/+10
| |
| * Merge branch 'develop' into wt-2025-inline-atomic-functionsKeith Bostic2015-08-132-3/+3
| |\
| * | WT-2025: Inline the gcc/lint (but not the Windows) atomic functions soKeith Bostic2015-08-063-10/+10
| | | | | | | | | | | | we get type checking and arguments are never evaluated more than once.
* | | Merge branch 'develop' into wt-1967-evict-anyKeith Bostic2015-08-132-3/+3
|\ \ \ | | |/ | |/|
| * | WT-2038 Use TAILQ for all lists. Otherwise remove is O(N), which may not ↵Michael Cahill2015-08-132-3/+3
| |/ | | | | | | show up until running real workloads (while holding locks).
* | Add a verbose message on close, it's useful to be able to match upKeith Bostic2015-08-081-0/+2
|/ | | | open/close pairs.
* SERVER-19751 Retry pthread_create on EAGAIN or EINTR.Michael Cahill2015-08-051-2/+4
|
* We've already tested the previous values for correctness, so we don'tKeith Bostic2015-08-031-12/+7
| | | | need to re-assign them before attempting the CAS instruction.
* Merge branch 'mutex-overflow' into rwlock-noshiftKeith Bostic2015-08-031-2/+2
|\
| * Try and clarify the second example is a continuing of the first one.Keith Bostic2015-08-031-2/+2
| |
* | WT-2023 Avoid bit shifts during read-write lock operations.Michael Cahill2015-08-031-26/+15
|/
* Merge pull request #2101 from wiredtiger/mutex-overflow-simplifyMichael Cahill2015-08-031-27/+11
|\ | | | | WT-2023: Mutex overflow simplify
| * Fix a commentKeith Bostic2015-08-021-1/+1
| |
| * An experimental change: replace the WT_ATOMIC_FETCH_ADD8 calls toKeith Bostic2015-08-021-27/+11
| | | | | | | | | | | | | | allocate and increment the ticket value with WT_ATOMIC_FETCH_ADD2 calls the allocate/increment just the ticket field. This means we don't overflow when acquiring tickets, which allows us to remove ticket overflow handling entirely.
* | Merge pull request #2099 from wiredtiger/mutex-overflow-cheap-testMichael Cahill2015-08-031-16/+24
|\ \ | | | | | | WT-2023: Avoid the CAS instruction in the trylock functions.
| * | There's a cheap register test we can do in both trylock cases to avoidKeith Bostic2015-08-021-16/+24
| |/ | | | | | | the CAS instruction.
* | Add a comment about why we need the memory barrier during write unlock.Keith Bostic2015-08-021-2/+6
|/
* Sleep eventually when waiting on a write lock, the same way we sleepKeith Bostic2015-08-021-6/+17
| | | | | | eventually when waiting on a read lock; a write lock has to wait on all previously ticketed readers and writers, so is likely to wait longer than a read lock.
* Margo's review: rename "us" to "wr", it's the writers and readers fieldKeith Bostic2015-08-021-36/+31
| | | | combined, rename "val" to "ticket", fix/reword some comments.
* Fix a couple of comments.Keith Bostic2015-08-021-2/+8
|
* Add another comment.Keith Bostic2015-08-011-0/+5
|
* I agree with Don, there's a bug in the __wt_try_readlock() function whereKeith Bostic2015-08-011-17/+141
| | | | | | | | | | | | | the readers value can overflow into the users value, and so it needs to be masked so it doesn't overflow. Use a masking operation rather than a separate variable. Rename the WT_RWLOCK.pad field to be "overflow", it's not a padding byte. Now that I've loaded this all into my head again, take Don's Wiki page as a starting point and comment the lock structure and how all of this stuff works, the web page we reference might not be around the next time we need to figure out what's going on.
* WT-2023: The users field intentionally overflows into pad; comment that.Don Anderson2015-07-281-5/+4
|
* WT-2023: Don't allow increments of 16-bit values to overflow into other fields.Don Anderson2015-07-271-2/+7
|
* WT-2016: Change wt_rwlock_t packing styleMark Benvenuto2015-07-211-1/+1
|
* Merge pull request #1991 from wiredtiger/open-directory-full-pathsueloverso2015-05-221-7/+0
|\ | | | | Sync the log directory when creating log files, not its parent
| * Sync the log directory when creating log files, not its parent. That code ↵Michael Cahill2015-05-221-7/+0
| | | | | | | | | | | | is the only caller of __wt_open with WT_FILE_TYPE_DIRECTORY and it passes in the correct path, but __open_directory was stripping the last component. refs WT-1941
* | Remove the WT_ATOMIC_CAS_VAL family of macros: use the true / false versions ↵Michael Cahill2015-05-201-2/+2
|/ | | | | | everywhere. refs WT-1938
* Revert part of the last change, Linux needs the cast to (void *).Keith Bostic2015-05-181-2/+5
|
* Solaris' pthread_t is an (unsigned int), #ifdef the thread ID code toKeith Bostic2015-05-181-8/+8
| | | | avoid gcc -Wint-to-pointer-cast warnings. Reference WT-1934.
* __wt_strdup() is a one-liner that calls __wt_strndup(), inline it.Keith Bostic2015-05-011-11/+0
|
* If our caller doesn't handle errors correctly, ensure a free won't fail.Keith Bostic2015-04-291-0/+6
|
* Don't set HAVE_POSIX_MEMALIGN automatically when TCMalloc supports it,Keith Bostic2015-04-251-6/+0
| | | | | | | | HAVE_POSIX_MEMALIGN is checked in other parts of the system, and, it configures Linux buffer alignment during auto-configuration. If we're ever on a system using autoconf that wants TCMalloc to provide the posix_memalign function, we'll need to fix up the autoconf script to handle that.
* TC MallocMark Benvenuto2015-04-241-0/+21
|
* The OS layer ftruncate() code sets the WT_FH file size, so whenKeith Bostic2015-04-121-3/+1
| | | | | | | | ftruncate is used to extend the file, we skip over the bytes added to the file during block allocation, and verify eventually fails because there are unverified blocks in the file. Reference #1871.
* Merge branch 'develop' into sweep-file-handle-minimumKeith Bostic2015-04-091-2/+1
|\
| * CID 72073 (#1 of 1): Redundant test (DEADCODE)Keith Bostic2015-04-091-2/+1
| | | | | | | | dead_error_condition: The condition session != NULL must be true.
* | Update the sweep handle minimum to not use a statistic.Alex Gorrod2015-04-091-3/+2
| | | | | | | | A couple of other tidy ups.
* | Merge branch 'develop' into sweep-file-handle-minimumKeith Bostic2015-04-082-3/+3
|\ \ | |/
View Lorry Controller Status