Nick Downing [Sat, 8 Jun 2019 06:21:36 +0000 (16:21 +1000)]
Unify process_alloc() and process_realloc() with PROCESS_ALLOC_MODE_REALLOC bit
Nick Downing [Sat, 8 Jun 2019 05:56:35 +0000 (15:56 +1000)]
Change clunky swap allocation stuff to use a normal realloc, but suppress copy
Nick Downing [Sat, 8 Jun 2019 05:39:07 +0000 (15:39 +1000)]
Simplify pool_alloc() calling convention so that offset is always specified for POOL_ALLOC_MODE_REALLOC, remove some special cases meant to aovid the move_up()
Nick Downing [Sat, 8 Jun 2019 01:50:40 +0000 (11:50 +1000)]
Change process_realloc() calling convention to be the same as process_alloc()
Nick Downing [Sat, 8 Jun 2019 01:31:15 +0000 (11:31 +1000)]
Change process->in_core_blocks to process->in_core_block relative to block_base
Nick Downing [Sat, 8 Jun 2019 01:25:54 +0000 (11:25 +1000)]
Fix wrong #ifdef INDIRECT_SWAP
Nick Downing [Wed, 5 Jun 2019 13:59:19 +0000 (23:59 +1000)]
Make do_swap_read_write(), do_swap_out(), and process_run() use process_calc()
Nick Downing [Tue, 4 Jun 2019 14:17:09 +0000 (00:17 +1000)]
Replace process->para, process->paras with process->para_base, process->para_limit such that these can be arbitrary (including negative) implying some fixed origin that keeps its relationship to the backing store under resizing, where previously process->para was restricted to the range 0..BLOCK_PARAS-1, rewrite horrible process_get_state() to process_calc() which converts these to blocks, majorly revamp the test script generator and runner to work natively with this scheme
Nick Downing [Mon, 3 Jun 2019 11:30:47 +0000 (21:30 +1000)]
Change process_get_state() calling convention slightly, replaces core_base and swap_base with core_block and swap_block, makes core_para and swap_para absolute
Nick Downing [Mon, 3 Jun 2019 11:10:41 +0000 (21:10 +1000)]
Give do_swap_read_write() routine more context so it can later do clipping etc
Nick Downing [Mon, 3 Jun 2019 11:06:32 +0000 (21:06 +1000)]
Rationalize use of core_blocks (becomes: block) and swap_blocks (becomes: swap_in when swapping in, unnecessary when swapping out), remove core_base, swap_base
Nick Downing [Mon, 3 Jun 2019 09:35:51 +0000 (19:35 +1000)]
Make swap_move() only be defined for INDIRECT_SWAP where it only moves the table memory, replace core_to_swap_copy() and swap_to_core_copy() with swap_read_write() which works in blocks and takes a direction flag, implement do_swap_read_write() to abstract out the swapping code which is independent of the direction
Nick Downing [Sun, 2 Jun 2019 10:38:33 +0000 (20:38 +1000)]
Change process->swap_blocks to process->core_blocks, stored iff INDIRECT_CORE
Nick Downing [Sun, 2 Jun 2019 10:08:22 +0000 (20:08 +1000)]
Fix minor spacing issue
Nick Downing [Sun, 2 Jun 2019 06:29:29 +0000 (16:29 +1000)]
Simplify the swap-in and swap-out loops, since have enough status information in the process that we do not need special loop entry code for new process/victim
Nick Downing [Sun, 2 Jun 2019 04:34:05 +0000 (14:34 +1000)]
Make all processes be in LRU list, with victim pointer separating fully in core processes from fully in swap processes, processes before pointer must have a core item (non-INDIRECT_CORE), process immediately before pointer may be partially swapped, and processes on/after pointer have no core item (non-INDIRECT_CORE)
Nick Downing [Sun, 2 Jun 2019 03:56:16 +0000 (13:56 +1000)]
Make PROCESS_FLAGS_CORE_ITEM only maintained for non-INDIRECT_CORE (moveable)
Nick Downing [Sun, 2 Jun 2019 03:39:27 +0000 (13:39 +1000)]
Make victim be LRU not process pointer, and use &lru_head instead of NULL
Nick Downing [Sun, 2 Jun 2019 03:27:36 +0000 (13:27 +1000)]
Get rid of runner pointer, was only needed for process_get_state() during move
Nick Downing [Sun, 2 Jun 2019 03:15:57 +0000 (13:15 +1000)]
Simplify process_free() and process_get_state() so that we rely completely on process->swap_blocks to distinguish between fully in core, runner, victim, etc
Nick Downing [Sun, 2 Jun 2019 02:30:47 +0000 (12:30 +1000)]
Cosmetic issue in a diagnostic printf()
Nick Downing [Sun, 2 Jun 2019 02:29:57 +0000 (12:29 +1000)]
For INDIRECT_CORE make process_realloc() free the physical blocks before resize
Nick Downing [Sun, 2 Jun 2019 02:21:22 +0000 (12:21 +1000)]
For INDIRECT_CORE defer swap out until after successful table memory allocation
Nick Downing [Sun, 2 Jun 2019 02:17:06 +0000 (12:17 +1000)]
Implement the core_move_up() routine for INDIRECT_CORE, needed for realloc_base
Nick Downing [Sun, 2 Jun 2019 01:55:28 +0000 (11:55 +1000)]
Tidy up process_alloc() / process_realloc() allocating/freeing physical blocks
Nick Downing [Sun, 2 Jun 2019 01:05:21 +0000 (11:05 +1000)]
Add process flags to track whether the process and/or the core_item are valid
Nick Downing [Sun, 2 Jun 2019 00:33:49 +0000 (10:33 +1000)]
Add per-process tracking of amount swapped (formerly was in victim_swap_blocks)
Nick Downing [Sun, 2 Jun 2019 01:03:46 +0000 (11:03 +1000)]
Update generated subdirectories to do the process->para and realloc_base tests
Nick Downing [Sun, 2 Jun 2019 00:19:44 +0000 (10:19 +1000)]
Fix a bug with using local copy of runner_swap_blocks not the global variable
Nick Downing [Sat, 1 Jun 2019 11:14:20 +0000 (21:14 +1000)]
Implement process->para so that start of data is independent of block boundary, implement realloc_base to extend start via direction flag for process_realloc()
Nick Downing [Sat, 1 Jun 2019 10:44:05 +0000 (20:44 +1000)]
Distinguish between swap reservation and occupation in "new core ..." printouts
Nick Downing [Sat, 1 Jun 2019 05:07:40 +0000 (15:07 +1000)]
Tidy up the actual_old_paras/actual_old_blocks stuff and make the test script include all the information used to calculate success values (for added checking)
Nick Downing [Sat, 1 Jun 2019 02:05:26 +0000 (12:05 +1000)]
Change all byte sizes to paras, BLOCK_SIZE to BLOCK_PARAS and so on, make core_block_mem/swap_block_mem use one int per simulated para not one uint8_t per byte
Nick Downing [Sat, 1 Jun 2019 00:21:47 +0000 (10:21 +1000)]
Simplify hashing function to eliminate wraparound carry and use of long long
Nick Downing [Fri, 31 May 2019 15:19:25 +0000 (01:19 +1000)]
Fix some defines and conditional compiles that got lost in the ifdef-generation
Nick Downing [Fri, 31 May 2019 15:05:56 +0000 (01:05 +1000)]
Turn things around so that swap is written from top down
Nick Downing [Fri, 31 May 2019 13:28:28 +0000 (23:28 +1000)]
Remove first/last transfer optimization, it will be done differently
Nick Downing [Fri, 31 May 2019 13:17:23 +0000 (23:17 +1000)]
Make victim_core_blocks be calculated on the fly, keep only victim_swap_blocks
Nick Downing [Fri, 31 May 2019 13:13:10 +0000 (23:13 +1000)]
Re-enable all diagnostic printfs
Nick Downing [Wed, 29 May 2019 13:32:00 +0000 (23:32 +1000)]
Remove support for INODE_SWAP, make PREALLOCATE_SWAP always, make PREALLOCATE_CORE when INDIRECT_CORE (have done this automatically, adds diff --ifdef scripts)
Nick Downing [Wed, 29 May 2019 13:05:36 +0000 (23:05 +1000)]
Add forgotten file
Nick Downing [Sat, 18 May 2019 02:34:56 +0000 (12:34 +1000)]
Implement binary format script (for use within kernel, where there's no scanf)
Nick Downing [Sat, 18 May 2019 01:23:12 +0000 (11:23 +1000)]
More printf and unnecessary header removal, improve preprocessed blank lines
Nick Downing [Sat, 18 May 2019 00:17:59 +0000 (10:17 +1000)]
Implement FUZIX build, reduces block sizes, simplifies hash function, fixes rassert issues
Nick Downing [Sat, 18 May 2019 00:14:01 +0000 (10:14 +1000)]
Remove all non-essential printfs (for code size reduction in FUZIX build)
Nick Downing [Thu, 9 May 2019 12:06:06 +0000 (22:06 +1000)]
Separate out block allocation bitmap stuff into a reuseable module block_pool.c
Nick Downing [Sun, 5 May 2019 11:55:47 +0000 (21:55 +1000)]
Rename pool_realloc() to pool_alloc() with a mode argument instead of ad-hoc schemes like sending an offset of INT_MIN or not filling in the move_up() routine
Nick Downing [Sun, 5 May 2019 03:31:21 +0000 (13:31 +1000)]
New combined routine that does all of pool_alloc(), pool_realloc(), pool_realloc_base() and does the best it can with whatever move()/move_up() routines found
Nick Downing [Mon, 22 Apr 2019 07:22:10 +0000 (17:22 +1000)]
Combine non-moveable with moveable pool, define both head->move and head->move_up to enable moveability, or just head->move for ordinary malloc-like behaviour
Nick Downing [Mon, 22 Apr 2019 06:23:04 +0000 (16:23 +1000)]
Back out change from commit
f4338d0 to process_avail, as process_avail was already defined to include the indirect blocks so no need to estimate their removal
Nick Downing [Mon, 22 Apr 2019 05:26:09 +0000 (15:26 +1000)]
Move swap space tests into do_swap_out(), re-evaluate for each victim to be less conservative in the inode swap case (allow only 2 blocks for partial indirect)
Nick Downing [Sun, 21 Apr 2019 11:38:36 +0000 (21:38 +1000)]
Do proper computations about disk free space including possible direct and indirect blocks (should revisit to be less conservative in transfer sizes by estimating each victim separately so we don't need to allow for so many partial direct/indirect blocks upfront), get NDEBUG working again (doesn't fill core with 0xaa)
Nick Downing [Sun, 21 Apr 2019 05:38:47 +0000 (15:38 +1000)]
More strictly declare variables at start of function, for ack
Nick Downing [Thu, 11 Apr 2019 13:26:26 +0000 (23:26 +1000)]
Fix inode leak caused by the f_trunc() modifications
Nick Downing [Thu, 11 Apr 2019 12:41:26 +0000 (22:41 +1000)]
Optimize out the classic or moveable allocator when not needed
Nick Downing [Thu, 11 Apr 2019 12:00:05 +0000 (22:00 +1000)]
Fix several more minor bugs, add gen.sh which pre-runs the C preprocessor to generate 6 specialized versions of the code
Nick Downing [Thu, 11 Apr 2019 11:12:55 +0000 (21:12 +1000)]
Fix INDIRECT_CORE with INODE_SWAP combination
Nick Downing [Thu, 11 Apr 2019 10:25:57 +0000 (20:25 +1000)]
Commit a forgotten file
Nick Downing [Tue, 9 Apr 2019 14:10:56 +0000 (00:10 +1000)]
First cut at inode swapping, works, may have an inode leak somewhere
Nick Downing [Tue, 9 Apr 2019 10:13:52 +0000 (20:13 +1000)]
Expand out the swap_read()/swap_write() functions inline, since they were only called once, and they were supposed to be abstract but now the abstract part is done by core_to_swap_copy()/swap_to_core_copy() and the rest is well defined -- makes core.h/swap.h and core.c/swap.c identical again except for core<->swap
Nick Downing [Sun, 7 Apr 2019 06:12:47 +0000 (16:12 +1000)]
Change order of adjusting process_core_blocks/process_swap_blocks/victim_core_blocks/victim_swap_blocks to reduce computation
Nick Downing [Sun, 7 Apr 2019 05:27:35 +0000 (15:27 +1000)]
Move PREALLOCATE_CORE/PREALLOCATE_SWAP defines into the respective header files and define victim_core_size/victim_swap_size there, remove old victim_xxx_item
Nick Downing [Sat, 6 Apr 2019 03:31:15 +0000 (14:31 +1100)]
More optimal handling of incomplete last block in swap_read()/swap_write()
Nick Downing [Sat, 6 Apr 2019 03:17:07 +0000 (14:17 +1100)]
Fix a bug in undo successful process realloc, fix a bug in number of blocks to transfer for swap_read() and swap_write(), fix minor inconsistencies
Nick Downing [Fri, 5 Apr 2019 23:11:14 +0000 (10:11 +1100)]
Move core_block_mem and swap_block_mem into pool_test_run.c and make core/swap memory accessor functions abstract, make core/swap move do incomplete last block
Nick Downing [Sat, 6 Apr 2019 02:47:04 +0000 (13:47 +1100)]
Get rid of the DEDICATED_xxx() macros in favour of more optimal inline code
Nick Downing [Fri, 5 Apr 2019 11:34:32 +0000 (22:34 +1100)]
Get rid of the dedicated and per-process core/swap item, just use per-process
Nick Downing [Thu, 4 Apr 2019 11:20:43 +0000 (22:20 +1100)]
Implement partial transfer to/from swap for incomplete last block
Nick Downing [Thu, 4 Apr 2019 10:20:03 +0000 (21:20 +1100)]
Make swap reading/writing take bytes not blocks, allowing incomplete last block
Nick Downing [Thu, 4 Apr 2019 10:01:31 +0000 (21:01 +1100)]
Make only the active (bytewise not blockwise) portion of process be hashed
Nick Downing [Thu, 4 Apr 2019 09:49:17 +0000 (20:49 +1100)]
Make process hash init and verify be bytewise rather than blockwise
Nick Downing [Thu, 4 Apr 2019 09:21:18 +0000 (20:21 +1100)]
Simplify the calling convention of the hash init and verify routines
Nick Downing [Thu, 4 Apr 2019 09:12:30 +0000 (20:12 +1100)]
Move the hash generation and verification stuff into process_test_run.c
Nick Downing [Thu, 4 Apr 2019 09:07:11 +0000 (20:07 +1100)]
Change process.blocks to process.size, do rounding and shifting at each access
Nick Downing [Thu, 4 Apr 2019 08:05:40 +0000 (19:05 +1100)]
Change test script to refer to bytes, although everything still done in blocks
Nick Downing [Thu, 4 Apr 2019 07:46:28 +0000 (18:46 +1100)]
In process related stuff, change size to blocks and SIZE to BLOCKS
Nick Downing [Sun, 31 Mar 2019 10:17:28 +0000 (21:17 +1100)]
Implement further macros DEDICATED_CORE_BASE() and DEDICATED_SWAP_LIMIT() which know whether the underlying items get updated during swap or whether they are preallocated at process creation time, and take a correction factor to apply in the latter case (by macro, so the correction is not evaluated in the former case)
Nick Downing [Sun, 31 Mar 2019 07:43:24 +0000 (18:43 +1100)]
Implement higher level macros to access process->core_item vs process->pool_item, victim->core_item vs victim_core_item and so on (depends on configuration)
Nick Downing [Sun, 31 Mar 2019 07:45:24 +0000 (18:45 +1100)]
Remove some bogus assertions that don't work in preallocated modes, tidy up
Nick Downing [Sun, 31 Mar 2019 02:40:20 +0000 (13:40 +1100)]
Change direction of swap in/out to forward, so that core is read/written in deque fashion (when swapping out, pop off the base, when swapping in, push onto the base) whereas swap is read/written in file fashion (when swapping out, append to the limit, when swapping in, truncate the limit), allows a swap-to-inode mode
Nick Downing [Sat, 30 Mar 2019 12:39:13 +0000 (23:39 +1100)]
Add pool_realloc_base() and pool_realloc_base_moveable() to allow a region to be used like a deque, new space can be pushed onto and contents popped off start
Nick Downing [Thu, 28 Mar 2019 12:30:27 +0000 (23:30 +1100)]
Add inode test script
Nick Downing [Sun, 24 Mar 2019 02:38:30 +0000 (13:38 +1100)]
Add UCP from FUZIX, with modifications to isolate FS stuff into new fuzix_fs.c
Nick Downing [Sat, 23 Mar 2019 22:55:43 +0000 (09:55 +1100)]
Implement simulated BLOCK_SIZE, hash create and check simulated block contents
Nick Downing [Sat, 23 Mar 2019 21:46:23 +0000 (08:46 +1100)]
Make n_blocks the primary size and table_size only given if indirect (was the other way around, making block stuff use either table_size or n_blocks), fix bugs
Nick Downing [Sat, 23 Mar 2019 06:22:24 +0000 (17:22 +1100)]
Subroutinize the hashing initialization and verification, to avoid duplication
Nick Downing [Sat, 23 Mar 2019 05:50:59 +0000 (16:50 +1100)]
Print message like 'alloc NN NN true|false' and then perform the operation and then '... ok', so that we can see clearly what messages come from what operation
Nick Downing [Sat, 23 Mar 2019 04:44:49 +0000 (15:44 +1100)]
Remove the reservation of spare space from pool object, core and swap modules
Nick Downing [Tue, 19 Mar 2019 13:13:23 +0000 (00:13 +1100)]
Implement INDIRECT_CORE, fix calculations so that PREALLOCATE_CORE make sense
Nick Downing [Tue, 19 Mar 2019 10:36:45 +0000 (21:36 +1100)]
For direct swap, store data in swap_block_mem not swap_table_mem, also for core
Nick Downing [Tue, 19 Mar 2019 10:11:51 +0000 (21:11 +1100)]
Add INDIRECT_SWAP option
Nick Downing [Mon, 18 Mar 2019 10:24:39 +0000 (21:24 +1100)]
Insert table_ in most core.c / swap.c functions e.g. core_alloc() becomes core_table_alloc(), because there will soon be an allocator for random block storage
Nick Downing [Sun, 17 Mar 2019 03:23:55 +0000 (14:23 +1100)]
Implement MOVEABLE_CORE and MOVEABLE_SWAP compile time options
Nick Downing [Sun, 17 Mar 2019 03:16:54 +0000 (14:16 +1100)]
Fix a bug of requiring a certain amount of free swap, when swap is preallocated
Nick Downing [Sun, 17 Mar 2019 03:00:27 +0000 (14:00 +1100)]
Fix a bug of victim_core_size being used uninitialized without PREALLOCATE_CORE
Nick Downing [Sun, 17 Mar 2019 02:24:46 +0000 (13:24 +1100)]
Implement PREALLOCATE_CORE and PREALLOCATE_SWAP compile time options
Nick Downing [Sat, 16 Mar 2019 22:27:26 +0000 (09:27 +1100)]
Improve use of alloc()/realloc()/free() to remove allocs and reallocs of size 0
Nick Downing [Sat, 16 Mar 2019 14:03:15 +0000 (01:03 +1100)]
Tidy up debugging code
Nick Downing [Sat, 16 Mar 2019 13:58:21 +0000 (00:58 +1100)]
Replace core_item and swap_item in the process structure with just pool_item
Nick Downing [Sat, 16 Mar 2019 13:47:56 +0000 (00:47 +1100)]
Keep process and victim pool items in dedicated storage during swapping (should generate smaller code, also ensures a process has only one pool item at a time)