Brian Norris [Fri, 27 Apr 2012 06:42:55 +0000 (23:42 -0700)]
model: backtracking messages are only for debugging
Brian Norris [Fri, 27 Apr 2012 06:42:12 +0000 (23:42 -0700)]
common: add simple DBG_ENABLED() macros
Brian Norris [Fri, 27 Apr 2012 06:36:26 +0000 (23:36 -0700)]
thread: remove dead Thread code
Brian Norris [Fri, 27 Apr 2012 06:35:44 +0000 (23:35 -0700)]
model: fixup thread ID selection
Brian Norris [Thu, 26 Apr 2012 23:52:20 +0000 (16:52 -0700)]
demote 'system_thread' to just 'system_context'
I don't need the extra 'class Thread' metadata for the ModelChecker 'system
thread,' and the dual use of class Thread for both system and user threads was
causing problems. So I separate them: class Thread is used only for user
threads. This changes a few interfaces throughout the code.
Brian Norris [Thu, 26 Apr 2012 23:16:23 +0000 (16:16 -0700)]
libatomic: add atomic_init()
Brian Norris [Thu, 26 Apr 2012 23:15:31 +0000 (16:15 -0700)]
userprog: print atomic load/store values
Brian Norris [Thu, 26 Apr 2012 23:14:27 +0000 (16:14 -0700)]
libatomic: use 'values' for atomic load/store
Brian Norris [Thu, 26 Apr 2012 19:05:08 +0000 (12:05 -0700)]
model: bugfix - detect conflicts properly
Finding a conflict from the same thread does not mean we should quit the search
with no found conflict; we just continue to the next search item.
Brian Norris [Thu, 26 Apr 2012 18:47:22 +0000 (11:47 -0700)]
libatomic: add DBG() calls
Brian Norris [Thu, 26 Apr 2012 18:38:36 +0000 (11:38 -0700)]
model: bugfix - reset the "current_action" after it has been processed
There are executions under which we may call ModelChecker::check_current_action()
twice in a row without an acutal ModelAction (curren_action) being set in
between. This causes the previous action to be queued twice in our trace
listing.
This can be solved 2 ways:
(1) set current_action to NULL after processing it
(2) move all calls to check_current_action() under the THREAD_READY codepath,
so that we guarantee that some Thread pushed an Action back to the
model-checking system
I do both.
Brian Norris [Thu, 26 Apr 2012 17:19:12 +0000 (10:19 -0700)]
librace: format DEBUG() prints properly
Apparently ISO C99 defines macros PRIuXX for proper printing of different-sized
uints.
Brian Norris [Wed, 25 Apr 2012 23:55:46 +0000 (16:55 -0700)]
librace: add load/store interface for checking data races
The compiler portion will need to replace loads/stores with calls to the
store_X() and load_X() functions.
Brian Norris [Tue, 24 Apr 2012 20:44:34 +0000 (13:44 -0700)]
model: remove unused definition
Brian Norris [Tue, 24 Apr 2012 20:38:37 +0000 (13:38 -0700)]
threads: allocate Threads in via userMalloc()
Thread state information should be held in the snapshotting region (when
implemented), so override new and delete.
Brian Norris [Tue, 24 Apr 2012 20:22:38 +0000 (13:22 -0700)]
common: introduce userMalloc() and userFree()
These are stubs for now, but I will need these.
Brian Norris [Tue, 24 Apr 2012 20:07:37 +0000 (13:07 -0700)]
libthreads: remove unnecessary 'extern'
Brian Norris [Tue, 24 Apr 2012 20:03:43 +0000 (13:03 -0700)]
run.sh: add simple 'run' script
Performs the step of setting LD_LIBRARY_PATH for simple runs.
Brian Norris [Tue, 24 Apr 2012 20:03:23 +0000 (13:03 -0700)]
userprog: use typedef'd thrd_start_t
Brian Norris [Tue, 24 Apr 2012 20:01:13 +0000 (13:01 -0700)]
.gitignore: add *.so
Brian Norris [Tue, 24 Apr 2012 20:00:35 +0000 (13:00 -0700)]
Makefile: build model checker as shared library (libmodel.so)
Brian Norris [Tue, 24 Apr 2012 19:06:21 +0000 (12:06 -0700)]
lib*.h: wrap C headers in `extern "C"'
Brian Norris [Tue, 24 Apr 2012 06:31:52 +0000 (23:31 -0700)]
libthreads: don't create ModelAction for thrd_join()
Temporarily? I may need to model-check join operations later.
Brian Norris [Tue, 24 Apr 2012 04:18:06 +0000 (21:18 -0700)]
model: rename print_trace() -> print_summary(), fixup summary printing
Brian Norris [Tue, 24 Apr 2012 03:39:01 +0000 (20:39 -0700)]
schedule: only print when DEBUG is enabled
Brian Norris [Tue, 24 Apr 2012 03:18:44 +0000 (20:18 -0700)]
tmp (model)
Brian Norris [Tue, 24 Apr 2012 03:16:05 +0000 (20:16 -0700)]
threads: cleanup inconsistencies in memory management
Brian Norris [Tue, 24 Apr 2012 03:13:02 +0000 (20:13 -0700)]
threads: set up Thread to be freed properly
To ensure that we can free memory properly, set up a destructor and rename
dispose() to complete().
Brian Norris [Tue, 24 Apr 2012 03:09:45 +0000 (20:09 -0700)]
schedule: reset scheduler when thread is removed
Ensure that the 'current' thread reference is removed properly.
Brian Norris [Tue, 24 Apr 2012 02:46:30 +0000 (19:46 -0700)]
threads: don't make direct call into scheduler
Since I need a ModelChecker::add_thread(Thread *) function anyway, I should
direct add_thread() calls through the model-checker framework rather than the
Scheduler class. This will make things a little cleaner and symmetric when I
add a remove_thread() function.
Brian Norris [Tue, 24 Apr 2012 02:22:52 +0000 (19:22 -0700)]
threads: save id within class Thread
There may be points at which the user's thread ID struct goes out of scope, but
I still need it internally. I think I'll just keep it internally, with a
duplicate in the user's copy.
Brian Norris [Tue, 24 Apr 2012 00:43:42 +0000 (17:43 -0700)]
threads: initialized Thread member variables
Brian Norris [Tue, 24 Apr 2012 00:34:06 +0000 (17:34 -0700)]
unify style for returning pointers
When returning a pointer from a function, always put a space between the type,
the '*', and the function name. E.g.:
type * func(args);
Brian Norris [Tue, 24 Apr 2012 00:14:45 +0000 (17:14 -0700)]
model: print 'number of executions'
In preparation of many executions, I establish a counting variable
'num_executions'.
Brian Norris [Tue, 24 Apr 2012 00:12:01 +0000 (17:12 -0700)]
threads: prepare system to loop over many executions
Brian Norris [Mon, 23 Apr 2012 23:00:06 +0000 (16:00 -0700)]
improve scheduler debugging
Brian Norris [Mon, 23 Apr 2012 22:40:42 +0000 (15:40 -0700)]
schedule: bugfix - set 'current' thread in all cases
When the scheduler was following a model-checking replay execution, it did not
set the 'current' thread properly.
Brian Norris [Mon, 23 Apr 2012 22:39:07 +0000 (15:39 -0700)]
schedule: refactor next_thread() for better debug printing
It makes more sense to call print() at the end of the the next_thread()
routine, to see consistent printing for all cases. So I refactor the control
flow to a single return statement.
Brian Norris [Mon, 23 Apr 2012 06:14:40 +0000 (23:14 -0700)]
schedule: print debug info
Brian Norris [Fri, 20 Apr 2012 17:59:02 +0000 (10:59 -0700)]
model: add simple comment
Brian Norris [Fri, 20 Apr 2012 17:58:40 +0000 (10:58 -0700)]
model: move public functions to private
Brian Norris [Fri, 20 Apr 2012 17:55:44 +0000 (10:55 -0700)]
model: implement get_next_replay() and advance_backtracking_state()
Brian Norris [Fri, 20 Apr 2012 17:50:29 +0000 (10:50 -0700)]
model: implement, use schedule_next_thread()
Relies on the rest of the model checker to set the next thread; simply returns
the chosen thread.
Brian Norris [Fri, 20 Apr 2012 17:47:10 +0000 (10:47 -0700)]
model: add prototypes to header
I will implement these functions in the next few commits.
Brian Norris [Fri, 20 Apr 2012 17:35:16 +0000 (10:35 -0700)]
model: add iteration routines for class Backtrack
Brian Norris [Fri, 20 Apr 2012 17:31:42 +0000 (10:31 -0700)]
model: include <cstddef>
NULL is not a builtin keyword in recent C++ revisions. I have to include
<cstddef>.
Brian Norris [Fri, 20 Apr 2012 17:30:58 +0000 (10:30 -0700)]
schedule: replace queue with list
I'll need random access to its elements later.
Brian Norris [Fri, 20 Apr 2012 06:28:41 +0000 (23:28 -0700)]
threads: introduce THREAD_ID_T_NONE
Represents a kind of 'NULL' thread, but for integer IDs.
Brian Norris [Fri, 20 Apr 2012 06:26:33 +0000 (23:26 -0700)]
threads: remove leftover class 'prototype'
Brian Norris [Thu, 19 Apr 2012 20:21:02 +0000 (13:21 -0700)]
model: stash each backtrack event in ModelChecker::backtrack_list
Brian Norris [Thu, 19 Apr 2012 20:20:47 +0000 (13:20 -0700)]
model: add class Backtrack
Brian Norris [Thu, 19 Apr 2012 20:00:26 +0000 (13:00 -0700)]
model: create 'action_list_t' typedef
'std::list<class ModelAction *>' is a long name for a repeated type. Replace
it.
Brian Norris [Thu, 19 Apr 2012 19:56:21 +0000 (12:56 -0700)]
model: convert 'action_trace' to pointer
We need to swap out this list with a new one when we reset, and we want to
retain references to this list easily, so make the ModelChecker::action_trace
member into a pointer to a list.
Brian Norris [Thu, 19 Apr 2012 18:31:41 +0000 (11:31 -0700)]
model: add TreeNode debugging information to print_trace()
Brian Norris [Thu, 19 Apr 2012 18:06:30 +0000 (11:06 -0700)]
model: implement get_last_conflict()
This performs a non-optimized check; that is, it uses a simple O(n) linear
search, when it could utilize O(1) lookups instead... but this isn't even the
full memory model yet, so I'm not working for perfection!
Brian Norris [Thu, 19 Apr 2012 18:02:48 +0000 (11:02 -0700)]
model: flesh out set_backtracking()
Relies on stub get_last_conflict() function.
Note: so far, implementing based on seq_cst model, not the full memory model...
Brian Norris [Thu, 19 Apr 2012 18:00:55 +0000 (11:00 -0700)]
model: flesh out check_current_action()
Relies on a stub set_backtracking() function.
Brian Norris [Thu, 19 Apr 2012 17:58:31 +0000 (10:58 -0700)]
model: add accessors for ModelAction variables
(+ add 'node' member, pointing to the state-space tree just before the action)
Brian Norris [Thu, 19 Apr 2012 17:47:11 +0000 (10:47 -0700)]
tree: bugfix - fix backtrack status
Brian Norris [Thu, 19 Apr 2012 07:10:17 +0000 (00:10 -0700)]
tree: don't use 'NULL' for a thread ID
Brian Norris [Wed, 18 Apr 2012 19:36:58 +0000 (12:36 -0700)]
libthreads: only record 'final' even for thrd_join
Brian Norris [Wed, 18 Apr 2012 01:23:25 +0000 (18:23 -0700)]
model: use TreeNode()
Very basic initialization...
Brian Norris [Wed, 18 Apr 2012 01:23:09 +0000 (18:23 -0700)]
tree: fix invalid free
Brian Norris [Wed, 18 Apr 2012 00:50:29 +0000 (17:50 -0700)]
tree: add class TreeNode
Untested, but ported from Nachos project.
Brian Norris [Tue, 17 Apr 2012 23:40:23 +0000 (16:40 -0700)]
rename threads_internal.h -> threads.h
Brian Norris [Tue, 17 Apr 2012 22:46:11 +0000 (15:46 -0700)]
threads/model: move switch_to_master from class Thread to class ModelChecker
Brian Norris [Mon, 16 Apr 2012 06:20:44 +0000 (23:20 -0700)]
rename binary: libthreads -> model
Brian Norris [Mon, 16 Apr 2012 06:15:24 +0000 (23:15 -0700)]
libthreads: split into libthreads and threads
For user-program vs. internal usage
Brian Norris [Mon, 16 Apr 2012 06:01:24 +0000 (23:01 -0700)]
major rewrite - 'struct thread' replaced with internal 'class Thread'
Rewrite code such that most thread data is kept internally, not within the user
program.
Brian Norris [Mon, 16 Apr 2012 05:32:55 +0000 (22:32 -0700)]
Makefile: add debug compilation flag
Brian Norris [Tue, 10 Apr 2012 23:26:26 +0000 (16:26 -0700)]
scheduler: kill 'replaceable' scheduler
I don't actually need a replaceable scheduler; I need a scheduler that can
switch policies at will. Just remove this level of abstraction for now (it can
be added later if needed).
Brian Norris [Tue, 10 Apr 2012 22:51:33 +0000 (15:51 -0700)]
libthreads: print out trace at end of execution
Brian Norris [Tue, 10 Apr 2012 22:51:05 +0000 (15:51 -0700)]
model: add print_trace() function
Prints out the current program trace.
Brian Norris [Tue, 10 Apr 2012 22:25:34 +0000 (15:25 -0700)]
libthreads: perform 'model checking' when moving to next thread
Simply use the hollow 'check_current_action()' interface to perform model
checking in the thread_system_next() call.
Right now, this just adds the previous (a.k.a. 'current') action to our log.
Brian Norris [Tue, 10 Apr 2012 22:19:29 +0000 (15:19 -0700)]
userprog: use both atomic loads and stores
(Meaningless change to 'user' program; just for testing purposes)
Brian Norris [Tue, 10 Apr 2012 22:14:45 +0000 (15:14 -0700)]
model: add check_current_action() function
This interface provides a wrapper for all our model checking operations on the
current action (at least, ideally this should work...I might need to work on
this a bit).
For now, we just record the program trace of all atomic/thread actions.
Brian Norris [Tue, 10 Apr 2012 22:08:23 +0000 (15:08 -0700)]
libthreads: thread_join: return 'int' as status
Brian Norris [Tue, 10 Apr 2012 22:00:13 +0000 (15:00 -0700)]
threads_internal: pass the current 'action' to the internal thread system
I build a ModelAction object to represent the properties of the current action
that are relevant to the model checker. This action should be passed from the
basic action (atomic_load, atomic_store, thread_yield, etc.) into our thread
library and eventually to the model checker.
I don't use these actions yet, but they will be shortly.
Brian Norris [Tue, 10 Apr 2012 21:51:30 +0000 (14:51 -0700)]
model: add class ModelAction
Represents a single action (sometimes called transition) in our program. This
can be a THREAD_* event (creation, join, etc.) or a relevant read/write
operation (e.g., atomic reads/writes). This object and interface is subject to
change a little...
Brian Norris [Tue, 10 Apr 2012 21:09:42 +0000 (14:09 -0700)]
threads_internal: add 'thread_switch_to_master()' internally
In progress: moving more model-checking/thread actions so that they are
performed only by a system thread, not by the user code.
For now, we replace our uses of thread_yield() (which simply switched to the
next user thread) while thread_switch_to_master() (which will later perform
intermediate actions before choosing the next user thread).
Also, we can begin creating internal and external interfaces in
threads_indernal.h and libthreads.h, respectively.
Brian Norris [Thu, 15 Mar 2012 22:42:36 +0000 (15:42 -0700)]
libthreads: make typedef for 'thread_id_t'
Just in case we need to change it later on, let's use it as a named type.
Brian Norris [Thu, 5 Apr 2012 03:14:27 +0000 (20:14 -0700)]
use my{Malloc,Free} in model-checking code
Also, fixes 'leak' where we forgot to free the main_thread.
Brian Norris [Thu, 5 Apr 2012 02:41:46 +0000 (19:41 -0700)]
malloc: override 'new' and 'delete' globally
Not sure if this is correct...
Q: do we also need to overload new[] and delete[]?
Brian Norris [Thu, 5 Apr 2012 02:35:41 +0000 (19:35 -0700)]
malloc: add myMalloc() and myFree()
Brian Norris [Thu, 5 Apr 2012 02:12:19 +0000 (19:12 -0700)]
common: rename "CONFIG" to "COMMON"
Brian Norris [Thu, 15 Mar 2012 18:12:44 +0000 (11:12 -0700)]
schedule: use STL 'queue' instead of 'list'
Brian Norris [Thu, 15 Mar 2012 06:01:27 +0000 (23:01 -0700)]
remove unnecessary #includes
Brian Norris [Thu, 15 Mar 2012 05:56:30 +0000 (22:56 -0700)]
schedule: make 'current' a private member of the scheduler
Brian Norris [Thu, 15 Mar 2012 05:52:42 +0000 (22:52 -0700)]
schedule: use STL list class instead of custom queue
Brian Norris [Wed, 14 Mar 2012 22:59:49 +0000 (15:59 -0700)]
rename other *.c to *.cc
userprog.c remains for now...
Brian Norris [Wed, 14 Mar 2012 22:56:56 +0000 (15:56 -0700)]
schedule: move schedule.c --> schedule.cc
Brian Norris [Wed, 14 Mar 2012 22:54:36 +0000 (15:54 -0700)]
schedule: create 'class Scheduler' with implementation 'class DefaultScheduler'
Brian Norris [Wed, 14 Mar 2012 22:10:31 +0000 (15:10 -0700)]
libthreads: delete allocated object
Brian Norris [Wed, 14 Mar 2012 22:08:25 +0000 (15:08 -0700)]
model: use 'this' uniformly
Brian Norris [Wed, 14 Mar 2012 22:07:04 +0000 (15:07 -0700)]
model: move model.c --> model.cc
Brian Norris [Wed, 14 Mar 2012 22:05:54 +0000 (15:05 -0700)]
model: change 'struct model_checker' to 'class ModelChecker'
Brian Norris [Wed, 14 Mar 2012 22:00:46 +0000 (15:00 -0700)]
Makefile: switch to C++ builds
Brian Norris [Tue, 13 Mar 2012 18:37:45 +0000 (11:37 -0700)]
stricter typing of function pointers for makecontext()
We need to use functions with no arguments (i.e., 'void (*)()') in order to
retain strict type-checking with makecontext(). Of course, we still circumvent
this type checking with casting, but we should straighten this out sometime...
Brian Norris [Tue, 13 Mar 2012 18:35:55 +0000 (11:35 -0700)]
C++: don't use C++ keywords as names (this, new, etc.)
Brian Norris [Tue, 13 Mar 2012 18:35:09 +0000 (11:35 -0700)]
C++: cast result of malloc