Commit Graph

419 Commits

Author SHA1 Message Date
c13d57f0f6 Fix for nOS-V with ovni.level=2
When running without all the subsystem events enabled in nOS-V, it is
possible to emit two consecutive VTx events, which would push twice the
same value ST_TASK_BODY into the subsystem channel. This change relaxes
the subsystem channel to accept duplicate stacked values. A regression
test is also added.

Reported-By: Raúl Peñacoba Veigas <raul.penacoba@bsc.es>
2023-11-14 17:26:43 +01:00
4b4f1bd218 Don't modify nOS-V subsystem state on task pause
In nOS-V, when a task was paused via the VTp event, two things were
happening: 1) the task state was set to pause and 2) the subsystem state
"Task: Running" was being popped.

This causes a problem when a task calls nosv_submit() in blocking mode,
as it will call nosv_pause() which will emit a VTp event from a
subsystem different than "Task: Running".

To solve this conflict, we handle the subsystems state and the task
state separately with the VTp and VTr events. The subsystem state "Task:
Running" no longer is connected to the state of the task and only shows
if we entered the body of the task or not. It has now been renamed to
"Task: In body".

The new state "Task: In body" represents that the task body has begun
the execution and is still in the stack, but the task may be paused. The
subsystem is not changed by the VTp (pause) or VTr (resume) events.

Fixes: https://pm.bsc.es/gitlab/rarias/ovni/-/issues/128
2023-10-23 11:26:17 +02:00
0df018cf5f Fix segfault reading empty streams
The check_clock_gate() was trying to access to the first event in all
streams. However, streams that are empty don't have any event, so they
cause a NULL dereference. Skipping not active streams avoids the
problem.
2023-10-23 10:07:44 +02:00
Kevin Sala
7b4ebca002 Add OVNI_TRACEDIR envar to change trace directory
This commit introduces the OVNI_TRACEDIR environment variable to change
the directory where traces are generated. By default, when the envar is
not defined, the trace is still generated in the ovni directory. The
envar can take a trace directory name, a relative path to the directory,
or its absolute path. In the first case, the directory is created in the
current path $PWD.

Both libovni (rt) and ovnisync read this environment variable.
2023-10-23 07:49:28 +00:00
Kevin Sala
24252ebbb3 Use common's progname functions in ovnisync
This commit adds the progname_get function in common.h header and
extends the ovnisync tool to use them
2023-10-23 07:49:28 +00:00
Kevin Sala
9a628e78ca Add MPI model with interface view 2023-09-06 18:11:18 +02:00
Kevin Sala
276afd5479 Add TAMPI model with subsystems view 2023-08-22 18:13:29 +02:00
Kevin Sala
9269dd7202 Add nosv create and destroy API subsystem states 2023-08-17 13:23:59 +02:00
a3ffb2443a Report the commit too with ovniver 2023-07-26 17:13:28 +02:00
bf8a2213c1 Add the ovniver program to report the version 2023-07-26 16:31:56 +02:00
97c936d41f Add ovni_version_get() function
Allows the reding the version of the libovni.so currently loaded.
2023-07-26 16:31:30 +02:00
f08595b3a7 Report open error for missing trace directory 2023-07-26 13:51:35 +02:00
f8c6b7c7c0 Improve the version error message
Use the terminology want/got to determine which version the user is
requested and the version the library provides.
2023-07-25 19:55:51 +02:00
48df02f102 Always use the project version
Don't duplicated versions so we avoid the problem of missing one
instance to be updated.
2023-07-07 10:42:40 +02:00
ee9111a1ec Move ovni.pc.in into src/ 2023-07-05 13:13:53 +02:00
4abd92e139 Make ovni.h a public header of ovni target 2023-07-05 13:11:18 +02:00
cd2e844066 Use the installation directories provided by cmake
Some systems will define the library path as lib64. Also update other
install commands to use the values discovered by the GNUInstallDirs
module.
2023-07-05 12:52:56 +02:00
5a9086e6d1 Fix ovnisort with flush events
Sort the events in the sorting region before injecting them back in the
stream. This solves the problem with flush events OF[ and OF].
2023-06-29 13:25:15 +02:00
05979df34d Install ovnitop.1 into man1 subdirectory 2023-06-02 17:12:17 +02:00
d9b82494d0 Free the table at the end in ovnitop 2023-06-01 18:38:26 +02:00
c1b0768e7d Add ovnitop(1) manual page 2023-05-31 17:23:25 +00:00
553d0a1c62 Add ovnitop command 2023-05-31 17:23:25 +00:00
44d120d7a8 Version 1.2.0 2023-05-01 09:19:05 +02:00
c2b6bfd3a8 Set the size of the channel property arrays
Arrays which don't initialize the last elements of the enumeration are
shorter and will cause a buffer overflow when read in a loop.
2023-04-24 12:37:20 +02:00
a8944b3763 Fix buffer overflow in sort_replace()
The access to the array was done before the out of bound check.
2023-04-24 11:47:46 +02:00
49f915fda5 Add FIXME notice for taskiter support
Set the task->thread to NULL if we re-execute a task from the Dead
state, so it follows the behavior of a Created task.
2023-04-17 15:58:19 +00:00
David Alvarez
26d01d18eb Support taskiter by allowing tasks to re-launch 2023-04-17 15:58:19 +00:00
1ec6f799f1 Add missing label for sponge state 2023-04-14 13:39:37 +00:00
b16c30a5e6 Add Absorbing state to the idle view in Nanos6 2023-04-14 13:39:37 +00:00
6cafd347ce Add support for sponge subsystem state in Nanos6 2023-04-14 13:39:37 +00:00
171e439bd4 Allow builds without MPI with -DUSE_MPI=OFF 2023-04-14 11:27:30 +00:00
6eead7e28e Warn if some looms don't have the rank
Fallback to sorting by loom name instead.
2023-04-14 12:13:49 +02:00
bfff1a815e Don't set rank_min multiple times 2023-04-14 12:01:47 +02:00
9568d4f379 Inform the user of the sorting method selected 2023-04-14 08:35:21 +00:00
676e30b4ca Order looms and processes by rank if given 2023-04-14 08:35:21 +00:00
80b8b209fa Skip repeated hostnames in ovnisync 2023-04-06 17:52:28 +02:00
9e105413fd Replace brk with breakdown 2023-04-06 17:38:37 +02:00
602daabe23 Fix copyright years 2023-04-06 15:33:15 +00:00
7aa3434e79 Remove ENABLE_DEBUG comments 2023-04-06 15:33:15 +00:00
29a7011141 Consider the -b option experimental 2023-04-06 15:33:15 +00:00
999a79b074 Run Nanos6 connect hook at init for breakdown 2023-04-06 15:33:15 +00:00
1c82470bfc Remove initial subsystem by default 2023-04-06 15:33:15 +00:00
b17ff6d166 Add unknown subsystem state for Nanos6 2023-04-06 15:33:15 +00:00
1582a60522 Set tr mux output to unknown subsystem on NULL 2023-04-06 15:33:15 +00:00
a827113c2a Increasing breakdown row padding to match CPU 2023-04-06 15:33:15 +00:00
881e556e06 Fix pcf_sdd_value() return check 2023-04-06 15:33:15 +00:00
cc09920c80 Remove unused i variable in emu/nanos6/breakdown.c 2023-04-06 15:33:15 +00:00
6711afdb58 Fix include headers following IWYU 2023-04-06 15:33:15 +00:00
5b64737ad9 Fix pcf_labels type 2023-04-06 15:33:15 +00:00
30f68ea5df Add missing headers 2023-04-06 15:33:15 +00:00
467cc9aa7e Fix value_str() usage 2023-04-06 15:33:15 +00:00
e287087d3b Use delta per stream in ovnidump 2023-04-06 15:33:15 +00:00
35de33e0f9 Add support for Nanos6 progress events 2023-04-06 15:33:15 +00:00
0134ceee9d Write .pcf and .row files for the breakdown 2023-04-06 15:33:15 +00:00
e838d687f0 Add missing Busy label 2023-04-06 15:33:15 +00:00
a90768c1d4 Add partial support for the breakdown model
For now only Runtime and Task breakdown trees are implemented. The Idle
state is not decomposed.
2023-04-06 15:33:15 +00:00
6b8099b6af Make the emulator extensible 2023-04-06 15:33:15 +00:00
35872354e0 Speedup the sort module
Improvements:

- Don't propagate values if they didn't change
- Use custom sort algorithm to speedup the sorting
- Allocate a contiguous array of channel outputs
2023-04-06 15:33:15 +00:00
1909d8106c Allow the select channel as input in a mux
Required to implement the tr channel in the breakdown model. See
https://pm.bsc.es/gitlab/rarias/ovni/-/issues/101
2023-04-06 15:33:15 +00:00
c624374cc8 Allow mux to specify the default value
Will be written to the output when the selection function returns NULL
as the selected input.
2023-04-06 15:33:15 +00:00
920866d8af Print remaining time with 0.1 second precision 2023-04-06 15:33:15 +00:00
4b053b9a38 Clear average line and increase elapsed precision 2023-04-06 15:33:15 +00:00
4f440c5104 Add some debug messages 2023-04-06 15:33:15 +00:00
f3bcdeccd1 Move chan_read() to a inline function in chan.h 2023-04-06 15:33:15 +00:00
43fa556fa9 Enable debug messages with -d
Allows the user to easily enable debug messages, which are the only way
to debug problems with channels.
2023-04-06 15:33:15 +00:00
31f24a2a55 Add breakdown option 2023-04-06 15:33:15 +00:00
029e185c6c Reserve first 1000 values for states
The task type hash starts at 1000, so we don't collide with states.
2023-04-06 15:33:15 +00:00
70b29b6459 Add sort channel module
Sorts the N input values and writes them in the N outputs in order.
2023-04-06 15:33:15 +00:00
a4ce0e2a1e Add Idle view in Nanos6 2023-04-06 15:33:15 +00:00
5c5f9f0f1a Use int64_t to represent the offset
Using size_t causes undefined behavior when the number is negative.
2023-04-05 12:47:36 +02:00
34a823f36d Fix ovnisort check operation
The stream_step() function will return 1 when reaching the end of the
stream, making the check return non-zero in each stream, even if all the
events have a monotonically increasing clock.
2023-03-24 12:08:45 +00:00
3d8e2000fd Don't use err() for non-error messages 2023-03-24 12:08:45 +00:00
f35741bac7 Allow setting the lock back limit in ovnisort 2023-03-24 12:08:45 +00:00
a04fc71ae9 Increase libovni version in CMakeLists.txt too 2023-03-24 12:08:45 +00:00
7827ef58e1 Fix minor version check 2023-03-24 12:08:45 +00:00
d86d5fec5f Fix overflow error for value_nextbuf 2023-03-24 12:08:45 +00:00
7f63d1e048 Remove unused thread_select_any() function 2023-03-24 12:08:45 +00:00
0041738578 Add optimization comment in stream_cmp 2023-03-24 12:08:45 +00:00
266b1638d4 Count the number of physical CPUs too 2023-03-24 12:08:45 +00:00
2c09e40c44 Use info() for information messages
Prefix err() messages with "ERROR:" too.
2023-03-24 12:08:45 +00:00
76efd7c216 Add FIXME notice in mux select chan_dirty() 2023-03-24 12:08:45 +00:00
d20483ed13 Fix compat.c headers 2023-03-24 12:08:45 +00:00
848a1de963 Constraint _GNU_SOURCE in compat.c only 2023-03-24 12:08:45 +00:00
ddbb7dd9f4 Add include-what-you-use
Adds forwards declarations in headers and includes all headers in
sources, even if they are found by transitive includes.
2023-03-24 12:08:45 +00:00
e471df9c1a Add ovniemu version 2023-03-24 12:08:45 +00:00
740d8b0d4b Only allow one rank per host in ovnisync 2023-03-24 12:08:45 +00:00
f22d9084ba Ensure all looms have at least one CPU 2023-03-24 12:08:45 +00:00
3d957f444c Only ignore clock table on ENOENT 2023-03-24 12:08:45 +00:00
648411c686 Remove trailing new line in err() and die() 2023-03-24 12:08:45 +00:00
3f44be9f2f Include path.h for loom.c 2023-03-24 12:08:45 +00:00
91957b4ed6 Remove commented code in loom.c 2023-03-24 12:08:45 +00:00
9a0d0df33d Use path_has_prefix in loom.c 2023-03-24 12:08:45 +00:00
7772541f7d Remove commented code in emu_stat.c 2023-03-24 12:08:45 +00:00
ad88e25278 Remove track mode argument from cpu_get_th_chan() 2023-03-24 12:08:45 +00:00
9a4a4a0005 Remove bay_remove() 2023-03-24 12:08:45 +00:00
5a03fd49e9 Remove function prefix from err() 2023-03-24 12:08:45 +00:00
9cb752e102 Emit warning when cfg directory exists 2023-03-24 12:08:45 +00:00
2e91d29ae9 Check fwrite return value 2023-03-24 12:08:45 +00:00
9304e84262 Fix spaces in ovni/setup.c 2023-03-24 12:08:45 +00:00
2be46393e1 Merge process_ev into model_ovni_event 2023-03-24 12:08:45 +00:00
06c250500c Warn when median burst time exceeds 100 ns 2023-03-24 12:08:45 +00:00
7b0d446b02 Add payload size checks for affinity events 2023-03-24 12:08:45 +00:00
9b209fab20 Remove spaces from kernel/setup.c 2023-03-24 12:08:45 +00:00
5556c521a0 Simplify context switch logic 2023-03-24 12:08:45 +00:00
74f253b575 Ignore Nanos6 progress events 2023-03-24 12:08:45 +00:00
96cfd6247e Remove switch in prv emit 2023-03-24 12:08:45 +00:00
9de0b18db2 Merge is_dup condition 2023-03-24 12:08:45 +00:00
38ebc3afce Protect return value with USE_RET 2023-03-24 12:08:45 +00:00
9572d79f72 Remove PRV_SKIPNULL flag 2023-03-24 12:08:45 +00:00
4a2a9ec685 Replace uses of die() with err() + return 2023-03-24 12:08:45 +00:00
0bdfea309a Simplify pcf_labels type 2023-03-24 12:08:45 +00:00
eaafd916be Fix wrong malloc error message 2023-03-24 12:08:45 +00:00
6deef0308a Fix underflow in path_remove_trailing() 2023-03-24 12:08:45 +00:00
8da0342760 Set thread and CPU channels to ignore duplicates
Some of the PRV channels no longer need to skip duplicates.
2023-03-24 12:08:45 +00:00
e625897766 Add CHAN_IGNORE_DUP to ignore duplicates 2023-03-24 12:08:45 +00:00
ab3e823134 Rename CHAN_DUPLICATES to CHAN_ALLOW_DUP 2023-03-24 12:08:45 +00:00
7fba5cbdcc Use a common variable to define the loom prefix 2023-03-24 12:08:45 +00:00
411ea962ff Don't assume chan value is int64 in chan_pop() 2023-03-24 12:08:45 +00:00
f61711047c Remove buffer argument from value_str()
Use a ring of buffers instead, so we can printf() up to 16 values in
the same call.
2023-03-24 12:08:45 +00:00
36bf4250f3 Remove die() usage from track.c 2023-03-24 12:08:45 +00:00
f957461c92 Fix emit comment in bay.c 2023-03-24 12:08:45 +00:00
6c7b64f7f2 Use memcmp() in value_is_equal() 2023-03-24 12:08:45 +00:00
2442d5decb Use 64 bit for type member in struct value
Now all the members are aligned to 64 bits, and there is no need to pack
the structure.
2023-03-24 12:08:45 +00:00
82f1cc287d Update copyright years to 2021-2023 2023-03-24 12:08:45 +00:00
6f31af0612 Remove ovni prefix from pcf.h guards 2023-03-24 12:08:45 +00:00
dc2a016c6b Add missing checks of return value 2023-03-24 12:08:45 +00:00
a0a70b1ffc Ensure the return value is used 2023-03-24 12:08:45 +00:00
30761c266b Emit the rank too when switching Nanos6 tasks
The task type and rank are allowed to emit PRV duplicates.
2023-03-24 12:08:45 +00:00
31b2ef5ffe Emit the appid and type in nosv task switch
Duplicates are allowed in the PRV for the task type, appid and rank, as
we may switch to inline if0 tasks with the same values.
2023-03-24 12:08:45 +00:00
17a0d241f6 Prefix model functions to prevent clashes 2023-03-24 12:08:45 +00:00
c508929835 Export the function to get the gid from a label 2023-03-24 12:08:45 +00:00
68690b4127 Define all static PRV types in a global enum 2023-03-24 12:08:45 +00:00
55624d4c68 Remove commented code from system.h 2023-03-24 12:08:45 +00:00
8c509475f9 Fix dbg() call 2023-03-24 12:08:45 +00:00
c39e75a8d5 Always compile dbg() statements 2023-03-24 12:08:45 +00:00
53c10aab2d Reuse ARRAYLEN() from common.h 2023-03-24 12:08:45 +00:00
d432595c7d Remove commented pcf code 2023-03-24 12:08:45 +00:00
a24477629d Set PRV flags explicitly per channel
Added PRV_EMITDUP and PRV_SKIPDUP to allow the check to be skipped. By
default it will fail to emit duplicated values.
2023-03-24 12:08:45 +00:00
48cd4de92e Use active thread for nodes subsystem
There is no reason to hide the thread (in the thread timeline) when
enters in Cooling or Warming states, as it won't cause any
oversubscription problem. CPU channel is still kept to track the running
thread.
2023-03-24 12:08:45 +00:00
15e1a5a0e9 Expand last stat message with event count 2023-03-24 12:08:45 +00:00
1d9d592b63 Fix context switch channel name 2023-03-24 12:08:45 +00:00
6039bf73c7 Ignore init time for speed stats 2023-03-24 12:08:45 +00:00
e907474ea9 Avoid double copy of mcv in emu_ev 2023-03-24 12:08:45 +00:00
c55a92872b Remove unused appid cpu channel
A similar channel is used in nosv model to report the appid of the
running task.
2023-03-24 12:08:45 +00:00
038498846e Use running thread for CPU context switches 2023-03-24 12:08:45 +00:00
8bf5d8c571 Fail gracefully when a unknown stream is found 2023-03-24 12:08:45 +00:00
0eb88af4b9 Use the CPU phyid for the name only
The logical index is used to change the affinity of the threads.
2023-03-24 12:08:45 +00:00
e0adf2f4a5 Don't run linter tests when interrupted 2023-03-24 12:08:45 +00:00
c2064c125e Run all finish() methods even after failures 2023-03-24 12:08:45 +00:00
2ebae042a8 Add ovnisync and install targets 2023-03-24 12:08:45 +00:00