By storing the per-process m_resident and m_virt values in the form
htop wants to display them in (KB, not pages), we no longer need to
have definitions of pageSize and pageSizeKB in the common CRT code.
These variables were never really CRT (i.e. display) related in the
first place. It turns out the darwin platform code doesn't need to
use these at all (the process values are extracted from the kernel
in bytes not pages) and the other platforms can each use their own
local pagesize variables, in more appropriate locations.
Some platforms were actually already doing this, so this change is
removing duplication of logic and variables there.
RichString_writeFrom takes a top spot during performance analysis due to the
calls to mbstowcs() and iswprint().
Most of the time we know in advance that we are only going to print regular
ASCII characters.
linux/SELinuxMeter.c: In function ‘hasSELinuxMount’:
linux/SELinuxMeter.c:38:21: warning: comparison of integer expressions of different signedness: ‘__fsword_t’ {aka ‘int’} and ‘unsigned int’ [-Wsign-compare]
38 | if (sfbuf.f_type != SELINUX_MAGIC) {
| ^~
Origin: 7df27b78e9/libselinux/src/init.c (L40)
linux/LinuxProcessList.c:1403:63: runtime error: division by zero
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior linux/LinuxProcessList.c:1403:63 in
The global ProcessList structure contains a couple of unused
fields. 'sharedMem' has never been used by any Meter, since
its not been anything other than zero in Linux /proc/meminfo
for many, many years. The freeMem field is only used in the
usedMem calculation, so it can reside on the stack like some
other memory variables used within-calculations-only and not
exposed to the user via a Meter.
taskstats is only checked on runtime if the column RCHAR, WCHAR, SYSCR,
SYSCW, RBYTES, WBYTES, CNCLWB, IO_READ_RATE, IO_WRITE_RATE or IO_RATE is
selected.
taskstats is currently enabled by default.
Drop the taskstats configuration switch, to reduce the maintenance cost.
cgroup is only checked on runtime if the column CGROUP is selected.
cgroup is currently disabled by default, but most distributions do
enable it.
Drop the cgroup configuration switch, to reduce the maintenance cost.
Move platform-specific code out of the htop.c main function
and into the platform sub-directories - primarily this is
the Linux procfs path check and sensors setup/teardown; not
needed on any other platforms. No functional changes here.
If currently two unsigned values are compared via `a - b`, in the case b
is actually bigger than a, the result will not be an negative number (as
-1 is expected) but a huge positive number as the subtraction is an
unsigned subtraction.
Avoid over-/underflow affected operations; use comparisons.
Modern compilers will generate sane code, like:
xor eax, eax
cmp rdi, rsi
seta al
sbb eax, 0
ret
- Add Settings forward declaration in Process.h
- Add assert.h include in XUitls.c
- Add conditional stdbool.h include in Object.h
- Drop unneeded stddef.h include in Richstring.c
- Drop unneeded unistd.h include in Process.h
- Drop unneeded string.h include in linux/Platform.c
- Use String_eq to avoid string.h include in Action.c
- Improve script to run custom iwyu version
- allow count out-parameter of String_split() to be NULL
- introduce xStrndup()
- do not allow NULL pointers passed to String_eq()
it is not used in any code
- implement String_startsWith(), String_contains_i() and String_eq()
as inline header functions
- adjust several conversion issues
Generic data, as CPU and memory usage, are used by Meters.
In paused mode they would stop receiving updates and especially Graph
Meters would stop showing continuous data.
Improves: #214Closes: #253
man:sysconf(3) states:
The values obtained from these functions are system configuration constants.
They do not change during the lifetime of a process.
Reported by UB sanitizer (alongside several other messages):
linux/LinuxProcessList.c:782:25: runtime error: member access within misaligned address 0x614000000264 for type 'struct taskstats', which requires 8 byte alignment
0x614000000264: note: pointer points here
64 01 03 00 0a 00 00 00 00 00 00 00 02 00 00 00 00 00 00 00 4b c8 2e 00 00 00 00 00 3e 45 3c fd
^
The issue doesn't cause trouble on x86, but any architecture with stricter memory alignment requirements may inadvertedly break.
Add a date meter and sort header and source files in Makefile
Change the lists of header and source files sorted alphabetical and one
file per line. This way diffs become better readable and merges easier.
Use NDEBUG conditional instead of DEBUG.
Do not call static functions in extern inline ones.
Vector.c:67:11: error: static function 'Vector_isConsistent' is used in an inline function with external linkage [-Werror,-Wstatic-in-inline]
- `CRT_fatalError()` is declared twice in CRT.h
- `Process_pidFormat`, `Process_writeField()` and `Process_compare` are
declared twice in Process.h
- `btime` is defined in LinuxProcess.c and also declared in
LinuxProcess.h, so drop in LinuxProcessList.h
This is a straightforward extension of the existing multi-column CPU meter
code, which now allows for up CPU meters to be displayed in up to 16 columns.
This also adds the meter declarations to all the platform-specific code.
This commit is based on a patch originally by @edef1c. The ZFS ARC is a cache
(it's in the name), which will be evicted by the kernel if memory pressure so
requires. Hence, the ARC should not be counted towards a system's total used
memory, and should instead be grouped with the other caches in the system.
Signed-off-by: edef <edef@edef.eu>
linux/Platform.c:142:17: warning: cast from function call of type 'double' to non-matching type 'int' [-Wbad-function-cast]
return (int) floor(uptime);
^~~~~~~~~~~~~
Use the more portable sysfs node /sys/devices/system/cpu/cpuX/cpufreq/scaling_cur_freq
to get the CPU frequency.
In case of an error fall back to /proc/cpuinfo .
Also use a fixed width of 4 for the frequency to avoid position jumps
in case the frequency moves in the range 900-1100 MHz.
unsigned overflow is well defined, but creates noise when using
sanitizers. unsigned overflow can be a symptom of logic issues of
counter, so its reasonable to use.
linux/LinuxProcessList.c:64:50: runtime error: unsigned integer overflow: 0 - 1 cannot be represented in type 'unsigned int'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior linux/LinuxProcessList.c:64:50 in
linux/LinuxProcessList.c:64:11: runtime error: implicit conversion from type 'unsigned int' of value 4294967295 (32-bit, unsigned) to type 'int' changed the value to -1 (32-bit, signed)
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior linux/LinuxProcessList.c:64:11 in
linux/LinuxProcessList.c:64:78: runtime error: unsigned integer overflow: 4 - 136 cannot be represented in type 'unsigned int'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior linux/LinuxProcessList.c:64:78 in
`env` is allocated by checked allocation functions and can not be NULL.
This checks confuses clang analyzer and causes a null-dereference
warning on `env[size-1]`.
The MIN, MAX, CLAMP, MINIMUM, and MAXIMUM macros appear
throughout the codebase with many re-definitions. Make
a single copy of each in a common header file, and use
the BSD variants of MINIMUM/MAXIMUM due to conflicts in
the system <sys/param.h> headers.
Reasoning:
- implementation was unsound -- broke down when I added a fairly
basic macro definition expanding to a struct initializer in a *.c
file.
- made it way too easy (e.g. via otherwise totally innocuous git
commands) to end up with timestamps such that it always ran
MakeHeader.py but never used its output, leading to overbuild noise
when running what should be a null 'make'.
- but mostly: it's just an awkward way of dealing with C code.
Promote the Arg union to a core data type in Object.c such
that it is visible everywhere (many source files need it),
and correct declarations of several functions that use it.
The Process_sendSignal function is also corrected to have
the expected return type (bool, not void) - an error being
masked by ignoring this not-quite-harmless warning. I've
also added error checking to the kill(2) call here, which
was previously overlooked / missing (?).
Extends the MakeHeader script to auto-generate correct "extern"
function declarations in some cases that it currently does not.
Related to https://github.com/hishamhm/htop/pull/981
openzfs_sysctl_init() now returns void instead of int.
The ZfsArcStats->enabled flag is set inside the init function
now, instead of having to be set from its return value.
Preparation for more flag setting in Compressed ARC commit.
ZfsArcMeter_readStats() added and all Meter->values[] setting
moved to it, eliminating duplicated code in
{darwin,freebsd,linux,solaris}/Platform.c.
The option is only implemented on Linux. On other platforms, and on Linuxes
that do not expose the relevant sysfs file, the frequency will be 0.
The "CPU average" meter does not show a frequency, only
the individual per-CPU meters.
If no pools are imported (ARC size == 0) or the
ZFS module is not in the kernel (/proc/spl/kstat/zfs/arcstats
does not exist), then the Meter reports "Unavailable".
The pressure stall information (PSI) metrics provide useful information
on delays caused by waiting for CPU, IO and memory. Particularly on busy
servers it can provide a quick overview of what's "slowing things down".
This feature is supported on Linux >= 4.20.
The interface is documented here:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/plain/Documentation/accounting/psi.txt
These links provide rationale:
https://lwn.net/Articles/759781/https://facebookmicrosites.github.io/psi/
The following metrics are added, corresponding to the currently exposed
lines (see `head /proc/pressure/*`):
- PressureStallCPUSome
- PressureStallIOSome
- PressureStallIOFull
- PressureStallMemorySome
- PressureStallMemoryFull
The color scheme is the same as that used for Load Average, however I
gave it separate entries just in case someone wants to change them
specifically.
Tested on 4.20.7-arch1-1-ARCH, on the linux platform.
Also tested that other platforms still compile (changed configure to use
the unsupported platform).
Closes#879.
Disable the follow process logic in Action_pickFromVector(), when
selecting sort order or user filter, since they don't apply on specific
process.
Fix#856
this way a remount of /proc will not reset starttimes
and we can also see startup times for processes started before the mount
of /proc
also record btime (boot time in seconds since epoch) as Linux semi-global
glibc 2.28 no longer defines 'major' and 'minor' in <sys/types.h> and
requires us to include <sys/sysmacros.h>. (glibc 2.25 starts
deprecating the macros in <sys/types.h>.) Now do include the latter if
found on the system.
At the moment, let's also utilize AC_HEADER_MAJOR in configure script.
However as Autoconf 2.69 has not yet updated the AC_HEADER_MAJOR macro
to reflect the glibc change [1], so add a workaround code.
Fixes#663. Supersedes pull request #729.
Reference:
[1] https://git.savannah.gnu.org/gitweb/?p=autoconf.git;a=commit;h=e17a30e987d7ee695fb4294a82d987ec3dc9b974
Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
The "if" tests if the character at index "5" is 'r', as a first quick
check. However at index "5" will always be a colon ":". This patch fixes
the off-by-one error. htop now shows proper values in the RD_SYSC
column.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Linux commit 06eb61844d841d0032a9950ce7f8e783ee49c0d0 ("sched/debug:
Add explicit TASK_IDLE printing") exposes kthreads idling using
TASK_IDLE in procfs as "I (idle)".
Until now, when sorting the STATE ("S") column, htop used the raw
value of the state character for comparison, however that led to the
undesirable effect of TASK_IDLE ('I') tasks being sorted above tasks
that were running ('R').
Thus, explicitly recognize the idle process state, and sort it below
others.
Adds support for showing columns with linux delay accounting.
This information can be read from the netlink interface, and thus we set up a socket to read from that when initializing the LinuxProcessList (LinuxProcessList_initNetlinkSocket). After that, for each process we call LinuxProcessList_readDelayAcctData, which sends a message thru the socket after setting up a callback to get the answer from the Kernel. That callback sets the process total delay time attribute. We then set the delay percent as the percentage of time process cpu time since last scan.
Nathan Scott