6d8841267a
preload: add assertion for performance hack
...
We steal two pointers temporarily, so ensure that we have the space to
overwrite a couple of addresses.
2020-03-05 15:22:17 -08:00
4af293a364
libsysprof: apply whole-system during capture replay
...
Fixes #31
2020-03-04 11:05:13 -08:00
351a2089ff
libsysprof-capture: remove unused API
...
We want this alter, but for now, we can skip it until there is a chance
to fully implement it.
2020-02-28 09:27:32 -08:00
58b9bc9992
libsysprof-ui: fix minute value
2020-02-26 21:59:03 -08:00
1fbeabf2a2
libsysprof: add context check for inline symbol decoding
2020-02-26 10:55:34 -08:00
0c698f2da1
libsysprof-ui: dont warn for not-supported error
2020-02-26 10:46:08 -08:00
f01298ead5
libsysprof: allow disabling the kernel symbol resolver
...
The kernel symbol resolver requires access to sysprofd, which might not
be available in some contexts (such as when no polkit agent is available).
This allows that to continue working by disabling the kernel with the
user-only setting.
2020-02-26 10:24:40 -08:00
14973bc2b9
callgraph: fix loading of stacks coming from backtrace()
2020-02-26 09:31:31 -08:00
26bf532a8c
libsysprof: include process ID when cmdline is not available
...
We shouldn't really hit this, but if we do, it's easy enough to synthesize
a real parent node for the process in question.
2020-02-26 08:48:05 -08:00
d34e277a6a
tests: initialize allocated value to 0
2020-02-24 15:27:58 -08:00
f1322de9e6
tests: add median to calculation
2020-02-24 15:15:20 -08:00
a9dcb58ca6
tests: add in-tree testing tool to list allocations within mark
...
This would be nice to have more generally useful via filters in the
allocations view. But for now, just make something simple work.
2020-02-24 15:11:01 -08:00
988ad116a4
libsysprof-capture: check alignment in read length
...
We require aligned frames, so we might as well enforce it in the reader.
2020-02-23 14:30:06 -08:00
fd215a39a3
libsysprof-capture: check frame length overruns
...
If we got a corrupted frame, just stop reading forward immediately.
2020-02-23 14:26:13 -08:00
98f6dc6642
tools: add --env=VAR=VALUE command line option
2020-02-21 12:40:55 -08:00
cd2c2c954a
libsysprof: extend LD_PRELOAD when pre-existing
2020-02-21 12:40:55 -08:00
9b52e88d3a
libsysprof-ui: add label for number of allocations
2020-02-21 12:39:13 -08:00
5b88e9c3aa
build: check for unw_set_cache_size()
...
This may not be available on older libunwind versions such as 1.2.
2020-02-20 11:22:56 -08:00
b0157683ef
preload: use #ifdef, not #if ENABLE_LIBUNWIND
2020-02-20 11:08:26 -08:00
348a1ef110
libsysprof: fix use of {0} initializers for older GCC
...
Older GCC versions complain about this so we might as well squash it.
2020-02-20 11:08:21 -08:00
0cb4bb61ac
libsysprof: remove use of GAtomicRCBox
...
Switching this to use an embedded ref count allows us to backport to
operating systems restricted to GLib 2.56.
2020-02-20 11:03:05 -08:00
f25c573ba6
libsysprof-ui: remove use of GAtomicRCBox
...
Switching this to use an embedded ref count allows us to backport to
operating systems restricted to GLib 2.56.
2020-02-20 11:02:49 -08:00
d89a689ab4
sysprofd: remove use of GAtomicRCBox
...
Using an embedded ref count allows us to backport to older operating
systems for which GLib is restricted to 2.56.
2020-02-20 11:01:02 -08:00
1017fed467
tests: add missing locale.h include
2020-02-20 10:59:57 -08:00
be6b979714
libsysprof-capture: remove use of GAtomicRCBox
...
If we backport to an older GLib release, we won't have the newer atomic
helpers available. It's really not too much burden to do that manually if
it means we can run on older systems.
2020-02-20 10:39:33 -08:00
cd8a99402f
build: fallback to __sync_synchronize()
...
If we're running on a GCC older than 4.9, then we won't have the
stdatomic.h available. We can just use a full barrier instead using
__sync_synchronize() to get the same effect, albeit slower.
2020-02-20 10:38:35 -08:00
2cd05c1bd8
libsysprof-capture: fallback to g_time_val_to_iso8601()
...
To allow us to backport to older systems, we can have a fallback here for
the time being.
2020-02-20 10:36:18 -08:00
a62664f8e5
memprof: add allocation instead of self to clipboard
2020-02-19 20:12:02 -08:00
8e81b1fcf9
libsysprof: add summary information for memprof profile
2020-02-19 00:16:36 -08:00
6b1872a5ce
libsysprof-ui: dont change pages if operation was cancelled
2020-02-18 21:03:12 -08:00
dd6b44cb2f
libsysprof-ui: move controls above stack
2020-02-18 19:56:30 -08:00
16e35c1fa6
libsysprof-ui: allow switching to temporary allocations
2020-02-18 19:20:38 -08:00
e3cb30e4ee
libsysprof: apply selection to temporary allocations
2020-02-18 19:19:01 -08:00
d525f065f4
libsysprof-ui: add API to reload pages
...
This can be useful when the page is interactive and needs to reload.
2020-02-18 19:17:38 -08:00
059508d582
memprof: fix whitespace
2020-02-18 17:22:32 -08:00
84e2c288dc
libsysprof: add support for calculating temporary allocations
...
This is useful to find allocations free'd right after they were created.
A temporary allocation is currently defined as a free() right after an
allocation of that same memory address. From a quick glance, that appears
to be similar to what I've been seeing in heaptrack all these years.
In the long term, I'd expect we can do something more useful such as
"freed from similar stack trace" since things like g_strdup_printf()
would obviously break important temporary allocations.
2020-02-18 17:18:04 -08:00
87004f5d24
preload: skip an additional stack frame
...
Now that we do some math here, we need to skip another frame to keep this
more useful and not show everything inside of sysprof_collector_allocate().
2020-02-18 14:11:09 -08:00
0d6a3d0183
libsysprof-capture: fix use of n_addrs
2020-02-18 14:05:26 -08:00
7490a774ab
libsysprof-capture: use signed int for backtrace return
...
This allows us to more safely subtract 1 from the unw_backtrace() to get
the proper number of frames (and detect it in the collector).
2020-02-18 14:03:19 -08:00
9f43bf2813
libsysprof: only add process cmdline info once
...
We might get this information from multiple sources (such as Linux's perf
or the proc data source). So only add this information once to avoid
having additional data we don't care about.
This also helps ensure we get a proper tree for the callgraph without
splitting things between updated cmdline information.
2020-02-18 13:45:24 -08:00
93acce520f
libsysprof: plug leak and do less strdups
...
Small leak based on the number of PROCESS frames we see. Easy to fix and
easy to not do the g_strdup_printf() at all in those cases.
2020-02-18 13:44:25 -08:00
fdf912b716
libsysprof-capture: do less work to calculate frame length
...
These values are already all aligned, so there is no need to realign them.
2020-02-18 13:35:45 -08:00
70bea64f88
libsysprof-capture: allow for backtrace skip optimization
...
We want to be backtracing directly into the capture buffer, but also need
to skip a small number of frames.
If we call the backtrace before filling in information, we can capture to
the position *before* ev->addrs and then overwrite that data right after.
2020-02-18 13:35:18 -08:00
b6dc058d62
libsysprof: check for time series after extracting processes
2020-02-18 13:17:49 -08:00
20f7c271ca
libsysprof-capture: clear SYSPROF_CONTROL_FD
2020-02-18 13:17:33 -08:00
4cfb51421b
libsysprof-capture: set tid to pid on non-Linux
2020-02-18 13:17:22 -08:00
4fd2068367
tests: ensure we sort by time after pid/tid
2020-02-18 12:00:08 -08:00
521024d51f
tests: add tool to list total/temp/leaked allocations
...
This is just a quick example of how to find temporary allocations.
2020-02-18 11:56:51 -08:00
77a13fe7f9
icons: provide helper icons
...
Fixes https://gitlab.gnome.org/GNOME/sysprof/issues/30
2020-02-18 12:05:55 +01:00
a37ad780ca
preload: use unw_backtrace()
...
This seems to be significantly faster than doing the manual stepping. A
quick look shows that it has a number of special cases which we'd have
to duplicate, so best to just use it directly.
2020-02-17 12:05:32 -08:00
