]> sourceware.org Git - systemtap.git/log
systemtap.git
10 years agoDon't give predefined types special treatment.
Stan Cox [Thu, 29 May 2014 15:11:40 +0000 (11:11 -0400)]
Don't give predefined types special treatment.

* dtrace.in (dtrace_bnf):  Remove type_spec.  Handle all types as type_name.

10 years agoFixed BZ1100813 by fixing build errors in nfsd.stp.
David Smith [Wed, 28 May 2014 21:00:44 +0000 (16:00 -0500)]
Fixed BZ1100813 by fixing build errors in nfsd.stp.

* tapset/linux/nfsd.stp (__rqstp_uid): Fix pointer casting, needed by
  32-bit platforms.
  (__rqstp_gid): Ditto.

10 years agoPR16991: tweak output for UNKNOWN root dentry; add hex address
Frank Ch. Eigler [Wed, 28 May 2014 16:23:04 +0000 (12:23 -0400)]
PR16991: tweak output for UNKNOWN root dentry; add hex address

10 years agoFixed PR16991 by avoiding an inifinte loop in task_dentry_path().
David Smith [Wed, 28 May 2014 15:54:57 +0000 (10:54 -0500)]
Fixed PR16991 by avoiding an inifinte loop in task_dentry_path().

* tapset/linux/dentry.stp (task_dentry_path): Return "UNKNOWN" for
  unmounted synthetic filesystems.
* testsuite/systemtap.base/task_dentry_path.exp: New file.
* testsuite/systemtap.base/task_dentry_path.sh: Ditto.
* testsuite/systemtap.base/task_dentry_path.stp: Ditto.

10 years agoBZ1099472: cast-scope.exp: adjust relative lineno
Jonathan Lebon [Wed, 28 May 2014 15:24:34 +0000 (11:24 -0400)]
BZ1099472: cast-scope.exp: adjust relative lineno

10 years agoBZ1099757: vars.exp: fix and strengthen
Jonathan Lebon [Wed, 28 May 2014 15:12:18 +0000 (11:12 -0400)]
BZ1099757: vars.exp: fix and strengthen

Since PR16615 (merge commit 4636ca3), linenos in statement probes have
become more precise and stringent. As a result, vars.exp was failing
because it was trying to probe a statement at a lineno with no line
records.

We fix this by first retrieving the first valid lineno using stap -l.
The test has also been refactored in general to be more robust. We can
now more easily add more functions to test (e.g. we now test both
bio_copy_user() and vfs_read()).

10 years agoman/stapprobes.3stap: clarify relative linenos
Jonathan Lebon [Wed, 28 May 2014 13:57:21 +0000 (09:57 -0400)]
man/stapprobes.3stap: clarify relative linenos

Since PR16615 (merge commit 4636ca3), relative line numbers are relative
to the decl_line of the function. Adjust man page to reflect this.

10 years agotapsets.cxx (trivial): clear up comment
Jonathan Lebon [Tue, 27 May 2014 21:32:54 +0000 (17:32 -0400)]
tapsets.cxx (trivial): clear up comment

We used to support both kernel mode and user mode before, but no more.
Update comment to not cause confusion.

10 years agodwflpp (trivial): fix missing _F()
Jonathan Lebon [Mon, 26 May 2014 18:05:58 +0000 (14:05 -0400)]
dwflpp (trivial): fix missing _F()

10 years agoNEWS: add blurb about PR16941
Jonathan Lebon [Mon, 26 May 2014 18:07:43 +0000 (14:07 -0400)]
NEWS: add blurb about PR16941

10 years agoMerge branch 'jlebon/listing' (PR16941)
Jonathan Lebon [Mon, 26 May 2014 14:43:45 +0000 (10:43 -0400)]
Merge branch 'jlebon/listing' (PR16941)

10 years agoMerge branch 'jlebon/library' into jlebon/listing
Jonathan Lebon [Mon, 26 May 2014 14:39:14 +0000 (10:39 -0400)]
Merge branch 'jlebon/library' into jlebon/listing

10 years agolist.exp: fix regexes and strengthen
Jonathan Lebon [Fri, 16 May 2014 14:45:27 +0000 (10:45 -0400)]
list.exp: fix regexes and strengthen

Fix regular expressions so that we expect the full path from stap rather
than just process("stap"). Also strengthen the testcase by making
matches against all output lines, rather than just checking if a single
line matches.

10 years agoderive_probes(): properly handle optionality
Jonathan Lebon [Tue, 13 May 2014 21:15:01 +0000 (17:15 -0400)]
derive_probes(): properly handle optionality

We previously overrode a location's optionality if it was derived from a
globby process probe. However, this causes further probe points to be
derived to also be optional, even though the 'higher level' probe point
had its optionality restored to what the user really wants. The end
result was that we would list probes as not optional even though the
endpoint probe was optional.

What we really mean when we say process probes born out of a higher
globby process probe should be optional is that only some of the matched
binaries will actually resolve, so it's OK if some don't resolve.
However, those that *do* resolve should respect the user's wish with
regards to optionality.

To do this, we must not override the probe_point's optional flag in
derive_probes.

This is mostly a cosmetic change, although some runtime elements do rely
on optionality to decide whether to emit a warning upon failing to
register a probe.

This patch also makes process.library probes born out of a higher globby
library probe follow the same logic.

10 years agolabels.exp: adjust lineno
Jonathan Lebon [Thu, 15 May 2014 15:56:10 +0000 (11:56 -0400)]
labels.exp: adjust lineno

10 years agolibrary probes: update stapprobes(3stap)
Jonathan Lebon [Thu, 8 May 2014 20:44:51 +0000 (16:44 -0400)]
library probes: update stapprobes(3stap)

Clarify that we normally search among NEEDED libraries but that
non-NEEDED libraries are also supported.

10 years agolabel probes: rework logic
Jonathan Lebon [Thu, 15 May 2014 15:08:36 +0000 (11:08 -0400)]
label probes: rework logic

Ever since PR16615 (merge commit 4636ca3), statement/function probes
where we specify the full "func@file:N" have been too precise for label
probes.

Imagine the following C program:

   ...
   <expression-statement>
mylabel:
   <expression-statement>
   ...

The line at which 'mylabel' occurs is not actually associated with any
line records. Thus, when using the new iterate_over_srcfile_lines(),
specifying the lineno of the label results in no records found, whereas
prior to PR16615, the next lineno would be picked up.

Regardless, the previous mechanism of using iterate_over_srcfile_lines()
to then call back to query_srcfile_label() which in turn called
iterate_over_labels() was a roundabout way of making sure that the
specified lineno lies in a filtered function's DIE.

We now rework iterate_over_labels() to directly handle lineno matching
when necessary. Supported cases:

process.function("foo@file.c:N").label("*")
   --> matches label in foo at N

process.function("foo@file.c+N").label("*")
   --> matches label in foo at N + decl_line of foo

process.function("foo@file.c:N-M").label("*")
   --> matches labels in foo between N and M

process.function("foo@file.c:*").label("*")
   --> matches all labels in foo

This patch also pushes down the priority of "func@file:N" type probes so
that if a .label or .callee[s] is present, then the relevant
iterate_over_*() is called, rather than iterate_over_srcfile_lines().

10 years agolibrary probes: suggest libraries on mismatch
Jonathan Lebon [Wed, 7 May 2014 18:27:29 +0000 (14:27 -0400)]
library probes: suggest libraries on mismatch

Suggest from the iterated libraries if no matching libraries were found.

10 years agoalias probes: mark final aliases as well-formed
Jonathan Lebon [Thu, 15 May 2014 22:06:17 +0000 (18:06 -0400)]
alias probes: mark final aliases as well-formed

When building probes from aliases, we mark them as well-formed if they
reached their final derivation so that in listing mode, we stop at this
final alias form.

10 years agolibrary probes: also try non-NEEDED libraries
Jonathan Lebon [Wed, 7 May 2014 18:16:24 +0000 (14:16 -0400)]
library probes: also try non-NEEDED libraries

This patch adds back support for non-NEEDED libraries even when the user
supplied a non-canonical path. However, a WARNING is emitted to notify
the user. This WARNING can be suppressed by specifying the fully
resolved path to the library.

10 years agoprocess probes: throw error upon unresolved executable
Jonathan Lebon [Mon, 26 May 2014 16:01:18 +0000 (12:01 -0400)]
process probes: throw error upon unresolved executable

When find_executable() fails to resolve something, it simply returns the
same path it was fed. Depending on the context, the caller may be able
to deal with an unresolved path.

In the case of process probes for dwarf_builder and other base_query
users, dealing with it meant simply that the dwflpp failed to find the
executable which meant iterate_over_modules() never even took off (and
we thus yielded the good ol' "no match" error).

However, in the case of library probes, this may not always work because
if the library path is fully specified, we will switch to it without
resorting to iterate_over_libraries() (see also previous patch) and
without ever actually checking if the process path was fully resolved,
thus allowing a probe such as the following to be created:

process("garbage").library("/full/path/to/library.so").function("foo")

So instead, we add explicit checks to verify that process probes were
properly resolved by find_executable(). As an added benefit, users get a
clearer reason of why the probe point failed:

BEFORE:

semantic error: while resolving probe point: identifier 'process' at <input>:1:7
        source: probe process("asdf").library("mylib.so").function("foo") {}
                      ^

semantic error: no match

AFTER:

semantic error: while resolving probe point: identifier 'process' at <input>:1:7
        source: probe process("asdf").library("mylib.so").function("foo") {}
                      ^

semantic error: cannot find executable 'asdf'

10 years agolibrary probes: strengthen library path resolution
Jonathan Lebon [Wed, 7 May 2014 14:59:13 +0000 (10:59 -0400)]
library probes: strengthen library path resolution

One of two things happens to library probes the first time we enter
dwarf_builder::build():
   - If the full path can be resolved using find_executable(), then we
     would right away switch focus to the library, rather than the
     process.
   - Otherwise, we would focus on the process and then use
     iterate_over_libraries() to find a library match.

These decisions are made in two spots: in dwarf_builder::build(), prior
to creating the dwflpp object, and in base_query::base_query().

When we resort to iterate_over_libraries(), new probes are created for
each matching library. And since the path inserted in those new probes
is the full path, find_executable() will find it upon recursing into
dwarf_builder::build().

One issue is that if find_executable() succeeds on the first time, we
will focus on the library, regardless of whether the library is NEEDED
or not. Whereas if find_executable() fails, only NEEDED libraries could
ever be picked up by iterate_over_libraries().

This is mostly fine, except that it creates an inconsistency. For
example, let's say executable 'bar' depends on 'libfoo.so' located in
/lib/. Doing a "libfoo.so" library probe would prioritize a libfoo.so
file in the current directory before the one in /lib/, which is the
'real' one. On the other hand, doing "libfoo.*" would match the
libfoo.so file in /lib/ because of the glob char.

It also fixes the potential issue of having a fully specified library
path with a bad process path: we would focus on the library yet in
query_module() try to iterate_over_libraries() on the library, thinking
that the library was not resolved.

In this series of patches, we change the behaviour as follow
   - Only switch to the library on the first entry of
     dwarf_builder::build() if the path is already fully resolved.
     Otherwise, always use iterate_over_libraries().
   - If iterate_over_libraries() fails, then try a direct
     find_executable() invocation, and warn that the found library is
     not a NEEDED one.

This patch does the first part only. The second part is done in the next
patch.

10 years agolibrary probes: don't add libraries to visited_modules
Jonathan Lebon [Tue, 6 May 2014 20:39:04 +0000 (16:39 -0400)]
library probes: don't add libraries to visited_modules

The dwarf_query's visited_modules set is used to keep track of which
modules are visited for suggestions. It works by emptying out the visited
modules into the dwarf_builder's modules_seen after we're done iterating
through the modules. By the end, modules_seen contains the full list of
modules to suggest from.

Since .library(glob) probes resolve to multiple probes for each matching
library, dwarf_builder's modules_seen will already pick each of them up
from the individually created dwarf_query objects.

So in the end, the modules_seen from which we make suggestions will
already contain them.

10 years agojava probes: mark initial pp as well-formed
Jonathan Lebon [Fri, 16 May 2014 19:23:59 +0000 (15:23 -0400)]
java probes: mark initial pp as well-formed

10 years agoprobe: remove unused create_alias() function
Jonathan Lebon [Thu, 15 May 2014 20:59:09 +0000 (16:59 -0400)]
probe: remove unused create_alias() function

10 years agodwarf_query: tweak final_function_name() signature
Jonathan Lebon [Thu, 15 May 2014 19:01:43 +0000 (15:01 -0400)]
dwarf_query: tweak final_function_name() signature

Almost all of the users of final_function_name() call it with a const
char* for the filename, rather than a string. There may be cases where
the filename is NULL (e.g. if the function was derived using the
symtab), in which case we don't want to cast to a string.

10 years agoquery_srcfile_line(): add well-formed probe point
Jonathan Lebon [Wed, 14 May 2014 15:24:19 +0000 (11:24 -0400)]
query_srcfile_line(): add well-formed probe point

We create the final well-formed probe point before actually creating the
derived probe.

10 years agoquery_plt_statement(): add well-formed probe point
Jonathan Lebon [Wed, 14 May 2014 14:48:39 +0000 (10:48 -0400)]
query_plt_statement(): add well-formed probe point

If the user provides a .plt[(string)] probe, then we'll end up in
query_one_plt during plt iteration, in which we create a specific_loc
where we replace the possibly globby string with the found plt. Mark
this specific loc as well-formed.

If the user provides a .plt[(string)].statement(number) probe, then
we'll end up in query_plt_statement() directly, in which case we need to
create the final well-formed probe point in which we omit the .plt part.

10 years agoquery_inline_instance_info(): add well-formed probe point
Jonathan Lebon [Wed, 14 May 2014 13:34:08 +0000 (09:34 -0400)]
query_inline_instance_info(): add well-formed probe point

We create the final well-formed probe point before actually creating the
derived probe.

10 years agoquery_func_info(): add well-formed probe point
Jonathan Lebon [Wed, 14 May 2014 04:14:48 +0000 (00:14 -0400)]
query_func_info(): add well-formed probe point

We create the final well-formed probe point before actually creating the
derived probe.

10 years agoquery_addr(): add well-formed probe point
Jonathan Lebon [Tue, 13 May 2014 21:40:52 +0000 (17:40 -0400)]
query_addr(): add well-formed probe point

We create the final well-formed probe point before actually creating the
derived probe.

10 years agoquery_label(): add well-formed probe point
Jonathan Lebon [Tue, 13 May 2014 20:57:12 +0000 (16:57 -0400)]
query_label(): add well-formed probe point

The probe points created in query_label() are already well-formed. We
use the new interface installed in dwarf_query to simplify it.

10 years agoquery_callee(): add well-formed probe point
Jonathan Lebon [Tue, 13 May 2014 20:40:10 +0000 (16:40 -0400)]
query_callee(): add well-formed probe point

The probe points created in query_callee() are already well-formed. We
use the new interface installed in dwarf_query to simplify it.

10 years agodwarf_query: add probe point editing helper functions
Jonathan Lebon [Wed, 14 May 2014 16:32:39 +0000 (12:32 -0400)]
dwarf_query: add probe point editing helper functions

As mentioned in the previous commit, these functions will help us to
modify the newly mounted probe point into a well-formed one.

10 years agodwarf_query: add mount/unmount_well_formed_probe_point()
Jonathan Lebon [Wed, 14 May 2014 16:20:36 +0000 (12:20 -0400)]
dwarf_query: add mount/unmount_well_formed_probe_point()

In commits e772a6e and 9891c54, we improved the way query_callee() and
query_label() add intermediate probes to the derivation chain by playing
with the dwarf_query. This technique will be required many times in
other places that we will need to fix up.

Therefore, we create two new dwarf_query helper functions. The
mount_well_formed_probe_point() function simply sets the current
base_loc and base_probe of the dwarf_query object to a new one, which is
denoted well-formed.

At first, this new location is simply a copy of the previous location,
with the exception of a fully resolved process/module path. Then,
depending on the probe point type, specific functors are replaced by
their final/non-globby versions to create a truly well-formed probe
point.

More helper functions to edit the functors will be added in the next
patch.

The unmount_well_formed_probe_point() refocuses the dwarf_query to the
previous base_loc/base_probe so that further probe points can be
derived.

10 years agokprobe_builder: remember the .call component
Jonathan Lebon [Tue, 13 May 2014 20:43:24 +0000 (16:43 -0400)]
kprobe_builder: remember the .call component

Although it doesn't make a difference, we remember whether .call was
specified and include it in the created probe so that it is listed as
well.

There's no need to mark it well-formed. It is the last destination, and
it is already well-formed.

10 years agosdt_query: set well-formed intermediate probe
Jonathan Lebon [Wed, 14 May 2014 14:31:07 +0000 (10:31 -0400)]
sdt_query: set well-formed intermediate probe

The specific_loc intermediate probe created in
sdt_query::convert_location() is already well-formed. We mark it as such
so that listings will use this probe.

10 years agoperf_builder: add well-formed intermediate probe
Jonathan Lebon [Tue, 13 May 2014 20:55:06 +0000 (16:55 -0400)]
perf_builder: add well-formed intermediate probe

10 years agohwbkpt_builder: add well-formed intermediate probe
Jonathan Lebon [Tue, 13 May 2014 20:24:35 +0000 (16:24 -0400)]
hwbkpt_builder: add well-formed intermediate probe

10 years agosimplify derived_probe::script_location()
Jonathan Lebon [Tue, 13 May 2014 20:18:43 +0000 (16:18 -0400)]
simplify derived_probe::script_location()

Rather than relying on yucky heuristics, we go the explicit way. The
script_location() function now relies on the new probe_point member
'well_formed', which is true if the probe point represents the 'final'
script-level expression, after wildcard expansion.

Since only each individual probe builder knows what constitutes a
well-formed probe point, it will be up to those to set the 'well_formed'
flag as needed. If the probe point that we want to list is also the
final one, which is the case for many simpler probe points, then no
further work is needed. The following patches modify the appropriate
builders to do just this.

10 years agolisting_mode.exp: new testcase
Jonathan Lebon [Mon, 12 May 2014 21:01:20 +0000 (17:01 -0400)]
listing_mode.exp: new testcase

10 years agolisting_mode.exp: rename to listing_mode_sanity.exp
Jonathan Lebon [Mon, 12 May 2014 20:59:28 +0000 (16:59 -0400)]
listing_mode.exp: rename to listing_mode_sanity.exp

10 years agond_syscall.exp: fix SYSTEMTAP_DEBUGINFO_PATH override
Jonathan Lebon [Mon, 26 May 2014 21:34:27 +0000 (17:34 -0400)]
nd_syscall.exp: fix SYSTEMTAP_DEBUGINFO_PATH override

This is a follow-up to commit 27ce133. Tcl's exec proc can't handle env
var setting as part of its args:

couldn't execute "SYSTEMTAP_DEBUGINFO_PATH=/dev/null": no such file or
directory
FAIL: testsuite/systemtap.syscall/nd_sys.stp compilation

We need to explicitly manipulate the env array to set the
SYSTEMTAP_DEBUGINFO_PATH and then restore it afterwards if needed.

10 years agoImprove BZ16956 by testing nd_syscall tapset without debuginfo.
David Smith [Wed, 21 May 2014 20:21:41 +0000 (15:21 -0500)]
Improve BZ16956 by testing nd_syscall tapset without debuginfo.

* testsuite/systemtap.syscall/nd_syscall.exp: Make sure the nd_syscall
  tapset isn't using debuginfo by setting SYSTEMTAP_DEBUGINFO_PATH to
  /dev/null.

10 years agoFix BZ16960 by handling !CONFIG_USER_NS better.
David Smith [Wed, 21 May 2014 19:09:41 +0000 (14:09 -0500)]
Fix BZ16960 by handling !CONFIG_USER_NS better.

* buildrun.cxx (compile_pass): Convert the from_kuid_munged() test to a
  compile test, instead of an export test.
* runtime/linux/runtime.h: Restore original CONFIG_USER_NS support.
* runtime/linux/autoconf-from_kuid_munged.c: New autoconf test.

10 years agolinux runtime: switch to vmalloc for context[] element allocation
Frank Ch. Eigler [Wed, 21 May 2014 17:43:48 +0000 (13:43 -0400)]
linux runtime: switch to vmalloc for context[] element allocation

The new eventcount.stp changes can result in relatively large context
structures (33712 bytes each on f19/x86-64), which the stp_kmalloc
widget can sometimes fail to allocate.  Switching to the numa-aware
vmalloc front-end lets these allocations succeed.

TBD: those contexts shouldn't be that large.

10 years agoeventcount.stp sample: add some more sorting options, config disclosure
Frank Ch. Eigler [Wed, 21 May 2014 17:11:26 +0000 (13:11 -0400)]
eventcount.stp sample: add some more sorting options, config disclosure

10 years agostapprobes.3stap: highlight obsolete nature of kernel.mark probes
Frank Ch. Eigler [Wed, 21 May 2014 13:20:28 +0000 (09:20 -0400)]
stapprobes.3stap: highlight obsolete nature of kernel.mark probes

10 years agoPR16955: adapt testsuite
Jonathan Lebon [Sun, 18 May 2014 19:29:03 +0000 (15:29 -0400)]
PR16955: adapt testsuite

- pr10854.exp: use -w to suppress warnings re. blacklisted functions
- rlimit.exp: ditto

10 years agoPR16955: improve blacklisting diagnostics
Jonathan Lebon [Fri, 16 May 2014 20:48:26 +0000 (16:48 -0400)]
PR16955: improve blacklisting diagnostics

Create a new enum to describe the reason why the function was
blacklisted. In dwarf_query::add_probe_point(), emit a warning if
blacklisted.

10 years agolinux/runtime.h: more CONFIG_USER_NS fallout
Frank Ch. Eigler [Sat, 17 May 2014 20:15:01 +0000 (16:15 -0400)]
linux/runtime.h: more CONFIG_USER_NS fallout

Partially unrolling commit 7d35a948401, because on modern kernels
(3.14+), the from_k?id_munged functions are defined as inlines, and
should be used regardless of CONFIG_USER_NS etc.  The buildrun.cxx
exportedness test for from_kuid_munged would fail, triggering our
runtime to use a do-nothing macro instead.

Until a more programmatic way is found for those unidentified older
kernels that have CONFIG_USER_NS but no from_k?id_munged inlines,
for now a user will have to work around the build problem with

  % stap -DSTAP_FAKE_KID_MUNGED

10 years agoMore linux 3.14 porting, sans CONFIG_USER_NS
Frank Ch. Eigler [Sat, 17 May 2014 19:36:46 +0000 (15:36 -0400)]
More linux 3.14 porting, sans CONFIG_USER_NS

This time, cover the remaining tapset/runtime stuff.

10 years agoPR16956: add "kernel<linux/types.h>" to all nd_syscall.* @cast ops
Frank Ch. Eigler [Sat, 17 May 2014 16:54:37 +0000 (12:54 -0400)]
PR16956: add "kernel<linux/types.h>" to all nd_syscall.* @cast ops

We can't take blood from stone - dwarf from a kernel whom we are
targeting for non-dwarf processing.  So generate dwarf from the
headers if necessary.

10 years agowarn about missing kernel symbol table
Frank Ch. Eigler [Sat, 17 May 2014 16:31:46 +0000 (12:31 -0400)]
warn about missing kernel symbol table

On some distros (e.g., arch linux), no System.map* file survives the
build process, but there may be ways to fabricate doppelgangers.

* session.cxx (systemtap_session::parse_kernel_functions): Warn if
  none found.
* warning::symbols: new man page
* warning::debuginfo: xref.

10 years agosession: don't suppress warnings in dump mode if -vv
Jonathan Lebon [Fri, 16 May 2014 21:11:16 +0000 (17:11 -0400)]
session: don't suppress warnings in dump mode if -vv

Just like how derive_probes() still prints errors in dump mode if -vv is
supplied, we do the same for warnings.

10 years agowhitespace tweak for examples html index
Frank Ch. Eigler [Fri, 16 May 2014 16:06:49 +0000 (12:06 -0400)]
whitespace tweak for examples html index

10 years agoAdd some popularity numbers for examples/category links.
Frank Ch. Eigler [Fri, 16 May 2014 16:05:17 +0000 (12:05 -0400)]
Add some popularity numbers for examples/category links.

10 years agoregenerate examples index
Frank Ch. Eigler [Fri, 16 May 2014 15:52:30 +0000 (11:52 -0400)]
regenerate examples index

10 years agopublican docs: define a BZURL
Frank Ch. Eigler [Thu, 15 May 2014 19:00:48 +0000 (15:00 -0400)]
publican docs: define a BZURL

10 years agoAdd pyparsing dependency for sdt-devel
Lukas Berk [Thu, 15 May 2014 18:36:11 +0000 (14:36 -0400)]
Add pyparsing dependency for sdt-devel

*systemtap.spec - add dependency for new dtrace functionality

10 years agoFix probes with no args for --no-pyparsing mode.
Stan Cox [Thu, 15 May 2014 17:26:39 +0000 (13:26 -0400)]
Fix probes with no args for --no-pyparsing mode.

* dtrace.in (_ReProvider): Check for no args.
* dtrace.exp: Fix test for probe with no args.

10 years agosdt_varname.exp: small tweaks
Jonathan Lebon [Mon, 12 May 2014 13:50:20 +0000 (09:50 -0400)]
sdt_varname.exp: small tweaks

1. Only run testcase if installtest
2. Uncomment the (mistakenly commented out) code that cleans up the
   testing directory upon exit
3. Enable -fPIC when compiling library, which can yield different forms
   of operands, thus improving the testing range

10 years agocallee.exp: simplify shlib testing further using -shared
Jonathan Lebon [Fri, 9 May 2014 20:34:41 +0000 (16:34 -0400)]
callee.exp: simplify shlib testing further using -shared

This is a follow-up to commit 72ad110. We can skip the library's
intermediate object file by directly using -shared in the first
invocation.

10 years agoMerge branch 'jlebon/pr13296'
Jonathan Lebon [Fri, 9 May 2014 19:57:28 +0000 (15:57 -0400)]
Merge branch 'jlebon/pr13296'

This merge gives SystemTap the ability to decode SDT operands that refer
to symbols. There are also a few code cleanups, such as:
   - Removal of SDT probing code that supports a hypothetical kernel
     mode.
   - Removal of old unused symbol table code for reading in symbols by
     reading output from nm or from a file.
   - Major refactoring of the sdt_uprobe_var_expanding_visitor::
     visit_target_symbo_arg() function to make it easier to understand
     the overall structure.
   - Strengthening of sections/note entry checks before parsing probe
     information.

10 years agoNEWS: add blurb about pr13296
Jonathan Lebon [Fri, 9 May 2014 19:49:36 +0000 (15:49 -0400)]
NEWS: add blurb about pr13296

10 years agosdt_varname.exp: new testcase for varname SDT probing
Jonathan Lebon [Fri, 9 May 2014 18:10:25 +0000 (14:10 -0400)]
sdt_varname.exp: new testcase for varname SDT probing

This new testcase verifies that SDT operands of the form
[off+]varname[+off][(reg)] are properly parsed by stap.

10 years agoSDT probes: remove kernel-specific handling
Jonathan Lebon [Tue, 6 May 2014 19:16:50 +0000 (15:16 -0400)]
SDT probes: remove kernel-specific handling

10 years agofactor out precision_to_function
Jonathan Lebon [Tue, 6 May 2014 18:35:13 +0000 (14:35 -0400)]
factor out precision_to_function

10 years agofactor out try_parse_arg_varname
Jonathan Lebon [Tue, 6 May 2014 18:27:54 +0000 (14:27 -0400)]
factor out try_parse_arg_varname

10 years agofactor out try_parse_arg_effective_addr
Jonathan Lebon [Tue, 6 May 2014 18:11:56 +0000 (14:11 -0400)]
factor out try_parse_arg_effective_addr

10 years agofactor out try_parse_arg_offset_register
Jonathan Lebon [Tue, 6 May 2014 18:01:47 +0000 (14:01 -0400)]
factor out try_parse_arg_offset_register

10 years agofactor out try_parse_arg_register
Jonathan Lebon [Tue, 6 May 2014 16:37:25 +0000 (12:37 -0400)]
factor out try_parse_arg_register

10 years agofactor out build_dwarf_registers
Jonathan Lebon [Tue, 6 May 2014 16:29:59 +0000 (12:29 -0400)]
factor out build_dwarf_registers

10 years agofactor out try_parse_arg_literal
Jonathan Lebon [Tue, 6 May 2014 16:06:46 +0000 (12:06 -0400)]
factor out try_parse_arg_literal

10 years agofactor out parse_out_arg_precision
Jonathan Lebon [Fri, 9 May 2014 19:28:10 +0000 (15:28 -0400)]
factor out parse_out_arg_precision

10 years agofactor out get_target_symbol_argno_and_validate
Jonathan Lebon [Tue, 6 May 2014 15:44:05 +0000 (11:44 -0400)]
factor out get_target_symbol_argno_and_validate

10 years agoupdate table of possible operand syntax
Jonathan Lebon [Tue, 6 May 2014 15:28:42 +0000 (11:28 -0400)]
update table of possible operand syntax

10 years agoPR13296: allow SDT operands using VARNAME
Jonathan Lebon [Thu, 1 May 2014 15:59:15 +0000 (11:59 -0400)]
PR13296: allow SDT operands using VARNAME

We may encounter SDT operands of the form [OFF+]VARNAME[+OFF][(REG)]
which refers to the address of a object in the symtab. On x86_64, the
REG can be '%rip' to denote RIP-relative addressing.

To do this, we also pass the current dwflpp object to
sdt_uprobe_var_expanding_visitor, which then tries symbol resolution
using the symbol table if it meets upon such an SDT operand.

10 years agosymbol_table::read_from_elf(): also pick up globals and locals
Jonathan Lebon [Thu, 1 May 2014 15:57:23 +0000 (11:57 -0400)]
symbol_table::read_from_elf(): also pick up globals and locals

These will be needed in upcoming patches.

10 years agosymbol_table: remove read_* methods
Jonathan Lebon [Thu, 1 May 2014 15:18:22 +0000 (11:18 -0400)]
symbol_table: remove read_* methods

These methods of retrieving symbols haven't been in use since 2008 (at
least according to the date of the comment mentioning its abandonment).
Rather than having it slowly rot away, let's remove it. Git will
remember it for us if we ever need it back.

10 years agotapsets.cxx: remove unneeded forward decls
Jonathan Lebon [Thu, 1 May 2014 15:08:22 +0000 (11:08 -0400)]
tapsets.cxx: remove unneeded forward decls

10 years agomodule_info::get_symtab(): don't require base_query
Jonathan Lebon [Thu, 1 May 2014 14:50:32 +0000 (10:50 -0400)]
module_info::get_symtab(): don't require base_query

The get_symtab() function originally required the 'sess' and 'dw'
members of dwarf_query (see commit 5f0a03a6). Nowadays, this is no
longer required.

10 years agosetup_note_probe_entry(): strengthen note checks
Jonathan Lebon [Wed, 23 Apr 2014 15:45:12 +0000 (11:45 -0400)]
setup_note_probe_entry(): strengthen note checks

In dwflpp::iterate_over_notes(), we call back to
sdt_query::setup_note_probe_entry() for every note entry found,
regardless of section name or note name, which is fine.

However, on the callback side in setup_note_probe_entry(), we never
actually check that the section name or note name is correct; we only
check that the note entry type was what we expected.

This means that it could have been possible for setup_note_probe_entry()
to try to blindly parse data from a note entry which was not even a
stapsdt note, which could be dangerous.

This patch adds explicit checks that the section name and the note name
are what we expect.

10 years agoCorrectly return task pointers by making them unsigned longs.
Tetsuo Handa [Fri, 9 May 2014 14:35:26 +0000 (09:35 -0500)]
Correctly return task pointers by making them unsigned longs.

* tapset/linux/task.stp (task_current): Return task strucct pointer as an
  unsigned long.
  (pid2task): Ditto.

10 years agocallee.exp: simplify shlib testing by using -rpath
Jonathan Lebon [Thu, 8 May 2014 19:16:36 +0000 (15:16 -0400)]
callee.exp: simplify shlib testing by using -rpath

We previously started stap separately from the target process because we
needed to set the LD_LIBRARY_PATH variable for it to run properly.

However, if we use the linker option -rpath to add the current directory
to the library path (as is done also in unprivileged_myproc.exp), this
is no longer necessary, thus greatly simplifying the testcase.

10 years agoNormalize "aarch64" to "arm64" to match SystemTap's idea of arch
William Cohen [Wed, 7 May 2014 15:48:17 +0000 (11:48 -0400)]
Normalize "aarch64" to "arm64" to match SystemTap's idea of arch

10 years agoerror::pass2 - note --poison-cache as possible caching bug workaround
Frank Ch. Eigler [Wed, 7 May 2014 14:29:36 +0000 (10:29 -0400)]
error::pass2 - note --poison-cache as possible caching bug workaround

10 years agomodule cache hashing: add systemtap version string
Frank Ch. Eigler [Wed, 7 May 2014 13:52:57 +0000 (09:52 -0400)]
module cache hashing: add systemtap version string

* session.cxx (version_string): New method.
  (version): Call it.
* hash.cxx (get_base_hash): Ditto.

10 years agoFix tcl error in testsuite/systemtap.base/target_set.exp
David Smith [Tue, 6 May 2014 20:50:11 +0000 (15:50 -0500)]
Fix tcl error in testsuite/systemtap.base/target_set.exp

* testsuite/systemtap.base/target_set.exp: Fix tcl error on script compile
  failure by providing fail reason.

10 years agoFix PR16914 by adding more aarch64 defines.
David Smith [Tue, 6 May 2014 20:42:11 +0000 (15:42 -0500)]
Fix PR16914 by adding more aarch64 defines.

* runtime/linux/compat_unistd.h: Add '__NR_open' and '__NR_futimesat'
  defines for aarch64.

10 years agoFlip condition so proper default picked for STAP_NOP in testsuite/sys/sdt.h
William Cohen [Tue, 6 May 2014 20:01:54 +0000 (16:01 -0400)]
Flip condition so proper default picked for STAP_NOP in testsuite/sys/sdt.h

A number of the uprobes code failed to build on aarch64 because the
define for STAP_NOP would default to the unusual "nop 0".  The test
has been flipped around so the default is usual "nop" if there is
not something explict.

10 years agoFix PR16914 by adding aarch64 support to runtime/linux/compat_unistd.h.
David Smith [Tue, 6 May 2014 18:34:18 +0000 (13:34 -0500)]
Fix PR16914 by adding aarch64 support to runtime/linux/compat_unistd.h.

10 years agoFix nd_syscall ustat32 probes.
David Smith [Tue, 6 May 2014 17:26:31 +0000 (12:26 -0500)]
Fix nd_syscall ustat32 probes.

* tapset/linux/nd_syscalls2.stp: Combine the 2 nd_syscall.ustat32 probes
  into one.

10 years agoPR16913: aarch64: don't require prelink
Martin Cermak [Tue, 6 May 2014 13:53:05 +0000 (09:53 -0400)]
PR16913: aarch64: don't require prelink

10 years agoMake syscall.sendfile optional.
David Smith [Mon, 5 May 2014 21:33:19 +0000 (16:33 -0500)]
Make syscall.sendfile optional.

10 years agoFix more compile error when we don't have a dwarf unwinder.
David Smith [Mon, 5 May 2014 21:25:25 +0000 (16:25 -0500)]
Fix more compile error when we don't have a dwarf unwinder.

* runtime/linux/runtime.h: Fix compilation on systems without
  STP_USE_DWARF_UNWINDER defined.
* runtime/unwind/unwind.h: For systems without STP_USE_DWARF_UNWINDER
  defined, just define a dummy unwind_context structure along with the
  real unwind_cache structure.

10 years agoFix compile error when we don't have a dwarf unwinder in _stp_tack_kernel_print
David Smith [Mon, 5 May 2014 20:46:40 +0000 (15:46 -0500)]
Fix compile error when we don't have a dwarf unwinder in _stp_tack_kernel_print

* runtime/stack.c (_stp_stack_kernel_print): Use 'c->kregs' instead of
  nonexistent 'regs' variable. This code is only compiled when
  STP_USE_DWARF_UNWINDER isn't defined.

10 years agosystemtap.spec: make java/emacs conditional on usual >=f19-rhel7
Frank Ch. Eigler [Mon, 5 May 2014 16:59:15 +0000 (12:59 -0400)]
systemtap.spec: make java/emacs conditional on usual >=f19-rhel7

10 years agobeginnner's guide configury: compatibility with older publican
Frank Ch. Eigler [Mon, 5 May 2014 16:49:10 +0000 (12:49 -0400)]
beginnner's guide configury: compatibility with older publican

Older publican doesn't support the --pdftool=fop option.  It's too
much hassle to autoautoconf this (considering the existing fop-related
conditionals at the top level configure.ac), so instead here we try
publican with and then without the --pdftool=fop option.

10 years agoFixed PR6791 by adding ia64 dwarfless register access.
David Smith [Mon, 5 May 2014 15:20:14 +0000 (10:20 -0500)]
Fixed PR6791 by adding ia64 dwarfless register access.

* tapset/ia64/registers.stp: Add dwarfless register access.
* runtime/regs.c (__stp_sign_extend32): Ensure it is present for ia64
  compiles.
* tapset/linux/nd_syscalls2.stp: Add ia64 support.
* testsuite/systemtap.syscall/nd_syscall.exp: Remove kfail for ia64.
* testsuite/buildok/nd_syscalls-detailed.stp: Make accept4 and fork tests
  optional.
* testsuite/systemtap.examples/process/strace.tcl: Removed file that
  kfailed ia64 for lack of dwarfless register access.
* testsuite/systemtap.examples/process/thread-business.tcl: Ditto.

This page took 0.06198 seconds and 5 git commands to generate.