]> sourceware.org Git - systemtap.git/log
systemtap.git
10 years agoparse: let autocast peeks detect deeper pretty-prints
Josh Stone [Wed, 14 May 2014 01:24:01 +0000 (18:24 -0700)]
parse: let autocast peeks detect deeper pretty-prints

It already noticed "(expr)$" but "(expr)$$" should be valid too.

10 years agoUnify checks for target_symbol::comp_pretty_print
Josh Stone [Wed, 14 May 2014 01:21:57 +0000 (18:21 -0700)]
Unify checks for target_symbol::comp_pretty_print

The slightly-complicated check for comp_pretty_print was repeated in a few
places.  These are now done via target_symbol::pretty_print_depth() and
::check_pretty_print(), and the latter handles common lvalue asserts too.

10 years agoHide the details of target_symbol_setter_functioncalls
Josh Stone [Tue, 13 May 2014 23:55:55 +0000 (16:55 -0700)]
Hide the details of target_symbol_setter_functioncalls

This is a somewhat ugly detail of var_expanding_visitor for turning
lvalue assignments into functioncalls, and the manipulation of
target_symbol_setter_functioncalls was duplicated in several places.
Make that manipulation a new method, provide_lvalue_call, and hide that
field as private.

10 years agoResolve function refs added by autocast expansion
Josh Stone [Tue, 13 May 2014 23:28:25 +0000 (16:28 -0700)]
Resolve function refs added by autocast expansion

* elaborate.cxx(autocast_expanding_visitor::resolve_functioncall): New,
  run symresolution and functioncall_traversing to make sure functions
  added by autocasting are resolved.
  (autocast_expanding_visitor::visit_autocast_op): Use it.

10 years agoelaborate: Propagate type_details in limited cases
Josh Stone [Fri, 9 May 2014 23:46:21 +0000 (16:46 -0700)]
elaborate: Propagate type_details in limited cases

First there's a new exp_type_null which acts as an explicit marker,
"there be no detailed types here!"  Apply this to globals, and it may
also be used combining disparate type details to degrade to pe_long.

Then there are three expression cases that copy type details around:

* Ternary (?:) - if the true/false branches agree on type details, then
  the expression as a whole carries that type too.

* Plain assignment (=) - the right side propagates to the left, but not
  vice versa.  The assignment itself gets the type details too.

* Symbol to its declaration, and vice versa.  If the details of either
  disagree, then both are demoted to exp_type_null, so that can trickle
  back to all uses of that symbol.

10 years agoelaborate: turn referent asserts into exceptions
Josh Stone [Fri, 9 May 2014 23:45:24 +0000 (16:45 -0700)]
elaborate: turn referent asserts into exceptions

It's a bit nicer to throw an "internal error" semantic error message,
rather than assert and leave dirty tmp et al, especially while I'm working
on (and breaking) type resolution. :)

10 years agoStart expanding autocast during type resolution
Josh Stone [Wed, 7 May 2014 20:07:37 +0000 (13:07 -0700)]
Start expanding autocast during type resolution

There's no real propagation yet, so you can only try directly
transferred types so far, like:

  probe oneshot {
    println( (& @cast(task_current(), "task_struct")) ->pid )
  }

So that captures the task_current() address *with* the type, and then
the "->pid" is working as a separate expression to dereference it.

* elaborate.cxx (semantic_pass_symbols): Move build_no_more from here...
  (semantic_pass): ... to here, well after all types are resolved.
  (autocast_expanding_visitor): New, call on the type_details to expand
  autocast_op into a functioncall, like any $var/@cast would.
  (semantic_pass_types): Use autocast_expanding_visitor whenever the
  normal type resolution indicated there might be something to do.
  (typeresolution_info::visit_autocast_op):  Don't fail immediately;
  just visit the operand, and note when type_details show up.
* tapsets.cxx (exp_type_dwarf::expand): Expand that autocast_op!
  (dwarf_autocast_expanding_visitor): Removed in favor of just having
  autocast_expanding_visitor call exp_type_ptr->expand().

10 years agoelaborate: mark functioncall type errors on the argument
Josh Stone [Tue, 6 May 2014 23:51:22 +0000 (16:51 -0700)]
elaborate: mark functioncall type errors on the argument

10 years agoSave type details in expressions
Josh Stone [Tue, 6 May 2014 22:28:21 +0000 (15:28 -0700)]
Save type details in expressions

* dwflpp.cxx (dwflpp::translate_final_fetch_or_store): Return the final
  typedie in memory provided by the caller.
  (dwflpp::literal_stmt_for_local): Pass through the final typedie.
  (dwflpp::literal_stmt_for_return): Ditto.
  (dwflpp::literal_stmt_for_pointer): Ditto.
* staptree.h (exp_type_details): New, generic type extras.
  (expression): Store optional exp_type_details.
  (expression_visitor): New, simple visitor to ease operating on all
  expressions equally.
* staptree.cxx (expression_visitor): Implement.
* tapsets.cxx (exp_type_dwarf): New DWARF type details.
  (synthetic_embedded_deref_call): Set the DWARF type on the resulting
  functioncall expression.
* elaborate.cxx (expression_build_no_more_visitor): New, clear extra
  type details from all expressions.
  (build_no_more): Clear expression types before calling build_no_more
  in the builders, so dwflpp lives long enough.

10 years agoFlatten dwflpp in add_unwindsym_ldd
Josh Stone [Sat, 3 May 2014 00:50:00 +0000 (17:50 -0700)]
Flatten dwflpp in add_unwindsym_ldd

There's no reason to new/delete this dwflpp instance; just create it
directly on the stack with no worries.

10 years agoMerge branch 'master' into autocast
Josh Stone [Fri, 2 May 2014 22:08:19 +0000 (15:08 -0700)]
Merge branch 'master' into autocast

10 years agoAdd test coverage for rhbz1027459
Martin Cermak [Fri, 2 May 2014 16:39:35 +0000 (18:39 +0200)]
Add test coverage for rhbz1027459

10 years agoMake the syscall.accept4 buildok test optional.
David Smith [Fri, 2 May 2014 14:49:36 +0000 (09:49 -0500)]
Make the syscall.accept4 buildok test optional.

10 years agoPR16894: new wordexp.exp to check -c COMMAND handling
Jonathan Lebon [Fri, 2 May 2014 14:19:06 +0000 (10:19 -0400)]
PR16894: new wordexp.exp to check -c COMMAND handling

This simple testcase checks that invalid -c COMMANDs for unspecified
process probes are properly handled (e.g. prior to commit eb94071, we
segfaulted).

10 years agoPR16894: give better errors re. -c commands
Jonathan Lebon [Fri, 2 May 2014 13:43:07 +0000 (09:43 -0400)]
PR16894: give better errors re. -c commands

This patch allows SystemTap to make the distinction when reporting an
error between no -c command given, and a given but invalid -c command.

- man/stapprobes.3stap: introduce 'unspecified' verbiage and clear up
  restrictions on unspecified process probes
- session.cxx: cmd_file() now throws errors upon wordexp() failures
- tapsets.cxx: (1) handle possibility of error from cmd_file() and chain
                   with own error if it occurs
               (2) separately handle no -c command / empty file
               (3) introduce 'unspecified' verbiage in error messages
                   and point to [man stapprobes]
- tapset-perfmon.cxx: ditto

10 years agoPR16716 partial fix: Fix types in syscall.clock_{getres,gettime,nanosleep}.
David Smith [Fri, 2 May 2014 13:20:21 +0000 (08:20 -0500)]
PR16716 partial fix: Fix types in syscall.clock_{getres,gettime,nanosleep}.

* tapset/linux/syscalls.stp (syscall.clock_getres): Fix compat support and
  types.
  (syscall.clock_gettime): Fix compat support and add 'tp_uaddr' variable.
  (syscall.clock_nanosleep): Fix types. Add 'clk_id', 'clk_id_str',
  'req_uaddr', and 'rem_uaddr' variables. Deprecate 'flag_str' in favor of
  'flags_str'.
  (syscall.compat_clock_nanosleep): Ditto.
  (syscall.clock_settime): Fix types.
* tapset/linux/nd_syscalls.stp (nd_syscall.clock_getres): Fix compat support.
  (nd_syscall.clock_gettime): Fix compat support and add 'tp_uaddr' variable.
  (nd_syscall.clock_nanosleep): Add 'clk_id', 'clk_id_str', 'req_uaddr',
  and 'rem_uaddr' variables. Deprecate 'flag_str' in favor of 'flags_str'.
  (nd_syscall.compat_clock_nanosleep): Ditto.
* tapset/linux/aux_syscalls.stp (_struct_timeval_u): Instead of "UNKNOWN",
  print the failing address.
  (_struct_compat_timeval_u): Ditto.
  (_struct_timezone_u): Ditto.
  (_struct_timespec_u): Ditto.
  (_struct_compat_timespec_u): Ditto.
  (_stp_clock_nanosleep_flags_str): New function.
  (_get_wc_str): Convert to use _stp_lookup_str().
* testsuite/buildok/aux_syscalls-embedded.stp: Add test for
  _stp_clock_nanosleep_flags_str().
* testsuite/buildok/syscalls-detailed.stp: Add tests for new variables.
* testsuite/buildok/nd_syscalls-detailed.stp: Ditto.
* testsuite/systemtap.syscall/clock.c: Add more error testing.
* testsuite/systemtap.syscall/futimes.c: Update expected output.
* testsuite/systemtap.syscall/net1.c: Ditto.
* testsuite/systemtap.syscall/recvmmsg.c: Ditto.
* testsuite/systemtap.syscall/select.c: Ditto.
* testsuite/systemtap.syscall/inotify.c: Fix "COVERAGE" comment formatting.

10 years agocmd_file(): properly free the words array
Jonathan Lebon [Thu, 1 May 2014 20:43:09 +0000 (16:43 -0400)]
cmd_file(): properly free the words array

We should free it regardless if rc > 0.

10 years agocmd_file(): only free words if succeeded
Jonathan Lebon [Thu, 1 May 2014 20:31:59 +0000 (16:31 -0400)]
cmd_file(): only free words if succeeded

Otherwise, we may accidentally summon Cthulhu.

10 years agofix typo in changelog (bad date)
Martin Cermak [Thu, 1 May 2014 06:42:41 +0000 (08:42 +0200)]
fix typo in changelog (bad date)

10 years agoversion bump and autoreconf
Jonathan Lebon [Wed, 30 Apr 2014 18:56:22 +0000 (14:56 -0400)]
version bump and autoreconf

10 years agoAdd MMPAP_SYSCALL_NO and other defines in runtime/syscall.h for aarch64
William Cohen [Wed, 30 Apr 2014 18:52:42 +0000 (14:52 -0400)]
Add MMPAP_SYSCALL_NO and other defines in runtime/syscall.h for aarch64

10 years agoPR16716 partial fix: Fix types in syscall.{compat_select,unlinkat}
David Smith [Wed, 30 Apr 2014 18:44:51 +0000 (13:44 -0500)]
PR16716 partial fix: Fix types in syscall.{compat_select,unlinkat}

* tapset/linux/syscalls2.stp: Fix types in syscall.{compat_select,unlinkat}.
* tapset/linux/nd_syscalls2.stp: Fix types in nd_syscall.unlinkat.

10 years agosystemtap.spec: pre-release log and bump release-2.5
Jonathan Lebon [Wed, 30 Apr 2014 18:16:20 +0000 (14:16 -0400)]
systemtap.spec: pre-release log and bump

10 years agoPR16716 partial fix: Fix types in syscall.{linkat,readlink,readlinkat}.
David Smith [Wed, 30 Apr 2014 15:06:39 +0000 (10:06 -0500)]
PR16716 partial fix: Fix types in syscall.{linkat,readlink,readlinkat}.

* tapset/linux/syscalls.stp: Fixed types in syscall.linkat.
* tapset/linux/syscalls2.stp: Fixed types in
  syscall.{readlink,readlinkat).
* testsuite/systemtap.syscall/link.c: Added tests.

10 years agoPR16716 partial fix: Fix types in syscall.{open,openat}.
David Smith [Tue, 29 Apr 2014 20:56:12 +0000 (15:56 -0500)]
PR16716 partial fix: Fix types in syscall.{open,openat}.

* tapset/linux/syscalls2.stp: Fixed types in syscall.{open,openat}.
* tapset/linux/nd_syscalls2.stp: Ditto.
* testsuite/systemtap.syscall/openclose.c: Added tests.

10 years agopyexample.stp: split into py2example.stp and py3example.stp
Jonathan Lebon [Tue, 29 Apr 2014 18:32:48 +0000 (14:32 -0400)]
pyexample.stp: split into py2example.stp and py3example.stp

In order to be compatible on systems with only python2 installed (or
only python3 installed), we split the pyexample.stp script into two
versions.

We also add a test_support line in the meta files so that check.exp
marks them as untested if either/both python versions are not present.

In the tapsets, we make py2/3 aliases that the scripts then use, but
also create general ones that refer to both the 2 and 3 variants.

The python3 version requires -g --suppress-time-limits, otherwise it's
very likely that a 'overhead exceeded' ERROR.

- py2example.stp: python 2 version of the script
- py2example.meta: meta file for python 2 version, with test_support
  check for python 2 specific SDT markers
- py3example.stp: python 3 version of the script
- py3example.meta: meta file for python 3 version, with test_support
  check for python 3 specific SDT markers
- tapset/python2.stp: make python2 specific aliases and alias the
  general ones to that
- tapset/python3.stp: make python3 specific aliases and alias the
  general ones to that

10 years agofix --suppress-time-limits usage typo too
Frank Ch. Eigler [Tue, 29 Apr 2014 20:25:04 +0000 (16:25 -0400)]
fix --suppress-time-limits usage typo too

10 years agoFix typo for suppress-time-limits in runtime headers
Lukas Berk [Tue, 29 Apr 2014 20:17:36 +0000 (16:17 -0400)]
Fix typo for suppress-time-limits in runtime headers

*man/stap.1   - update man page to properly list macro's
  and polarity of statement
*runtime/common_probe_context.h - fix STAP_SUPPRESS_TIME_LIMITS_ENABLE typo
*runtime/runtime_defines.h  - ditto

10 years agoFix PR16884 by updating [nd_]syscall.compat_sys_shmctl.
David Smith [Tue, 29 Apr 2014 19:31:09 +0000 (14:31 -0500)]
Fix PR16884 by updating [nd_]syscall.compat_sys_shmctl.

* tapset/linux/syscalls2.stp (syscall.compat_sys_shmctl): Make supported
  variables match up with syscall.shmctl.
* tapset/linux/nd_syscalls2.stp (nd_syscall.compat_sys_shmctl): Ditto.
* testsuite/buildok/syscalls2-detailed.stp: Updated
  syscall.compat_sys_shmctl test.
* testsuite/buildok/nd_syscalls2-detailed.stp: Ditto.
* NEWS: Mention variable deprecations.

10 years agoPR16716 partial fix: Fix types in syscall.{chmod,fchmodat}.
David Smith [Tue, 29 Apr 2014 18:24:33 +0000 (13:24 -0500)]
PR16716 partial fix: Fix types in syscall.{chmod,fchmodat}.

* tapset/linux/syscalls.stp: Fixed types in syscall.{chmod,fchmodat}.
* tapset/linux/nd_syscalls.stp: Removed old comment.

10 years agoPR16716 partial fix: Fix types in syscall.sysfs.
David Smith [Tue, 29 Apr 2014 17:25:51 +0000 (12:25 -0500)]
PR16716 partial fix: Fix types in syscall.sysfs.

* tapset/linux/syscalls2.stp: Fixed types in syscall.sysfs. Improved
  output formatting.
* tapset/linux/syscalls2.stp: Improved output formatting in
  nd_syscall.sysfs.
* testsuite/systemtap.syscall/sysfs.c: New test case.

10 years agoNEWS: typos in --dump-functions blurb
Jonathan Lebon [Tue, 29 Apr 2014 16:15:59 +0000 (12:15 -0400)]
NEWS: typos in --dump-functions blurb

10 years agoPR16716 partial fix: Fix types in syscall.{dup,dup2,dup3}.
David Smith [Tue, 29 Apr 2014 15:54:27 +0000 (10:54 -0500)]
PR16716 partial fix: Fix types in syscall.{dup,dup2,dup3}.

* tapset/linux/syscalls.stp: Fixed types in syscall.{dup,dup2,dup3}.
* tapset/linux/aux_syscalls.stp (_dup3_flag_str): Just use the open mode
  flags list.
* testsuite/systemtap.syscall/dup.c: Add more tests.

10 years agoupdate copyrights
Jonathan Lebon [Tue, 29 Apr 2014 15:42:41 +0000 (11:42 -0400)]
update copyrights

Update the Red Hat copyrights using:

scripts/update-copyright "Red Hat" redhat.com

10 years agonew helper script scripts/update-copyrights
Jonathan Lebon [Thu, 24 Apr 2014 16:53:06 +0000 (12:53 -0400)]
new helper script scripts/update-copyrights

This script looks at all the tracked files in the tree that are
copyrighted by <company> and updates the copyright year if they've been
modified this year by a user at <domain>.

Usage example: scripts/update-copyrights "Red Hat" redhat.com

10 years agoBZ1092147: ppc64le: don't require prelink
Brent Baude [Mon, 28 Apr 2014 20:39:19 +0000 (16:39 -0400)]
BZ1092147: ppc64le: don't require prelink

10 years agoPR16406: fix build-id false mismatch for userspace-program vs. kernel-module
Frank Ch. Eigler [Mon, 28 Apr 2014 19:55:56 +0000 (15:55 -0400)]
PR16406: fix build-id false mismatch for userspace-program vs. kernel-module

The translate.cxx-emitted _stp_module array has two name fields, .name
and .path, which were ambiguous as to whether each was a kernel module
or a userspace executable.  The .name string was just the basename,
and it was used in the near-startup-time _stp_module_check().  If the
userspace program matched a kernel module, bangorama (a false build-id
mismatch error at startup) .. unless --all-modules was given, in
which case the kernel modules won the duplication race.

The correction has a couple of parts.  First, the _stp_module .name is
made the same as the fully-/-qualified .path for userspace programs.
Second, _stp_module_check checks all kernel+module bits it knows
about, not just the first one.  Hand-tested with a variety of runs,
and with the *myproc.exp test, which is now forced to try to conflict.

* translate.cxx (dump_unwindsym_ctx): Make userspace _stp_module->name's
  fully-/-qualified.
* runtime/sym.c (_stp_module_check): Check all and only kernel-space
  build-ids.  Filter by _stp_module->name.
  (stp_kmodule_check, stp_usermodule_check): WARN_ON wrong-polarity
  incoming module/path name.
* testsuite/.../unprivileged_myproc.exp: Run 'modprobe loop'

10 years agounprivileged_myproc.exp: only KFAIL .callee(string)
Jonathan Lebon [Mon, 28 Apr 2014 21:00:39 +0000 (17:00 -0400)]
unprivileged_myproc.exp: only KFAIL .callee(string)

This is a followup to the followup commit 9d4241d. The .callees and
.callees(number) probes will always work because they will match ibar().
Only the .callee(string), which is turned to .callee("tbar"), should be
KFAILed, since it will definitely fail for GCC < 4.7.

10 years agounprivileged_myproc.exp: adjust loop.c
Jonathan Lebon [Mon, 28 Apr 2014 20:24:32 +0000 (16:24 -0400)]
unprivileged_myproc.exp: adjust loop.c

The process("./loop").function("ibar").inline probe was failing in this
test because of a bug in GCC (RHBZ1092144). In a nutshell, GCC unrolls
the loop once, resulting in ibar() being called twice: once outside the
loop and once inside. However, only the former gets a corresponding
inlined_subroutine DIE. As a result, the SystemTap probe only fires once
at startup rather than at every iteration.

This patch takes out the if-expression in the loop so that GCC doesn't
need to do any unrolling and we end up with two inlined_subroutine DIEs.

10 years agotask.stp: fix typo in task_rlimit() description
Jonathan Lebon [Mon, 28 Apr 2014 14:52:58 +0000 (10:52 -0400)]
task.stp: fix typo in task_rlimit() description

10 years agoPR16716 partial fix: Fix types in syscall.{inotify_add_watch,inotify_rm_watch}
David Smith [Mon, 28 Apr 2014 20:32:32 +0000 (15:32 -0500)]
PR16716 partial fix: Fix types in syscall.{inotify_add_watch,inotify_rm_watch}

* tapset/linux/syscalls.stp: Fix types in
  syscall.{inotify_add_watch,inotify_rm_watch}.
* tapset/linux/nd_syscalls.stp: Fix types in nd_syscall.inotify_rm_watch.
* tapset/linux/aux_syscalls.stp (_inotify_watch_mask_str): Updated and
  improved to display unknown values.
  (_inotify_init1_flag_str): Ditto.
* testsuite/systemtap.syscall/inotify.c: Add more tests.

10 years agoPR16716 partial fix: Fix types in syscall.{close,creat}.
David Smith [Mon, 28 Apr 2014 18:22:26 +0000 (13:22 -0500)]
PR16716 partial fix: Fix types in syscall.{close,creat}.

* tapset/linux/syscalls.stp: Fixed types in syscall.{close,creat}.
* tapset/linux/syscalls.stp: Fixed types in syscall.creat.
* tapset/linux/aux_syscalls.stp (_sys_open_flag_str): Updated and improved
  to display unknown values.
* testsuite/systemtap.syscall/openclose.c: Added more tests.

10 years agoAdd java backtracing and self unwinding to NEWS
Lukas Berk [Mon, 28 Apr 2014 15:26:24 +0000 (11:26 -0400)]
Add java backtracing and self unwinding to NEWS

*NEWS - add news blurbs

10 years agosyscall testsuite: accept4 SOCK_CLOEXEC
Frank Ch. Eigler [Mon, 28 Apr 2014 14:55:19 +0000 (10:55 -0400)]
syscall testsuite: accept4 SOCK_CLOEXEC

Add a conditional #define for RHEL5 era kernels.

10 years agosyscall testsuite test.tcl: use staprun -R for reused module
Frank Ch. Eigler [Mon, 28 Apr 2014 14:53:30 +0000 (10:53 -0400)]
syscall testsuite test.tcl: use staprun -R for reused module

It's been occasionally observed that a single syscall test case can
leave the [nd_]syscall.ko module loaded, which blocks further testing.
staprun -R can rename the .ko, to avoid the chain failure.

10 years agounprivileged_myproc testsuite: mark .callees tests kfail for gcc < 4.7
Frank Ch. Eigler [Sun, 27 Apr 2014 22:20:26 +0000 (18:20 -0400)]
unprivileged_myproc testsuite: mark .callees tests kfail for gcc < 4.7

This is a followup to commit 081e9217c8d0, this time on rhel6.

10 years agoruntime rhel6: avoid kernel crash with buildid checks for process(PID#) probes
Frank Ch. Eigler [Sun, 27 Apr 2014 21:16:29 +0000 (17:16 -0400)]
runtime rhel6: avoid kernel crash with buildid checks for process(PID#) probes

Crash observed on rhel6 (utrace -> process(NUM).statement(ADDR).absolute probes)
in unprivileged_myproc.exp test case.

* runtime/linux/uprobes-common.c (stap_uprobe_change_plus): Don't try to check
  process against _stp_module entry, if we don't even have a process pathname.
* runtime/sym.c (_stp_usermodule_check): WARN_ON incoming null pathname.

10 years agotapset: Deal with missing fs in i386/register.stp
Josh Stone [Sat, 26 Apr 2014 00:17:03 +0000 (17:17 -0700)]
tapset: Deal with missing fs in i386/register.stp

Older kernels didn't have fs in pt_regs, so the following _reg_offsets
were incorrect.  Make fs conditional on stapconf, and use a counter to
more easily keep track of the differences.

10 years agoruntime: Fix synthesized regs on ix86
Josh Stone [Sat, 26 Apr 2014 00:12:18 +0000 (17:12 -0700)]
runtime: Fix synthesized regs on ix86

The ix86 register names and offsets in the inline asm of
arch_unw_init_frame_info are now fixed, including stapconf to deal with
fs/xfs/gs changes over time.

x86_64 is updated to match the same style of access.

10 years agobuildrun: Don't bother filtering -fno-asynchonous-unwind-tables
Josh Stone [Fri, 25 Apr 2014 23:05:50 +0000 (16:05 -0700)]
buildrun: Don't bother filtering -fno-asynchonous-unwind-tables

Our -fasynchronous-unwind-tables in EXTRA_CFLAGS takes precedence just
fine over kbuild's -fno-asynchonous-unwind-tables setting.

10 years agoruntime: Extend sect_attrs use back to 2.6.11
Josh Stone [Fri, 25 Apr 2014 23:01:06 +0000 (16:01 -0700)]
runtime: Extend sect_attrs use back to 2.6.11

Only 2.6.19+ has sect_attrs->nsections, but from 2.6.11 one can count
where the NULL terminator is on sect_attrs->grp.attrs.

10 years agoruntime: Initialize _stp_module_self from THIS_MODULE
Josh Stone [Fri, 25 Apr 2014 22:25:10 +0000 (15:25 -0700)]
runtime: Initialize _stp_module_self from THIS_MODULE

By iterating the sections of THIS_MODULE directly, rather than from
STP_RELOCATION messages, we can more safely contain the eh_frame length
determination.  As an upper bound, the length can be no more than the
distance to the next section in memory.  Then when we walk fde records,
we can sanity check that the offsets make sense.

10 years agosyscalls tapset: adapt to absence of uselib(2) syscall
Frank Ch. Eigler [Fri, 25 Apr 2014 23:14:39 +0000 (19:14 -0400)]
syscalls tapset: adapt to absence of uselib(2) syscall

LKML commit 69369a70 made it possible to configure a kernel without
uselib(2) support.  Mark these tapset aliases optional so syscall.*
type probing works on !CONFIG_USELIB kernels.

10 years agoPR6961 cont'd: add a terminating 0x00000000 word to module .ko .eh_frame
Frank Ch. Eigler [Fri, 25 Apr 2014 19:46:06 +0000 (15:46 -0400)]
PR6961 cont'd: add a terminating 0x00000000 word to module .ko .eh_frame

This is necessary because now we parse our own module's .eh_frame
lightly at startup time, and must reliably find the end of the
entries.  (We don't get the crtend.o goodies from userspace.)
We also reject 8-byte extended-length cases, should they ever come up.

* session.* (op_create_auxiliary): Make possible to designate aux files
  as trailers (coming in after the main generated .o file) instead of before.
* buildrun.cxx (compile_pass): Add s.auxiliary_output trailers.
* translate.cxx (emit_symbol_data_done): T-800 returns from the future.
* runtime/sym.c (_stp_kmodule_update_address): Reject 8-byte extended-length
  fields in .eh_frame.

10 years agostmt_inlines.exp: mark untested for older GCCs
Jonathan Lebon [Fri, 25 Apr 2014 19:35:27 +0000 (15:35 -0400)]
stmt_inlines.exp: mark untested for older GCCs

The RHEL5 GCC does not output the necessary debuginfo in order to do
this test, so mark it untested.

10 years agostatement.exp: adapt for RHEL5
Jonathan Lebon [Fri, 25 Apr 2014 19:00:07 +0000 (15:00 -0400)]
statement.exp: adapt for RHEL5

In GCC < 4.4, the decl_line is the line holding the opening brace, not the
line with the identifier. Since the decl_line is the basis for relative
line number statement probes, this discrepancy was causing statement.exp
to have some RELATIVE subtest failures.

Add another empty line so that main@statement.c+4 always falls on an
empty line, regardless of which is the decl_line.

10 years agotestsuite: adapt to RHEL5
Jonathan Lebon [Fri, 25 Apr 2014 16:01:52 +0000 (12:01 -0400)]
testsuite: adapt to RHEL5

- callee.exp: set as untested if GCC or elfutils too old
- dump_functions.exp: RHEL5 exec doesn't support -ignorestderr
- dump_probe_aliases.exp: ditto

10 years agolib/systemtap.exp: add strverscmp proc
Jonathan Lebon [Fri, 25 Apr 2014 15:59:31 +0000 (11:59 -0400)]
lib/systemtap.exp: add strverscmp proc

There are times when we want to set tests as 'untested' if certain
component versions are not met. The new proc strverscmp will serve well
there.

10 years agolib/systemtap.exp: get better GCC and elfutils version info
Jonathan Lebon [Fri, 25 Apr 2014 15:53:12 +0000 (11:53 -0400)]
lib/systemtap.exp: get better GCC and elfutils version info

Rather than having the global GCC_Version hold the whole
"gcc --version | head -n 1" string, create a new GCC_FullVersion for
that and make GCC_Version only hold the raw version (e.g. "4.1.2").
This will be useful later on for version comparison.

Create new ELF_Version which holds the elfutils library version against
which SystemTap was compiled.

10 years agosyscall testsuite: syslog.c: make nonblocking
Frank Ch. Eigler [Thu, 24 Apr 2014 20:33:43 +0000 (16:33 -0400)]
syscall testsuite: syslog.c: make nonblocking

We need to use SYSLOG_ACTION_READ_ALL, not SYSLOG_ACTION_READ,
to ensure our test program quits quickly.

10 years agoChange how we assign _stp_module_self name/path to use THIS_MODULE
Lukas Berk [Thu, 24 Apr 2014 20:22:18 +0000 (16:22 -0400)]
Change how we assign _stp_module_self name/path to use THIS_MODULE

*runtime/sym.c - change _stp_module_self name/path member variables to
         be assigned using THIS_MODULE->name instead of strcpy

10 years agoMerge branch 'master' of ssh://sourceware.org/git/systemtap
David Smith [Wed, 23 Apr 2014 21:00:58 +0000 (16:00 -0500)]
Merge branch 'master' of ssh://sourceware.org/git/systemtap

10 years agoPR16716partial fix: Fix types in syscall.{fchmod,fchown,fchownat}.
David Smith [Wed, 23 Apr 2014 21:00:31 +0000 (16:00 -0500)]
PR16716partial fix: Fix types in syscall.{fchmod,fchown,fchownat}.

*tapset/linux/syscalls.stp: Fix types in syscall.{fchmod,fchown,fchownat}.
*testsuite/systemtap.syscall/chmod.c: Added more tests.

10 years agoMerge branch 'jistone/pr16844'
Josh Stone [Wed, 23 Apr 2014 20:36:22 +0000 (13:36 -0700)]
Merge branch 'jistone/pr16844'

10 years agoMerge branch 'master' of ssh://sourceware.org/git/systemtap into lberk/backtrace
Lukas Berk [Wed, 23 Apr 2014 20:18:58 +0000 (16:18 -0400)]
Merge branch 'master' of ssh://sourceware.org/git/systemtap into lberk/backtrace

10 years agopr16806.exp: new testcase
Jonathan Lebon [Wed, 23 Apr 2014 19:48:37 +0000 (15:48 -0400)]
pr16806.exp: new testcase

- pr16806.exp: new testcase to verify proper tracepoint unregistration
  and utrace shutdown.
- loop.c: don't call usleep() if NOSLEEP is defined. This is used by
  pr16806.exp, which requires it to make conditions more racy.
- unprivileged_myproc.exp: simplify path to C file.
- unprivileged_probes.exp: ditto.

10 years agostmt_inlines.exp: rename cleanup proc
Jonathan Lebon [Wed, 23 Apr 2014 18:55:13 +0000 (14:55 -0400)]
stmt_inlines.exp: rename cleanup proc

There is already a proc named cleanup in lib/systemtap.exp. Rename the
one in stmt_inlines.exp so we don't accidentally redefine it.

10 years agoFix typo in systemtap.spec file rpm scriptlet
Lukas Berk [Wed, 23 Apr 2014 19:59:03 +0000 (15:59 -0400)]
Fix typo in systemtap.spec file rpm scriptlet

*systemtap.spec - switch directory check condition to use '${archdir}'
  variable instead of %{archdir}

10 years agoPR16716 partial fix: Better types passed to _dfd_str() from syscall probes.
David Smith [Wed, 23 Apr 2014 19:37:51 +0000 (14:37 -0500)]
PR16716 partial fix: Better types passed to _dfd_str() from syscall probes.

* tapset/linux/syscalls.stp: Fixed calls to _dfd_str() in
  facessat, fchownat, futimesat, compat_futimesat, and linkat syscall
  probes.
  (syscall.futimesat): Added 'dirfd_str' variable. Fixed nesting.
  (syscall.compat_futimesat): Ditto.
* tapset/linux/nd_syscalls.stp (nd_syscall.futimesat): Added 'dirfd_str'
  variable. Fixed nesting.
  (nd_syscall.compat_futimesat): Ditto.
* tapset/linux/syscalls2.stp: Fixed calls to _dfd_str() in openat,
  readlinkat, renameat, symlinkat, unlinkat, and utimensat syscall
  probes.
  (syscall.openat): Added 'dfd' and 'dfd_str' variables.
  (syscall.readlinkat): Added 'dfd_str' variable.
  (syscall.utimensat): Added 'dfd', 'dfd_str', 'filename_uaddr',
  'filename', 'tsp_uaddr', 'flags', and 'flags_str' variables.
  (syscall.compat_utimensat): Ditto.
* tapset/linux/nd_syscalls2.stp (nd_syscall.openat): Added 'dfd' and
  'dfd_str' variables.
  (nd_syscall.readlinkat): Added 'dfd_str' variable.
  (nd_syscall.utimensat): Added 'dfd', 'dfd_str', 'filename_uaddr',
  'filename', 'tsp_uaddr', 'flags', and 'flags_str' variables.
  (nd_syscall.compat_utimensat): Ditto.
* tapset/linux/aux_syscalls.stp (_dfd_str): Improve to not ignore
  compatibility mode wrong value (instead, values passed in were fixed).
* runtime/linux/compat_unistd.h: Add '__NR_compat_futimesat' define.
* testsuite/systemtap.syscall/alarm.c: Add more tests.
* testsuite/systemtap.syscall/futimes.c: Ditto.
* testsuite/buildok/syscalls-detailed.stp: Added basic test for new
  variables.
* testsuite/buildok/syscalls2-detailed.stp: Ditto.
* testsuite/buildok/nd_syscalls-detailed.stp: Ditto.
* testsuite/buildok/nd_syscalls2-detailed.stp: Ditto.

10 years agoPR16844: Direct messages to _stp_dbug
Josh Stone [Wed, 23 Apr 2014 00:14:36 +0000 (17:14 -0700)]
PR16844: Direct messages to _stp_dbug

Adds dbug_tp(), with a new DEBUG_TRACEPOINTS flag to control verbosity.
All printk and WARNs are removed in favor of dbug_tp.

10 years agoPR16844: Use _stp_kmalloc_gfp and _stp_kfree
Josh Stone [Tue, 22 Apr 2014 21:27:26 +0000 (14:27 -0700)]
PR16844: Use _stp_kmalloc_gfp and _stp_kfree

10 years agoPR16844: Initial adaptation for kernel 3.15 tracepoints
Josh Stone [Tue, 22 Apr 2014 16:07:42 +0000 (09:07 -0700)]
PR16844: Initial adaptation for kernel 3.15 tracepoints

This hides the tracepoint version adaptations in stp_tracepoint.h.  For
older kernel versions, that header is sufficient.  With 3.15 we have to
do extra bookkeeping ourselves, so it pulls in stp_tracepoints.c, which
is borrowed almost verbatim from lttng.  It seems to work well, but we
can still decide if we want to take a different approach.

10 years agotestsuite: don't mandate presence of -m32/-m64 libc/stdc++ devel
Frank Ch. Eigler [Tue, 22 Apr 2014 21:11:00 +0000 (17:11 -0400)]
testsuite: don't mandate presence of -m32/-m64 libc/stdc++ devel

Some distros (I'm looking at you, Ubuntu 14) don't make it easy to
build several permutations of -m32 / -m64 binaries.

10 years agoCorrect the architecture for the libHelper_ppc64.so name
Lukas Berk [Tue, 22 Apr 2014 16:36:57 +0000 (12:36 -0400)]
Correct the architecture for the libHelper_ppc64.so name

*java/Makefile.am - override the java_arch name (powerpc64) with ppc64
*java/Makefile.in - generated from an autoreconf

10 years agostmt_inlines.exp: testcase for statement probes in inlines
Jonathan Lebon [Tue, 22 Apr 2014 16:16:49 +0000 (12:16 -0400)]
stmt_inlines.exp: testcase for statement probes in inlines

This testcase verifies our ability to place probes on statements in
functions that have been inlined multiple times.

10 years agostatement.exp and stmt_rel.exp: a few tweaks
Jonathan Lebon [Fri, 18 Apr 2014 04:49:20 +0000 (00:49 -0400)]
statement.exp and stmt_rel.exp: a few tweaks

- stmt_rel.exp: small fix
- statement.exp: test a few more things

10 years agocallee.exp: fix inline compilation
Jonathan Lebon [Tue, 22 Apr 2014 15:08:30 +0000 (11:08 -0400)]
callee.exp: fix inline compilation

GCC won't emit a warning if we also use the inline keyword.

10 years agostapprobes.3stap: give @defined() hint for $return
Jonathan Lebon [Tue, 22 Apr 2014 14:49:14 +0000 (10:49 -0400)]
stapprobes.3stap: give @defined() hint for $return

10 years agoPR16716 partial fix: Better types in 'syscall.{access,faccessat}'.
David Smith [Tue, 22 Apr 2014 15:32:05 +0000 (10:32 -0500)]
PR16716 partial fix: Better types in 'syscall.{access,faccessat}'.

* tapset/linux/syscalls.stp: Fix 'mode' variable type in
  'syscall.{access,faccessat}'.
* tapset/linux/aux_syscall.stp: Convert to use _stp_lookup_or_str().
* testsuite/systemtap.syscall/access.c (main): Add tests.

10 years agosession.cxx: print found System.map file name for report, not first one tried
Frank Ch. Eigler [Tue, 22 Apr 2014 15:06:50 +0000 (11:06 -0400)]
session.cxx: print found System.map file name for report, not first one tried

10 years agosession.cxx: also print kernel_functions[] sizes at -vvv verbosity
Frank Ch. Eigler [Tue, 22 Apr 2014 15:04:54 +0000 (11:04 -0400)]
session.cxx: also print kernel_functions[] sizes at -vvv verbosity

10 years agoStapbm uses netstat, require it for runtime-java
Lukas Berk [Tue, 22 Apr 2014 14:32:36 +0000 (10:32 -0400)]
Stapbm uses netstat, require it for runtime-java

*systemtap.spec - require net-tools for netstat use in stapbm/runtime-java

10 years agoMerge branch 'master' of ssh://sourceware.org/git/systemtap
David Smith [Tue, 22 Apr 2014 13:43:26 +0000 (08:43 -0500)]
Merge branch 'master' of ssh://sourceware.org/git/systemtap

10 years agoPR16716 partial fix: Better types in 'syscall.{swapon,syslog}'.
David Smith [Tue, 22 Apr 2014 13:42:59 +0000 (08:42 -0500)]
PR16716 partial fix: Better types in 'syscall.{swapon,syslog}'.

* tapset/linux/syscalls2.stp: Fix types in 'syscall.syslog' and
  'syscall.swapon'. In 'syscall.swapon', decode the flags in the new
  'swapflags' variable.
* tapset/linux/nd_syscalls2.stp: In 'nd_syscall.swapon', decode the flags
  in the new 'swapflags' variable.
* tapset/linux/aux_syscalls.stp (_swapon_flags_str): New function.
* tapset/uconversions.stp (user_string_n2_quoted): If we're in a compat
  task, when printing the pointer value as a number, don't expand it to
  64-bits.
* testsuite/systemtap.syscall/swap.c: Added more tests.
* testsuite/systemtap.syscall/syslog.c: New test case.
* testsuite/buildok/syscalls2-detailed.stp: Added new 'swapflags' variable.
* testsuite/buildok/nd_syscalls2-detailed.stp: Ditto.

10 years agoAdd python support tapset example.
Stan Cox [Tue, 22 Apr 2014 13:37:16 +0000 (09:37 -0400)]
Add python support tapset example.

* systemtap.examples/general/tapset/(python2.stp,python3.stp,python.stpm): Python support tapset
* systemtap.examples/general/(pyexample.meta,pyexample.stp,pyexample.py):  Use it.

10 years agoPR16716 partial fix: Better types in 'syscall.{getpriority,setpriority}'.
David Smith [Mon, 21 Apr 2014 18:43:33 +0000 (13:43 -0500)]
PR16716 partial fix: Better types in 'syscall.{getpriority,setpriority}'.

* tapset/linux/syscalls.stp: Fix types in 'syscall.getpriority'.
* tapset/linux/syscalls2.stp: Fix types in 'syscall.setpriority'.
* tapset/linux/aux_syscalls.stp (_priority_which_str): Convert to use
  _stp_lookup_str().
* testsuite/systemtap.syscall/getpriority.c: New test case.
* testsuite/systemtap.syscall/setpriority.c: Ditto.

10 years agoPR16716 partial fix: Better types in 'syscall.{getpgid,reboot,setpgid}'.
David Smith [Mon, 21 Apr 2014 17:52:26 +0000 (12:52 -0500)]
PR16716 partial fix: Better types in 'syscall.{getpgid,reboot,setpgid}'.

* tapset/linux/syscalls.stp: Fix types in 'syscall.getpgid'.
* tapset/linux/syscalls2.stp: Fix types in 'syscall.reboot' and
  'syscall.setpgid'.
* tapset/linux/x86_64/syscalls.stp: Fix nesting in 'syscall.mmap2'.
* tapset/linux/x86_64/nd_syscalls.stp: Ditto.
* tapset/linux/aux_syscalls.stp (_reboot_magic_str): Convert to use
  _stp_lookup_str().
  (_reboot_flag_str): Ditto.
* runtime/linux/compat_unistd.h: Added mmap2 defines.
* testsuite/systemtap.syscall/pgid.c (main): Add more tests.
* testsuite/systemtap.syscall/reboot.c: New test case.

10 years agotapset: set _shutdown_how_str and _madvice_advice_str :string
Josh Stone [Fri, 18 Apr 2014 01:27:01 +0000 (18:27 -0700)]
tapset: set _shutdown_how_str and _madvice_advice_str :string

These were failing buildok/aux_syscalls-embedded.stp, because print()
can't figure out their types implicitly.

10 years agotapset: mark _sched_policy_str() as pure and :string
Josh Stone [Fri, 18 Apr 2014 01:14:06 +0000 (18:14 -0700)]
tapset: mark _sched_policy_str() as pure and :string

This was causing issues in any script that pulled in aux_syscalls, but
didn't use _sched_policy_str().  Since it wasn't pure, it couldn't be
elided, and since it wasn't explicitly declared :string, there was no
type deduction to define its STAP_RETVALUE.

10 years agotestsuite: tighten warn_overflow.exp verbosity to just pass-5
Josh Stone [Fri, 18 Apr 2014 01:14:57 +0000 (18:14 -0700)]
testsuite: tighten warn_overflow.exp verbosity to just pass-5

10 years agotestsuite: don't mix -p2 and -l in labels.exp
Josh Stone [Fri, 18 Apr 2014 00:54:01 +0000 (17:54 -0700)]
testsuite: don't mix -p2 and -l in labels.exp

Since commit bba368c5c5fea, -p2 and -l are exclusive options, even
though internally they are similar.

Furthermore, "labels exe .statement" and "labels so .statement" had this
failure masked, because they were looking for the absence of the error
"semantic error: no match" rather than the presence of actual matches.
Thus the option-validating error didn't trigger failure before.

10 years agoPrevent lock-recursion in _stp_ctl_send
Josh Stone [Thu, 17 Apr 2014 21:32:40 +0000 (14:32 -0700)]
Prevent lock-recursion in _stp_ctl_send

In rare cases, we may hit a probe while the transport layer is holding a
spinlock, and that probe may call _stp_ctl_send which tries to grab the
same and deadlocks.  This is a bit easier to trigger on lockdep-enabled
kernels with the lock_acquired tracepoint.

This patch refactors the context->busy state management into get/put
context, and those areas which grab probe-sensitive locks now wrap
themselves with a context to stay comfortably free of probes.

Tangentially, the dyninst side abandons the busy flag, as it already had
a tls_context pointer to prevent direct recursion and a mutex for
exclusive access across all processes.

DEBUG_REENTRANCY is an unfortunate casualty, because we can't safely
call _stp_warn when the busy context may be from those held locks.

10 years agoNEWS: add blurb re. statement probes
Jonathan Lebon [Thu, 17 Apr 2014 21:43:29 +0000 (17:43 -0400)]
NEWS: add blurb re. statement probes

10 years agoMerge branch 'jlebon/pr16615'
Jonathan Lebon [Thu, 17 Apr 2014 21:42:52 +0000 (17:42 -0400)]
Merge branch 'jlebon/pr16615'

10 years agostmt_rel.exp: improve coverage
Jonathan Lebon [Thu, 17 Apr 2014 21:34:24 +0000 (17:34 -0400)]
stmt_rel.exp: improve coverage

The testcase previously only tested that specific relative linenos in
bio_init() were valid and that there were at least 3 linenos available
for probing.

We now improve this test by checking that probes listed by the wildcard
lineno are all accessible by relative numbering as well. As a sanity
check, we check that bio_init() has at least 3 linenos available, as
before.

10 years agostmt_rel_user.exp: generalize it
Jonathan Lebon [Thu, 17 Apr 2014 21:32:44 +0000 (17:32 -0400)]
stmt_rel_user.exp: generalize it

Since only stmt_rel_user.exp uses stmt_rel.c, it's more fitting to
rename it and make the testcase refer to it in a more general way.

10 years agotestsuite: adapt to new statement probes
Jonathan Lebon [Thu, 17 Apr 2014 18:05:38 +0000 (14:05 -0400)]
testsuite: adapt to new statement probes

- implicitptr.exp: set lineno to 13. 7 falls outside the function.
- library.stp: adjust relative lineno
- stmt_rel_user.exp: ditto

10 years agoadd DWARF_LINE* macros to help diagnosis
Jonathan Lebon [Thu, 17 Apr 2014 16:21:25 +0000 (12:21 -0400)]
add DWARF_LINE* macros to help diagnosis

Similarly to the previous commit, we modify the new safe_dwarf_line*()
functions so that they carry __FILE__ and __LINE__ information into the
error. This new information is filled in when using the new DWARF_LINE*
macros.

Before:

semantic error: libdw failure (dwarf_lineaddr): no error

After:

semantic error: libdw failure (dwarf_lineaddr): no error
   thrown from: ../systemtap/dwflpp.cxx:2242

10 years agoadd DWFL_ASSERT and DWARF_ASSERT to help diagnosis
Jonathan Lebon [Thu, 17 Apr 2014 16:08:00 +0000 (12:08 -0400)]
add DWFL_ASSERT and DWARF_ASSERT to help diagnosis

Semantic errors thrown from dwfl_assert() and dwarf_assert() lacked any
positional information to help track down where the assertion failed. We
create two new macros, DWFL_ASSERT and DWARF_ASSERT, which carry down
the __FILE__ and __LINE__ information so that the semantic_error created
contains that information, which can be printed out using -vv.

Before:

semantic error: libdwfl failure (asserting!): no error

After:

semantic error: libdwfl failure (asserting!): no error
   thrown from: ../systemtap/tapsets.cxx:7183

10 years agostatement.exp: rework and expand
Jonathan Lebon [Wed, 16 Apr 2014 18:02:21 +0000 (14:02 -0400)]
statement.exp: rework and expand

The statement.exp test case previously only tested a few specific cases.
We now introduce a new test program, 'statement.c', on which we can test
for all the things we previously tested, allowing us to remove the other
test programs. Furthermore, we extend coverage to test many other
possible combinations.

This page took 0.071122 seconds and 5 git commands to generate.