The SystemTap team announces release 5.0
Faster & more reliable kernel-user transport; support for forthcoming
debuginfod "metadata" searches to mass-probe potentially installed
binaries; deprecated support for RHEL6 2.6.32 and earlier kernels;
more DWARF5 debuginfo format support.
= Where to get it
https://sourceware.org/systemtap/ - our project page
https://sourceware.org/systemtap/ftp/releases/
https://koji.fedoraproject.org/koji/packageinfo?packageID=615
git tag release-5.0 (commit cfc2c1d53924face11e3fab78ded61c359778eb9)
There have been over 110 commits since the last release.
There have been over 35 bugs fixed / features added since the last release.
= SystemTap frontend (stap) changes
- More probe point process details are printed in "-L" list mode with
more "-v".
- For the case where newer kernels may break systemtap runtime APIs,
better pass-4 failure diagnostics are printed.
- The target(s) of process probes may be specified by path name globs,
as located selected debuginfod servers. This requires the
debuginfod servers to support "metadata" queries. At the time of
this systematp release, this debuginfod capability is under review
in elfutils, so not yet merged into elfutils master/releases.
# export DEBUGINFOD_URLS="URL1 URL2 ..."
# stap -e 'probe debuginfod.process("/usr/*/curl").begin { log("hi") }'
= SystemTap backend changes
- Performance improvements in uprobe registration and module startup.
- The kernel-user message transport system added framing codes, making
the transport more reliable, but becoming incompatible across
pre-5.0 versions. Use matching versions of stap and staprun.
- RHEL6, kernel 2.6.32* as well as older releases have been deprecated
from the codebase. RHEL7 and kernel version 3.10 are now the oldest
supported versions for Systemtap.
- New runtime macro STP_TIMING_NSECS is now supported for reporting probe timing
stats in nsecs instead of cycles. This may become default later.
- Add new runtime macro STP_FORCE_STDOUT_TTY to override STP_STDOUT_NOT_ATTY.
= SystemTap tapset changes
- Tapset function print_ubacktrace_fileline() now understands DWARF5.
- The testsuite Makefile has been simplified to remove concurrency,
so "parallel" and "resume" modes are gone.
- Numerous kernel porting tweaks.
= SystemTap sample scripts
* All 190+ examples can be found at https://sourceware.org/systemtap/examples/
= Examples of tested kernel versions
3.10.0 (RHEL7)
4.18.0 (RHEL8 + CentOS Stream 8 x86_64, aarch64, ppc64le, s390x)
5.14.0 (RHEL9 + CentOS Stream 9 x86_64, aarch64, ppc64le, s390x)
6.4.15 (Fedora 38 x86_64)
6.5.6 (Fedora 39 x86_64)
6.6.0-rc (Fedora rawhide x86_64)
= Known issues with this release
None, it's practically perfect.
= Contributors for this release
Aaron Merey, Aliaksandr Valialkin*, Frank Ch. Eigler, Housam Alamour*,
Martin Cermak, Ryan Goldberg, Serhei Makarov, William Cohen, Yaakov
Selkowitz, Yichun Zhang (agentzh), lijunlong*
Special thanks to new contributors, marked with '*' above.
= Bugs fixed for this release <https://sourceware.org/PR#####>
30396 kernel 6.3.* objtool/ibt warnings during pass 4
30406 C symbol conflicts when using Fedora 36's 6.2 kernels and -DDEBUG_MEM
20735 "soft lockup" bug on RHEL7 ppc64
30405 Kernel errors in kallsyms_on_each_symbol() with Fedora 36's 6.2/6.1 debug kernels
30395 Regex code has invalid memory reads caught by KASAN
30408 Always fail to read userland memory (read faults) inside perf event probes with 6.2/6.1 kernels
30454 The stap runtime does not call nmi_uaccess_okay() to protect data races against CR3 switching on X86
29765 Convert NFS from readpages to readahead => tapset/linux/nfs.stp breakage
30484 stap-report scrapes too much /sys /proc
30415 conflicting types for ‘kallsyms_on_each_symbol’
30459 Ko compilation failures with kernel 6.1/6.2 kernel headers
30570 Type inference for elided try-catch block has unresolved error variable type
30617 Systemtap unable to successfully build kernel modules for linux-6.5
30634 sprintf_traceback* can flush content to stapio stream
30442 optional statement probes still throw exceptions that kill pass-2
29108 persistent intermittent bufhdr-corruption / transport-failures, even with large -s
30749 stap --sign-module doesn't sign the module, as one would expect
30456 Kernel 5.18+ retired set_fs() and replaced it with the user_access_begin/end() pair for userland memory accesses
30777 Systemtap modules unable to run on systemtap supporting Intel CET IBT
27410 rpm-set path based probes
30858 improve diagnostics for kernel unsupported version
30864 json-c required, but not during configure
30831 Systemtap scripts fail to compile on newest linux 6.6 kernels
30416 unable to find member 'request' for struct scsi_cmnd
30434 deprecate rhel6 (2.6.32*) support
27791 hw_breakpoint.exp fails to build
26152 _struct_compat_timex_u() function doesn't work on rhel8
13480 ARM missing syscall.sigaltstack
30401 s390x specific: invalid application of ‘sizeof’ to incomplete type ‘struct stack_frame’
30407 Tapset function print_ubacktrace_fileline() no longer works with DWARF5
30987 Addition of strlcopy and strlcat to glibc prevents compilation of dyninst based systemtap instrumentation
30803 Extend debuginfod.process probe to glob by package name too
31011 Memory leaks happen when stap's -t option is specified and the stdout stream is not a tty
31013 Use of sleeping _stp_stat_del() operations in atomic contexts when cleaning up for -t
31012 Bug: kernel module loading might block a CPU more than 1.5ms on very fast machines
31018 Map operations might get no lock protections due to "pushdown lock" bugs
31028 Systemtap doesn't find probe points for process.function("*").callee("*")' in code with DWARF5 debuginfo