[Bug tapsets/28443] New: Provide syscall_any tapset for bpf

["wcohen at redhat dot com" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=28443

            Bug ID: 28443
           Summary: Provide syscall_any tapset for bpf
           Product: systemtap
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: tapsets
          Assignee: systemtap at sourceware dot org
          Reporter: wcohen at redhat dot com
  Target Milestone: ---

A number of systemtap examples use the syscall_any tapset.  Providing a BPF
version of the syscall_any tapset would allow additional examples to run with
as BPF.

Looked into how to implement this.  The raw tracepoints sys_enter and sys_exit
make the pt_regs available.  The sys_enter also has $id (don't know why they
didn't decide to include $id in the sys_exit). A helper function can pick out
the appropriate register from the pt_regs to get the syscall number. The bpf
backend also provides information about which target architecture the machine
is.  As experiment I was ableo to put guards around the initiatization code in
each of the tapset/linux/*/syscall_num.stp files to made the initializer only
run on the specific architecture and then have symbolic links from tapset/bpf
to the arch-specific syscall_num.stp files.  Will need to add the generation of
these guards to scripts/dump-syscalls.sh.

Haven't seen a way to get information whether this is 32-bit user-space code
running on 64-bit platform.  The existing tapset code to do that is using C
include files to get bit mask to determine whether things are running in compat
mode.  There doesn't seem to be a BPF function to extract that information.

-- 
You are receiving this mail because:
You are the assignee for the bug.