[Bug runtime/31117] transport failures

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

https://sourceware.org/bugzilla/show_bug.cgi?id=31117

--- Comment #6 from Martin Cermak <mcermak at redhat dot com> ---
Created attachment 15287
  --> https://sourceware.org/bugzilla/attachment.cgi?id=15287&action=edit
possible patch for the 32-bit tests subproblem

(In reply to William Cohen from comment #5)
> The attached patch improves the test results for the systemtap.map tests
> that were transporting more than PAGE_SIZE amount of data.  It also improves
> the test results for the syscall tests.

Thank-you for looking into this Will!

> However, on RHEL9/f39 there
> are a number of unrelated failures on x86_64 for the 32-bit tests because
> the check for the syscall names/numbers added by git commit
> 7fe1b7bb8e665c9edd38eaf34a499618ee763293

Looking closer... Apparently on RHEL9 the @__compat_task macro <=
_stp_is_compat_task()  <= _stp_is_compat_task2() in runtime/compatdefs.h
doesn't work right.  Wow, I think I have a fix for that see attached.  Here is
how my thoughts went:

I've tested the problem is in _stp_is_compat_task2(), that's in
runtime/compatdefs.h.  I've found that a related systemtap commit is
ea81249b90b92a589d304c051efbc5bbad7ce073.  This mentions kernel commit
ff170cd05953, so I've looked at that, and found that kernel function
is_64bit_mm() seems to be doing what I need.  I've then looked at the modern
implementation of this function, and used that back in systemtap's
_stp_is_compat_task2().  And that seems to work on my specific test system,
which is 5.14.0-402.el9.x86_64.

The attached patch works for me on both RHEL8 and RHEL9.  BUT ... this seems to
be a case where this upstream kernel change introduced in vanilla v6.3 was
backported to lower distribution kernel version, so the condition "#if
LINUX_VERSION_CODE < KERNEL_VERSION(5,0,0)" is a bit problematic..

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