git: c328f64d8107 - main - arm64: Fix COMPAT_FREEBSD32.

[Mark Millard]

Jessica Clarke jrtc27 at freebsd.org wrote on
Wed Mar 10 18:36:39 UTC 2021 :

> On 10 Mar 2021, at 18:07, Olivier Houchard <cognet at FreeBSD.org> wrote:
> > 
> > The branch main has been updated by cognet:
> > 
> > URL: https://cgit.FreeBSD.org/src/commit/?id=c328f64d81079bad5064c8a387883df50ab5aaed
> > 
> > commit c328f64d81079bad5064c8a387883df50ab5aaed
> > Author:     Olivier Houchard <cognet at FreeBSD.org>
> > AuthorDate: 2021-03-10 18:01:41 +0000
> > Commit:     Olivier Houchard <cognet at FreeBSD.org>
> > CommitDate: 2021-03-10 18:06:42 +0000
> > 
> >    arm64: Fix COMPAT_FREEBSD32.
> > 
> >    The ENTRY() macro was modified by commit
> >    28d945204ea1014d7de6906af8470ed8b3311335 to add an optional NOP instruction
> >    at the beginning of the function. It is of course an arm64 instruction, so
> >    unsuitable for the 32bits sigcode. So just use EENTRY() instead for
> >    aarch32_sigcode. This should fix receiving signals when running 32bits
> >    binaries on FreeBSD/arm64.
> 
> Hmm, that's a good point, does the extra nop for the AArch64 sigcode
> cause issues for gdb detecting it?
> 
> Perhaps we should upstream CheriBSD's SIGCODE macros?..

I'll note that, in my reports on the lists about aarch64 having
problems with armv7 port building failures, I used lldb because
gdb did not report backtraces beyond #0 but lldb did report
more complete backtraces.

However, for lldb part of my notes were a report of a
potential missing frame as well:

QUOTE
I'll note that when I looked at detail as the assembler level
it appeared that there was a frame not shown between #0 and #1
in lldb's output: Frame #1 "->" was indicating the instruction
after a simple bl to a not-shown subroutine.

For building textproc/libxslt :
(jobserver_acquire not shown between #0 and #1)

(lldb) bt
* thread #1, name = 'gmake', stop reason = signal SIGSEGV
 * frame #0: 0xffffe190
   frame #1: 0x0003b5f8 gmake`new_job(file=<unavailable>) at job.c:1870:21
   frame #2: 0x0002db80 gmake`execute_file_commands(file=<unavailable>) at commands.c:476:3 [artificial]
   frame #3: 0x00049acc gmake`update_file [inlined] remake_file(file=0x400a9700) at remake.c:1234:11
   frame #4: 0x00049a84 gmake`update_file [inlined] update_file_1(file=<unavailable>, depth=6) at remake.c:835
   frame #5: 0x000494ec gmake`update_file(file=<unavailable>, depth=<unavailable>) at remake.c:336
   frame #6: 0x0004b08c gmake`check_dep(file=0x400a9700, depth=<unavailable>, this_mtime=1, must_make_ptr=0xffffc4ac) at remake.c:1024:20
   frame #7: 0x00049074 gmake`update_file at remake.c:572:17
   frame #8: 0x00048b80 gmake`update_file(file=<unavailable>, depth=<unavailable>) at remake.c:336
   frame #9: 0x0004b08c gmake`check_dep(file=0x400a9400, depth=<unavailable>, this_mtime=1, must_make_ptr=0xffffc564) at remake.c:1024:20
   frame #10: 0x00049074 gmake`update_file at remake.c:572:17
   frame #11: 0x00048b80 gmake`update_file(file=<unavailable>, depth=<unavailable>) at remake.c:336
   frame #12: 0x0004b08c gmake`check_dep(file=0x400a8f20, depth=<unavailable>, this_mtime=1, must_make_ptr=0xffffc61c) at remake.c:1024:20
   frame #13: 0x00049074 gmake`update_file at remake.c:572:17
   frame #14: 0x00048b80 gmake`update_file(file=<unavailable>, depth=<unavailable>) at remake.c:336
   frame #15: 0x000487e0 gmake`update_goal_chain(goaldeps=<unavailable>) at remake.c:151:22
   frame #16: 0x0003f25c gmake`main(argc=2, argv=0xffffd470, envp=0xffffffff) at main.c:2589:13
   frame #17: 0x0002c0fc gmake`__start(argc=2, argv=<unavailable>, env=<unavailable>, ps_strings=<unavailable>, obj=0x400c4004, cleanup=0x40091aa0) at crt1_c.c:92:7

-> 1870	        got_token = jobserver_acquire (waiting_jobs != NULL);

   0x3b5f4 <+1288>: bl     0x50078                   ; jobserver_acquire at posixos.c:265
->  0x3b5f8 <+1292>: cmp    r0, #1
END QUOTE

The backtrace example above did not show
jobserver_acquire as having a frame.

The original note had more examples: all the
example process failures had an apparently
missing frame in the lldb backtrace.


===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)