| Message ID | alpine.LFD.2.21.2002081532520.18621@redsun52.ssa.fujisawa.hgst.com |
|---|---|
| State | Not Applicable |
| Headers | show |
| Series | configure: Re-disable building cross-gdbserver | expand |
>>>>> "Maciej" == Maciej W Rozycki <macro@wdc.com> writes:
Maciej> Correct fallout from commit 919adfe84092 ("Move gdbserver to top level")
Maciej> and revert to not building `gdbserver' in a cross-configuration, that is
Maciej> where host != target, matching the documented behaviour. We have no way
Maciej> to support non-native `gdbserver', and native `gdbserver' is usually of
Maciej> no use with cross-GDB of the chosen host.
Pedro had a different way to do this, that keeps the decision under
gdbserver's control:
https://sourceware.org/ml/gdb-patches/2020-02/msg00383.html
Also note that I haven't sent the top-level configury patches upstream
to gcc yet. I am going to do this all at once, at the end. I think
there's one more patch that will be needed, once gdbserver depends on
the top-level gdbsupport.
Tom
On Tue, 11 Feb 2020, Tom Tromey wrote: > Maciej> Correct fallout from commit 919adfe84092 ("Move gdbserver to top level") > Maciej> and revert to not building `gdbserver' in a cross-configuration, that is > Maciej> where host != target, matching the documented behaviour. We have no way > Maciej> to support non-native `gdbserver', and native `gdbserver' is usually of > Maciej> no use with cross-GDB of the chosen host. > > Pedro had a different way to do this, that keeps the decision under > gdbserver's control: > > https://sourceware.org/ml/gdb-patches/2020-02/msg00383.html That's actually quite similar to what I considered first, before I changed my mind. Whatever. However I would expect `exit' not to be what we want in a sourced script (I did this differently; see below). > Also note that I haven't sent the top-level configury patches upstream > to gcc yet. I am going to do this all at once, at the end. I think > there's one more patch that will be needed, once gdbserver depends on > the top-level gdbsupport. Ack. Maciej --- gdbserver/configure.srv | 8 ++++++++ 1 file changed, 8 insertions(+) Index: binutils-gdb/gdbserver/configure.srv =================================================================== --- binutils-gdb.orig/gdbserver/configure.srv +++ binutils-gdb/gdbserver/configure.srv @@ -36,6 +36,14 @@ srv_linux_obj="linux-low.o nat/linux-osd # Input is taken from the "${host}" variable. case "${host}" in + ${target}) + gdbserver_host=${host} + ;; + *) + gdbserver_host=NONE + ;; +esac +case "${gdbserver_host}" in aarch64*-*-linux*) srv_tgtobj="linux-aarch64-low.o" srv_tgtobj="$srv_tgtobj nat/aarch64-linux-hw-point.o" srv_tgtobj="$srv_tgtobj linux-aarch32-low.o"
On 2/11/20 9:01 PM, Maciej W. Rozycki wrote: > On Tue, 11 Feb 2020, Tom Tromey wrote: > >> Maciej> Correct fallout from commit 919adfe84092 ("Move gdbserver to top level") >> Maciej> and revert to not building `gdbserver' in a cross-configuration, that is >> Maciej> where host != target, matching the documented behaviour. We have no way >> Maciej> to support non-native `gdbserver', and native `gdbserver' is usually of >> Maciej> no use with cross-GDB of the chosen host. >> >> Pedro had a different way to do this, that keeps the decision under >> gdbserver's control: >> >> https://sourceware.org/ml/gdb-patches/2020-02/msg00383.html > > That's actually quite similar to what I considered first, before I > changed my mind. Whatever. Doing it in gdbserver/ has the advantage that it stays under gdbserver's control, so it doesn't need syncing code with the gcc tree. I know of at least one off-tree port that uses gdbserver in a host != target scenario, so I imagine that this condition will evolve over time. Also, this way, changes in this area don't require running autoconf to regenerate configure. I'm not seeing any downside. > > However I would expect `exit' not to be what we want in a sourced script > (I did this differently; see below). Good point, somehow did not think of that. It worked in my patch because we source the script in a sub-shell. But it's clearer/better to not rely on that. > > case "${host}" in > + ${target}) > + gdbserver_host=${host} > + ;; > + *) > + gdbserver_host=NONE > + ;; if/else reads more to-the-point to me, so I tweaked it that way, and merged it in (to binutils-gdb), like below. I'm sorry for not noticing your earlier patch. From f20e3e823d56e54ffe56792ea6a2fe947c2dec0d Mon Sep 17 00:00:00 2001 From: "Maciej W. Rozycki" <macro@wdc.com> Date: Wed, 12 Feb 2020 13:50:30 +0000 Subject: [PATCH] Disable gdbserver on host != target configurations Correct fallout from commit 919adfe84092 ("Move gdbserver to top level") and revert to not building `gdbserver' in a cross-configuration, that is where host != target, matching the documented behaviour. We have no way to support non-native `gdbserver', and native `gdbserver' is usually of no use with cross-GDB of the chosen host. gdbserver/ChangeLog: 2020-02-12 Maciej W. Rozycki <macro@wdc.com> Pedro Alves <palves@redhat.com> Skip building gdbserver in a cross-configuration. * configure.srv: Set $gdbserver_host depending on whether $target is $host. Use $gdbserver_host instead of $host. --- gdbserver/ChangeLog | 7 +++++++ gdbserver/configure.srv | 11 +++++++++-- 2 files changed, 16 insertions(+), 2 deletions(-) diff --git a/gdbserver/ChangeLog b/gdbserver/ChangeLog index 09707067730..709ef23674c 100644 --- a/gdbserver/ChangeLog +++ b/gdbserver/ChangeLog @@ -1,3 +1,10 @@ +2020-02-12 Maciej W. Rozycki <macro@wdc.com> + Pedro Alves <palves@redhat.com> + + Skip building gdbserver in a cross-configuration. + * configure.srv: Set $gdbserver_host depending on whether $target + is $host. Use $gdbserver_host instead of $host. + 2020-02-11 Simon Marchi <simon.marchi@efficios.com> * configure: Re-generate. diff --git a/gdbserver/configure.srv b/gdbserver/configure.srv index 2e83cbdc07f..375ac0aeb2a 100644 --- a/gdbserver/configure.srv +++ b/gdbserver/configure.srv @@ -33,9 +33,16 @@ ipa_ppc_linux_regobj="powerpc-32l-ipa.o powerpc-altivec32l-ipa.o powerpc-vsx32l- # these files over and over again. srv_linux_obj="linux-low.o nat/linux-osdata.o nat/linux-procfs.o nat/linux-ptrace.o nat/linux-waitpid.o nat/linux-personality.o nat/linux-namespaces.o fork-child.o nat/fork-inferior.o" -# Input is taken from the "${host}" variable. +# Input is taken from the "${host}" and "${target}" variables. -case "${host}" in +# GDBserver can only debug native programs. +if test "${target}" = "${host}"; then + gdbserver_host=${host} +else + gdbserver_host= +fi + +case "${gdbserver_host}" in aarch64*-*-linux*) srv_tgtobj="linux-aarch64-low.o" srv_tgtobj="$srv_tgtobj nat/aarch64-linux-hw-point.o" srv_tgtobj="$srv_tgtobj linux-aarch32-low.o" base-commit: 38de8abe21fe17c31888094bd860a84f88cb5749
On Wed, 12 Feb 2020, Pedro Alves wrote: > > That's actually quite similar to what I considered first, before I > > changed my mind. Whatever. > > Doing it in gdbserver/ has the advantage that it stays under gdbserver's > control, so it doesn't need syncing code with the gcc tree. I know of at > least one off-tree port that uses gdbserver in a host != target scenario, > so I imagine that this condition will evolve over time. Sure, that makes sense to me. > > case "${host}" in > > + ${target}) > > + gdbserver_host=${host} > > + ;; > > + *) > > + gdbserver_host=NONE > > + ;; > if/else reads more to-the-point to me, so I tweaked it that > way, and merged it in (to binutils-gdb), like below. Great, thanks for handling this! > I'm sorry for not noticing your earlier patch. No worries, I'm glad we've got this sorted. Maciej
Index: binutils-gdb/configure =================================================================== --- binutils-gdb.orig/configure +++ binutils-gdb/configure @@ -3538,12 +3538,13 @@ case "${target}" in ;; esac -# Only allow gdbserver on some systems. +# Only allow native gdbserver and then only on some systems. if test -d ${srcdir}/gdbserver; then if test x$enable_gdbserver = x; then { $as_echo "$as_me:${as_lineno-$LINENO}: checking for gdbserver support" >&5 $as_echo_n "checking for gdbserver support... " >&6; } - if (srcdir=${srcdir}/gdbserver; \ + if test x${is_cross_compiler} = xyes \ + || (srcdir=${srcdir}/gdbserver; \ . ${srcdir}/configure.srv; \ test -n "$UNSUPPORTED") then Index: binutils-gdb/configure.ac =================================================================== --- binutils-gdb.orig/configure.ac +++ binutils-gdb/configure.ac @@ -782,11 +782,12 @@ case "${target}" in ;; esac -# Only allow gdbserver on some systems. +# Only allow native gdbserver and then only on some systems. if test -d ${srcdir}/gdbserver; then if test x$enable_gdbserver = x; then AC_MSG_CHECKING([for gdbserver support]) - if (srcdir=${srcdir}/gdbserver; \ + if test x${is_cross_compiler} = xyes \ + || (srcdir=${srcdir}/gdbserver; \ . ${srcdir}/configure.srv; \ test -n "$UNSUPPORTED") then
Correct fallout from commit 919adfe84092 ("Move gdbserver to top level") and revert to not building `gdbserver' in a cross-configuration, that is where host != target, matching the documented behaviour. We have no way to support non-native `gdbserver', and native `gdbserver' is usually of no use with cross-GDB of the chosen host. * configure.ac: Do not build `gdbserver' if $is_cross_compiler. * configure: Regenerate. --- Hi, Verified with a native build, a crossed build of a native configuration and a build of a cross-debugger; Canadian Cross not checked. OK to apply? NB I have noticed that the gdbserver(1) man page is still being built and installed (as ${target_alias}-gdbserver.1, if building a cross-debugger), even if `gdbserver' itself is not (and it is not built if only `gdbserver' is while GDB is not). Obviously this is due to the man page still living under gdb/doc/, and I presume it will be addressed sometime soon, by moving the man page somewhere under gdbserver/, right? Also there are currently a number of mismatches in configure.ac between the gcc and the binutils-gdb repositories; what is the plan to eliminate them? Maciej --- configure | 5 +++-- configure.ac | 5 +++-- 2 files changed, 6 insertions(+), 4 deletions(-) gdbserver-no-cross.diff