| Message ID | 20220420065013.222816-1-npiggin@gmail.com |
|---|---|
| Headers | show |
| Series | gdbserver multi-threaded debugging and POWER9/10 support | expand |
On Wed, 20 Apr 2022 at 06:50, Nicholas Piggin <npiggin@gmail.com> wrote: > > POWER10 is working well with sbefifo (modulo the start SBE command > being buggy with stop states -- make sure you use powersave=off when > testing this, until those can be fixed). POWER9 similarly works well, > with the exception of breakpoints causing PSI interrupt storm. > > gdbserver itself was due for a bit of attention so there's a bunch of > improvements and fixes. > > Multi-threaded debugging is implemented later in the series using the > gdb "all-stop" threads model (all target threads are stopped if any is > stopped). > > Since v1: > - Made changes suggested by Joel (thanks). > - Made a generated target, split out the regen of the stale > gdb_parser_precompile.c into that patch. > - Make the double-ctrl-C really immediately kill. > - Implemented multi-threaded debugging. > - A few more smaller improvements (put_mem handling, breakpoint > command handling, returning more registers). > - Redudce round trips further by returning registers with stop reason. > - POWER9 support as well. > - Proper SPATTN / attn detection and clearing for POWER9/10. Looking good. Unless there's things you want to fix on the first go, I'll merge these and you can send any improvements as follow up patches? Cheers, Joel > > Thanks, > Nick > > > Nicholas Piggin (39): > libpdbg: fix p9chip and p10chip thread_stop > gdbserver: add a 'generated' make target, update > gdb_parser_precompile.c > gdbserver: rename detach command handler > gdbserver: fix put_mem command parsing > gdbserver: include <port> argument in command line help text > gdbserver: Document problem and workaround for client remote timeouts > gdbserver: print some client/server info > gdbserver: Make command callback array static > gdbserver: use standard compatibility test calls > gdbserver: fix read buffer overflow > gdbserver: reduce PR_INFO noise > gdbserver: fix interrupt double stop reason response > libpdbg: thread_regs don't print regs > gdbserver: lexer fixes > gdbserver: put_mem handle arbitrary lengths > gdbserver: implement NoAckMode > libpdbg: Remove enable_attn target command > gdbserver: disable attn after breaking > gdbserver: use read-modify-write for put_mem that is not 8-byte > aligned > gdbserver: enable POWER9 support > gdbserver: Add POWER10 support > gdbserver: check for attn using the SPATTN register > gdbserver: breakpoint instruction test current host endian when it is > required > gdbserver: catch ctrl-C to clean up host state > gdbserver: return more registers > gdbserver: initial thread support > gdbserver: track attn enablement by breakpoints > gdbproxy: implement get_thread / set_thread commands > gdbserver: add multi-threaded start/stop helper functions > gdbserver: tidy poll() > gdbserver: multi-thread polling and stop-reason > gdbserver: track stop reason per thread > gdbserver: multithread debugging support > gdbserver: allow gdbserver to start with targets running > gdbserver: better deal with threads initially stopped > gdbserver: send regs with stop reason > gdbserver: use thread_stop_all() / thread_start_all() > gdbserver: handle gdb breakpoint commands > gdbserver: remove fallback breakpoint support in put_mem > > Makefile.am | 12 + > README.md | 45 +- > libpdbg/chip.c | 2 - > libpdbg/hwunit.c | 1 + > libpdbg/hwunit.h | 2 +- > libpdbg/p10chip.c | 8 +- > libpdbg/p8chip.c | 31 - > libpdbg/p9chip.c | 8 +- > libpdbg/thread.c | 7 +- > src/gdb_parser.rl | 124 +++- > src/gdb_parser_precompile.c | 478 ++++++++++----- > src/main.c | 2 +- > src/pdbgproxy.c | 1121 +++++++++++++++++++++++++++++++---- > src/pdbgproxy.h | 11 +- > src/thread.c | 2 + > 15 files changed, 1495 insertions(+), 359 deletions(-) > > -- > 2.35.1 > > -- > Pdbg mailing list > Pdbg@lists.ozlabs.org > https://lists.ozlabs.org/listinfo/pdbg
Excerpts from Joel Stanley's message of May 3, 2022 5:47 pm: > On Wed, 20 Apr 2022 at 06:50, Nicholas Piggin <npiggin@gmail.com> wrote: >> >> POWER10 is working well with sbefifo (modulo the start SBE command >> being buggy with stop states -- make sure you use powersave=off when >> testing this, until those can be fixed). POWER9 similarly works well, >> with the exception of breakpoints causing PSI interrupt storm. >> >> gdbserver itself was due for a bit of attention so there's a bunch of >> improvements and fixes. >> >> Multi-threaded debugging is implemented later in the series using the >> gdb "all-stop" threads model (all target threads are stopped if any is >> stopped). >> >> Since v1: >> - Made changes suggested by Joel (thanks). >> - Made a generated target, split out the regen of the stale >> gdb_parser_precompile.c into that patch. >> - Make the double-ctrl-C really immediately kill. >> - Implemented multi-threaded debugging. >> - A few more smaller improvements (put_mem handling, breakpoint >> command handling, returning more registers). >> - Redudce round trips further by returning registers with stop reason. >> - POWER9 support as well. >> - Proper SPATTN / attn detection and clearing for POWER9/10. > > Looking good. Unless there's things you want to fix on the first go, > I'll merge these and you can send any improvements as follow up > patches? Thanks for the review, sorry it took so long to get around to replying. You raised some good points but if I can address them incrementally might be easier. Thanks, Nick
On Tue, 10 May 2022 at 09:47, Nicholas Piggin <npiggin@gmail.com> wrote: > > Excerpts from Joel Stanley's message of May 3, 2022 5:47 pm: > > On Wed, 20 Apr 2022 at 06:50, Nicholas Piggin <npiggin@gmail.com> wrote: > >> > >> POWER10 is working well with sbefifo (modulo the start SBE command > >> being buggy with stop states -- make sure you use powersave=off when > >> testing this, until those can be fixed). POWER9 similarly works well, > >> with the exception of breakpoints causing PSI interrupt storm. > >> > >> gdbserver itself was due for a bit of attention so there's a bunch of > >> improvements and fixes. > >> > >> Multi-threaded debugging is implemented later in the series using the > >> gdb "all-stop" threads model (all target threads are stopped if any is > >> stopped). > >> > >> Since v1: > >> - Made changes suggested by Joel (thanks). > >> - Made a generated target, split out the regen of the stale > >> gdb_parser_precompile.c into that patch. > >> - Make the double-ctrl-C really immediately kill. > >> - Implemented multi-threaded debugging. > >> - A few more smaller improvements (put_mem handling, breakpoint > >> command handling, returning more registers). > >> - Redudce round trips further by returning registers with stop reason. > >> - POWER9 support as well. > >> - Proper SPATTN / attn detection and clearing for POWER9/10. > > > > Looking good. Unless there's things you want to fix on the first go, > > I'll merge these and you can send any improvements as follow up > > patches? > > Thanks for the review, sorry it took so long to get around to replying. > > You raised some good points but if I can address them incrementally > might be easier. I've merged them. I'll do a bit more testing before pushing them out. Thanks for all the hard work. Cheers, Joel
POWER10 is working well with sbefifo (modulo the start SBE command being buggy with stop states -- make sure you use powersave=off when testing this, until those can be fixed). POWER9 similarly works well, with the exception of breakpoints causing PSI interrupt storm. gdbserver itself was due for a bit of attention so there's a bunch of improvements and fixes. Multi-threaded debugging is implemented later in the series using the gdb "all-stop" threads model (all target threads are stopped if any is stopped). Since v1: - Made changes suggested by Joel (thanks). - Made a generated target, split out the regen of the stale gdb_parser_precompile.c into that patch. - Make the double-ctrl-C really immediately kill. - Implemented multi-threaded debugging. - A few more smaller improvements (put_mem handling, breakpoint command handling, returning more registers). - Redudce round trips further by returning registers with stop reason. - POWER9 support as well. - Proper SPATTN / attn detection and clearing for POWER9/10. Thanks, Nick Nicholas Piggin (39): libpdbg: fix p9chip and p10chip thread_stop gdbserver: add a 'generated' make target, update gdb_parser_precompile.c gdbserver: rename detach command handler gdbserver: fix put_mem command parsing gdbserver: include <port> argument in command line help text gdbserver: Document problem and workaround for client remote timeouts gdbserver: print some client/server info gdbserver: Make command callback array static gdbserver: use standard compatibility test calls gdbserver: fix read buffer overflow gdbserver: reduce PR_INFO noise gdbserver: fix interrupt double stop reason response libpdbg: thread_regs don't print regs gdbserver: lexer fixes gdbserver: put_mem handle arbitrary lengths gdbserver: implement NoAckMode libpdbg: Remove enable_attn target command gdbserver: disable attn after breaking gdbserver: use read-modify-write for put_mem that is not 8-byte aligned gdbserver: enable POWER9 support gdbserver: Add POWER10 support gdbserver: check for attn using the SPATTN register gdbserver: breakpoint instruction test current host endian when it is required gdbserver: catch ctrl-C to clean up host state gdbserver: return more registers gdbserver: initial thread support gdbserver: track attn enablement by breakpoints gdbproxy: implement get_thread / set_thread commands gdbserver: add multi-threaded start/stop helper functions gdbserver: tidy poll() gdbserver: multi-thread polling and stop-reason gdbserver: track stop reason per thread gdbserver: multithread debugging support gdbserver: allow gdbserver to start with targets running gdbserver: better deal with threads initially stopped gdbserver: send regs with stop reason gdbserver: use thread_stop_all() / thread_start_all() gdbserver: handle gdb breakpoint commands gdbserver: remove fallback breakpoint support in put_mem Makefile.am | 12 + README.md | 45 +- libpdbg/chip.c | 2 - libpdbg/hwunit.c | 1 + libpdbg/hwunit.h | 2 +- libpdbg/p10chip.c | 8 +- libpdbg/p8chip.c | 31 - libpdbg/p9chip.c | 8 +- libpdbg/thread.c | 7 +- src/gdb_parser.rl | 124 +++- src/gdb_parser_precompile.c | 478 ++++++++++----- src/main.c | 2 +- src/pdbgproxy.c | 1121 +++++++++++++++++++++++++++++++---- src/pdbgproxy.h | 11 +- src/thread.c | 2 + 15 files changed, 1495 insertions(+), 359 deletions(-)