From: Pedro Alves <palves@redhat.com>
To: Yao Qi <qiyaoltc@gmail.com>, gdb-patches@sourceware.org
Subject: Re: [PATCH 3/7] Force to insert software single step breakpoint
Date: Mon, 11 Apr 2016 14:31:00 -0000 [thread overview]
Message-ID: <570BB52F.7@redhat.com> (raw)
In-Reply-To: <1458749384-19793-4-git-send-email-yao.qi@linaro.org>
On 03/23/2016 04:09 PM, Yao Qi wrote:
> GDB doesn't insert software single step breakpoint if the instruction
> branches to itself, so that the program can't stop after command "si".
>
> (gdb) b 32
> Breakpoint 2 at 0x8680: file git/gdb/testsuite/gdb.base/branch-to-self.c, line 32.
> (gdb) c
> Continuing.
>
> Breakpoint 2, main () at gdb/git/gdb/testsuite/gdb.base/branch-to-self.c:32
> 32 asm (".Lhere: " BRANCH_INSN " .Lhere"); /* loop-line */
> (gdb) si
> infrun: clear_proceed_status_thread (Thread 3991.3991)
> infrun: proceed (addr=0xffffffff, signal=GDB_SIGNAL_DEFAULT)
> infrun: step-over queue now empty
> infrun: resuming [Thread 3991.3991] for step-over
> infrun: skipping breakpoint: stepping past insn at: 0x8680
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> Sending packet: $Z0,8678,4#f3...Packet received: OK
> infrun: skipping breakpoint: stepping past insn at: 0x8680
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> Sending packet: $Z0,b6fe86c8,4#82...Packet received: OK
> infrun: resume (step=1, signal=GDB_SIGNAL_0), trap_expected=1, current thread [Thread 3991.3991] at 0x868
>
> breakpoint.c:should_be_inserted thinks the breakpoint shouldn't be
> inserted, which is wrong. This patch restrict the condition that only
> return false if breakpoint is NOT single step breakpoint.
>
> gdb:
>
> 2016-03-23 Yao Qi <yao.qi@linaro.org>
>
> * breakpoint.c (should_be_inserted): Don't return 0 if single
> step breakpoint is inserted at the address we're stepping over.
> * gdbarch.sh (software_single_step): Update comments.
> * gdbarch.h: Regenerated.
> ---
> gdb/breakpoint.c | 9 ++++++++-
> gdb/gdbarch.h | 5 ++++-
> gdb/gdbarch.sh | 5 ++++-
> 3 files changed, 16 insertions(+), 3 deletions(-)
>
> diff --git a/gdb/breakpoint.c b/gdb/breakpoint.c
> index f99a7ab..9ecfb07 100644
> --- a/gdb/breakpoint.c
> +++ b/gdb/breakpoint.c
> @@ -2219,9 +2219,16 @@ should_be_inserted (struct bp_location *bl)
> return 0;
>
> /* Don't insert a breakpoint if we're trying to step past its
> - location. */
> + location except single step breakpoint, because the single step
> + breakpoint may be inserted at the location we're trying to step
> + if the instruction branches to itself. However, the instruction
> + won't be executed at all and it may break the semantics of the
> + instruction, for example, the instruction is a conditional
> + branch or updates some flags. We can't fix it unless GDB is able
> + to emulate the instruction or switch to displaced stepping. */
> if ((bl->loc_type == bp_loc_software_breakpoint
> || bl->loc_type == bp_loc_hardware_breakpoint)
> + && bl->owner->type != bp_single_step
> && stepping_past_instruction_at (bl->pspace->aspace,
> bl->address))
Another scenario occurred to me:
- Thread A is software single-stepping.
- Thread B hits single-step breakpoint of thread A.
- We pause all threads and set thread B stepping past the
single-step breakpoint of thread A.
But if the single-step breakpoint is for another thread, then
we won't actually manage to have thread B step past it, resulting
in spurious re-hits and no-guaranteed forward progress. See
e.g., non-stop-fair-events.exp:
# On software single-step targets that don't support displaced
# stepping, threads keep hitting each others' single-step
# breakpoints, and then GDB needs to pause all threads to step
# past those. The end result is that progress in the main
# thread will be slower and it may take a bit longer for the
# signal to be queued; bump the timeout.
Sounds like we may need to look at the single-step breakpoint's thread
id, and only insert it if it is for the thread that is going to be
doing the step-over? We may need to record that in step_over_info and
pass more info to stepping_past_instruction_at.
> --- a/gdb/gdbarch.sh
> +++ b/gdb/gdbarch.sh
> @@ -609,7 +609,10 @@ m:CORE_ADDR:addr_bits_remove:CORE_ADDR addr:addr::core_addr_identity::0
> # target can single step. If not, then implement single step using breakpoints.
> #
> # A return value of 1 means that the software_single_step breakpoints
> -# were inserted; 0 means they were not.
> +# were inserted; 0 means they were not. Multiple breakpoints may be
> +# inserted for some instructions such as conditional branch. However,
> +# each implementation must always evaluate the condition and only put
> +# the breakpoint at the branch destination if the condition is true.
I'd add:
(...) condition is true, so that we ensure forward progress when
stepping past a conditional branch to self.
This will help porters evaluate whether that's really necessary
for their ports.
Thanks,
Pedro Alves
next prev parent reply other threads:[~2016-04-11 14:31 UTC|newest]
Thread overview: 31+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-03-23 16:10 [PATCH 0/7 V2] Step over instruction branches to itself Yao Qi
2016-03-23 16:10 ` [PATCH 6/7] Resume the inferior with signal rather than stepping over Yao Qi
2016-04-11 15:29 ` Pedro Alves
2016-03-23 16:10 ` [PATCH 4/7] Insert breakpoint even when the raw breakpoint is found Yao Qi
2016-04-11 14:41 ` Pedro Alves
2016-04-12 9:04 ` Yao Qi
2016-04-12 9:41 ` Pedro Alves
2016-04-25 8:45 ` Yao Qi
2016-03-23 16:10 ` [PATCH 2/7] Deliver signal in hardware single step Yao Qi
2016-04-11 14:10 ` Pedro Alves
2016-04-22 10:54 ` Yao Qi
2016-03-23 16:10 ` [PATCH 1/7] New test case gdb.trace/signal.exp Yao Qi
2016-04-08 16:52 ` Pedro Alves
2016-04-11 8:41 ` Yao Qi
2016-04-11 14:04 ` Pedro Alves
2016-04-22 10:53 ` Yao Qi
2016-04-26 12:57 ` Pedro Alves
2016-04-11 14:08 ` Pedro Alves
2016-03-23 16:10 ` [PATCH 5/7] [GDBserver] Don't error in reinsert_raw_breakpoint if bp->inserted Yao Qi
2016-04-11 14:54 ` Pedro Alves
2016-03-23 16:10 ` [PATCH 3/7] Force to insert software single step breakpoint Yao Qi
2016-04-11 14:31 ` Pedro Alves [this message]
2016-04-13 16:21 ` Yao Qi
2016-04-19 14:54 ` Yao Qi
2016-04-19 15:17 ` Pedro Alves
2016-04-20 7:50 ` Yao Qi
2016-04-22 16:36 ` Pedro Alves
2016-04-25 8:40 ` Yao Qi
2016-03-23 16:26 ` [PATCH 7/7] New test case gdb.base/branch-to-self.exp Yao Qi
2016-04-11 15:34 ` Pedro Alves
2016-04-25 8:58 ` Yao Qi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=570BB52F.7@redhat.com \
--to=palves@redhat.com \
--cc=gdb-patches@sourceware.org \
--cc=qiyaoltc@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).