Hi All,
GDB semihosting support is broken in the current trunk. When debugging a
basic "Hello, World" application via the QEMU GDB stub:
$ qemu-system-arm -s -S -M integratorcp -cpu any --semihosting
--monitor null --serial null -kernel hello
GDB (7.2.50) receives an interrupt before anything is printed:
Program received signal SIGINT, Interrupt.
As an example, consider this processing of the 'isatty' syscall.
'gdb_do_syscall' gets executed while 'current_run_state == RUN_STATE_RUNNING'.
It then changes the VM and GDB server state with:
s->state = RS_SYSCALL;
vm_stop(RUN_STATE_DEBUG);
s->state = RS_IDLE;
which leaves the GDB server state in 'RS_IDLE', but since the syscall
happens while in the TCG thread the 'RUN_STATE_DEBUG' stop does not
happen immediately. This leads to the following course of events:
1. 'gdb_do_syscall' finishes up by sending a 'Fisatty,00000001' packet to the
GDB client.
2. 'RUN_STATE_DEBUG' get pulled from the event queue.
3. 'gdb_vm_state_change' is invoked thus causing a 'T05thread:01;' reply
to be sent back to the GDB client.
4. A payload of '+$F1#77+' comes in containing the syscall reply and the
'T05' ACK.
5. '+$F1#77' is processed and 'gdb_handle_packet' transitions the
state back 'RUN_STATE_RUNNING'.
6. 'gdb_read_byte' starts processing the remaining payload of '+',
but the current state is 'RUN_STATE_RUNNING' so 'gdb_read_byte' issues
a 'vm_stop(RUN_STATE_PAUSED)' which takes effect immediately.
7. 'gdb_vm_state_change' is invoked thus causing a 'T02thread:01;' reply
to be sent back to the GDB client.
8. The GDB client interrupts and breaks the semihosting flow.
This patch fixes the problem be staying in the 'RS_SYSCALL' state until next
packet read comes in. Therefore keeping any 'T' statuses from being sent
back to the GDB client while the syscall is still being processed.
Meador Inge (1):
gdbserver: Keep VM state status replies from happening during a
syscall
gdbstub.c | 3 ++-
1 files changed, 2 insertions(+), 1 deletions(-)
--
1.7.7.6
