Re: [avr-gcc-list] another bug with gdb 5.3 :-(

[Theodore A. Roth]

On Sun, 4 May 2003, Klaus Rudolph wrote:

:)Hi all,
:)
:)
:)I have another reproduceable error in gdb 5.3.
:)
:)Everytime the simulavr raise an irq and jumps to the vector
:)while gdb is looking for the next instruction with a "$s#xx"
:)there come a lot of T.... , thats fine, but then
:)the next T... comes with the irqvector in the PC. (4000000 -> Vector
:)Address 0x40
:)Also OK. But now gdb tries to read memory
:)m00,27 means read memory from flash 0x00 with 0x27 bytes long
:)next is
:)m27,19 means from 0x27 flash address with 0x19 bytes size.
:)after receiving that data the gdb allways crashes.

I think the crash is due to a buffer overrun in avr-tdep.c when reading in 
the prologue when avr-gdb hits an interrupt as a breakpoint. I extended the 
prologue array in avr_scan_prologue() and the crashes went away. I haven't 
had time to follow up on this though.

:)
:)Have you any idea how to debug gdb with gdb?
:)I tried it, but sending a singint to the subprocess 
:)is received by the debugging debugger itself :-(

Use the gdb signal command to send SIGINT to the debuggee (avr-gdb in this 
case.

:)
:)If I attach from the outside world with gdb --pid xxxx to
:)the avr-gdb I got sometimes the same problem, sometimes
:)I got a communication protocol error between the 
:)avr-gdb and simulavr :-(

Could you send me some more details on this comm error?

Ted Roth