[regression] Do not read from catchpoint/watchpoint locations' addresses when checkin...
authorLuis Machado <lgustavo@codesourcery.com>
Wed, 12 Aug 2015 08:36:09 +0000 (05:36 -0300)
committerLuis Machado <lgustavo@codesourcery.com>
Wed, 12 Aug 2015 08:36:09 +0000 (05:36 -0300)
commit244558af868d5427903c35c5105bf5499639f81f
tree4d837e5976c06fc18f619815cf393d2f7a5f9330
parent0eace2105dc6c005fdc908414ad564d318b71540
[regression] Do not read from catchpoint/watchpoint locations' addresses when checking for a permanent breakpoint

While running bare-metal tests with GDB i noticed some failures in
gdb.base/break.exp, related to the use of the catch commands.

It turns out GDB tries to access memory address 0x0 whenever one tries
to insert a catchpoint, which should obviously not happen.

This was introduced with the changes for permanent breakpoints. In special,
bp_loc_is_permanent tries to check if there is a breakpoint inserted at
the same address as the current breakpoint's location's address. In the
case of catchpoints, this is 0x0.

(top-gdb) catch fork
Sending packet: $m0,1#fa...Packet received: E01
Catchpoint 4 (fork)

(top-gdb) catch vfork
Sending packet: $m0,1#fa...Packet received: E01
Catchpoint 5 (vfork)

It is not obvious to detect because this fails silently for Linux. For our
bare-metal testing, though, this fails with a clear error message from the
target about not being able to read such address.

The attached patch addresses this by bailing out of bp_loc_is_permanent (...)
if the location address is not meaningful. I also took the opportunity to
update the comment for breakpoint_address_is_meaningful, which mentioned
breakpoint addresses as opposed to their locations' addresses.

gdb/ChangeLog:

2015-08-11  Luis Machado  <lgustavo@codesourcery.com>

* breakpoint.c (bp_loc_is_permanent): Return 0 when breakpoint
location address is not meaningful.
(breakpoint_address_is_meaningful): Update comment.
gdb/ChangeLog
gdb/breakpoint.c