Looks like we got our first SoC that actually insists on using
word-sized accesses for its MMIO registers with the Rk3288. This patch
changes the GDB command handler for reading and writing memory to always
perform word-sized accesses. This isn't really perfect since the remote
GDB interface is just not really meant to interact with MMIO (e.g. you
shouldn't use this on something with read side effects), but for most
of our purposes it should be good enough.
BUG=chrome-os-partner:18390
TEST=Remote GDB works on Veyron even when writing MMIO registers.
Original-Change-Id: I2ae52636593499f70701582811f1b692c1ea8fcc
Original-Signed-off-by: Julius Werner <jwerner@chromium.org>
Original-Reviewed-on: https://chromium-review.googlesource.com/208554
Original-Reviewed-by: David Hendricks <dhendrix@chromium.org>
(cherry picked from commit 028940934e6b45a02122b61bb859588bf8671938)
Signed-off-by: Marc Jones <marc.jones@se-eng.com>
Change-Id: I4185a6efe9a5211525781acd0a167b821e854211
Reviewed-on: http://review.coreboot.org/8130
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Tested-by: build bot (Jenkins)
This patch adds the ability to attach a GDB host through the UART to a
running payload. Libpayload implements a small stub that can parse and
respond to the GDB remote protocol and provide the required primitives
(reading/writing registers/memory, etc.) to allow GDB to control
execution.
The goal of this implementation is to be as small and uninvasive as
possible. It implements only the minimum amount of primitives required,
and relies on GDB's impressive workaround capabilities (such as
emulating breakpoints by temporarily replacing instructions) for the
more complicated features. This way, a relatively tiny amount of code on
the firmware side opens a vast range of capabilities to the user, not
just in debugging but also in remote-controlling the firmware to change
its behavior (e.g. through GDBs ability to modify variables and call
functions).
By default, a system with the REMOTEGDB Kconfig will only trap into GDB
when executing halt() (including the calls from die_if(), assert(), and
exception handlers). In addition, payloads can manually call gdb_enter()
if desired. It will print a final "Ready for GDB connection." on the
serial, detach the normal serial output driver and wait for the commands
that GDB starts sending on attach.
Based on original implementation by Gabe Black <gabeblack@chromium.org>.
BUG=chrome-os-partner:18390
TEST=Boot a GDB enabled image in recovery mode (or get it to hit a
halt()), close your terminal, execute '<toolchain>-gdb --symbols
/build/<board>/firmware/depthcharge_gdb/depthcharge.elf --directory
~/trunk/src/third_party/coreboot/payloads/libpayload --directory
~/trunk/src/platform/depthcharge --directory
~/trunk/src/platform/vboot_reference --ex "target remote
<cpu_uart_pty>"' and behold the magic.
(You can also SIGSTOP your terminal's parent shell and the terminal
itself, and SIGCONT them in reverse order after GDB exits. More
convenient wrapper tools to do all this automatically coming soon.)
Original-Change-Id: Ib440d1804126cdfdac4a8801f5015b4487e25269
Original-Signed-off-by: Julius Werner <jwerner@chromium.org>
Original-Reviewed-on: https://chromium-review.googlesource.com/202563
Original-Reviewed-by: Stefan Reinauer <reinauer@chromium.org>
(cherry picked from commit 9c4a642c7be2faf122fef39bdfaddd64aec68b77)
Signed-off-by: Marc Jones <marc.jones@se-eng.com>
Change-Id: I9238b4eb19d3ab2c98e4e1c5946cd7d252ca3c3b
Reviewed-on: http://review.coreboot.org/8119
Tested-by: build bot (Jenkins)
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Julius Werner