ARM: 8425/1: kgdb: Don't try to stop the machine when setting breakpoints
authorDoug Anderson <armlinux@m.disordat.com>
Wed, 26 Aug 2015 17:26:49 +0000 (18:26 +0100)
committerRussell King <rmk+kernel@arm.linux.org.uk>
Wed, 16 Sep 2015 22:58:46 +0000 (23:58 +0100)
commit7ae85dc7687c7e7119053d83d02c560ea217b772
tree1663450cdac1b5c0f4a9cdd5a73061d0e5b88d69
parent90cde5584a34e210850728bebee4c6fb1a7124ba
ARM: 8425/1: kgdb: Don't try to stop the machine when setting breakpoints

In (23a4e40 arm: kgdb: Handle read-only text / modules) we moved to
using patch_text() to set breakpoints so that we could handle the case
when we had CONFIG_DEBUG_RODATA.  That patch used patch_text().
Unfortunately, patch_text() assumes that we're not in atomic context
when it runs since it needs to grab a mutex and also wait for other
CPUs to stop (which it does with a completion).

This would result in a stack crawl if you had
CONFIG_DEBUG_ATOMIC_SLEEP and tried to set a breakpoint in kgdb.  The
crawl looked something like:

 BUG: scheduling while atomic: swapper/0/0/0x00010007
 CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.2.0-rc7-00133-geb63b34 #1073
 Hardware name: Rockchip (Device Tree)
  (unwind_backtrace) from [<c00133d4>] (show_stack+0x20/0x24)
  (show_stack) from [<c05400e8>] (dump_stack+0x84/0xb8)
  (dump_stack) from [<c004913c>] (__schedule_bug+0x54/0x6c)
  (__schedule_bug) from [<c054065c>] (__schedule+0x80/0x668)
  (__schedule) from [<c0540cfc>] (schedule+0xb8/0xd4)
  (schedule) from [<c0543a3c>] (schedule_timeout+0x2c/0x234)
  (schedule_timeout) from [<c05417c0>] (wait_for_common+0xf4/0x188)
  (wait_for_common) from [<c0541874>] (wait_for_completion+0x20/0x24)
  (wait_for_completion) from [<c00a0104>] (__stop_cpus+0x58/0x70)
  (__stop_cpus) from [<c00a0580>] (stop_cpus+0x3c/0x54)
  (stop_cpus) from [<c00a06c4>] (__stop_machine+0xcc/0xe8)
  (__stop_machine) from [<c00a0714>] (stop_machine+0x34/0x44)
  (stop_machine) from [<c00173e8>] (patch_text+0x28/0x34)
  (patch_text) from [<c001733c>] (kgdb_arch_set_breakpoint+0x40/0x4c)
  (kgdb_arch_set_breakpoint) from [<c00a0d68>] (kgdb_validate_break_address+0x2c/0x60)
  (kgdb_validate_break_address) from [<c00a0e90>] (dbg_set_sw_break+0x1c/0xdc)
  (dbg_set_sw_break) from [<c00a2e88>] (gdb_serial_stub+0x9c4/0xba4)
  (gdb_serial_stub) from [<c00a11cc>] (kgdb_cpu_enter+0x1f8/0x60c)
  (kgdb_cpu_enter) from [<c00a18cc>] (kgdb_handle_exception+0x19c/0x1d0)
  (kgdb_handle_exception) from [<c0016f7c>] (kgdb_compiled_brk_fn+0x30/0x3c)
  (kgdb_compiled_brk_fn) from [<c00091a4>] (do_undefinstr+0x1a4/0x20c)
  (do_undefinstr) from [<c001400c>] (__und_svc_finish+0x0/0x34)

It turns out that when we're in kgdb all the CPUs are stopped anyway
so there's no reason we should be calling patch_text().  We can
instead directly call __patch_text() which assumes that CPUs have
already been stopped.

Fixes: 23a4e4050ba9 ("arm: kgdb: Handle read-only text / modules")
Reported-by: Aapo Vienamo <avienamo@nvidia.com>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
arch/arm/kernel/kgdb.c