[PATCH qemu v4 0/2] ARM: Add support for V8M special registers in GDB stub

~dreiss-meta posted 2 patches 1 year, 2 months ago
target/arm/cpu.h      |  25 ++++-
target/arm/gdbstub.c  | 241 ++++++++++++++++++++++++++++++++++++++++++
target/arm/m_helper.c |  87 +++++++--------
3 files changed, 305 insertions(+), 48 deletions(-)
[PATCH qemu v4 0/2] ARM: Add support for V8M special registers in GDB stub
Posted by ~dreiss-meta 1 year, 2 months ago
New in v4: Moved arm_v7m_mrs_control out of the `#ifdef
CONFIG_USER_ONLY` block, unbreaking the user-only build.  The downside
is that this function is now taking up space in the user-only binary,
but it can (presumably?) never be used because there are no user modes
for v8m cores.  Let me know if you'd prefer for me to wrap `#ifdef
CONFIG_USER_ONLY` around the v8m registers in the gdb stub.  Also, let
me know if you'd prefer a separate commit for renaming and moving
v7m_mrs_control.

David Reiss (2):
  target/arm/gdbstub: Support reading M system registers from GDB
  target/arm/gdbstub: Support reading M security extension registers
    from GDB

 target/arm/cpu.h      |  25 ++++-
 target/arm/gdbstub.c  | 241 ++++++++++++++++++++++++++++++++++++++++++
 target/arm/m_helper.c |  87 +++++++--------
 3 files changed, 305 insertions(+), 48 deletions(-)

-- 
2.34.5
Re: [PATCH qemu v4 0/2] ARM: Add support for V8M special registers in GDB stub
Posted by Peter Maydell 1 year, 2 months ago
On Tue, 7 Feb 2023 at 17:37, ~dreiss-meta <dreiss-meta@git.sr.ht> wrote:
>
> New in v4: Moved arm_v7m_mrs_control out of the `#ifdef
> CONFIG_USER_ONLY` block, unbreaking the user-only build.  The downside
> is that this function is now taking up space in the user-only binary,
> but it can (presumably?) never be used because there are no user modes
> for v8m cores.  Let me know if you'd prefer for me to wrap `#ifdef
> CONFIG_USER_ONLY` around the v8m registers in the gdb stub.  Also, let
> me know if you'd prefer a separate commit for renaming and moving
> v7m_mrs_control.

We do support the M-profile cores in the linux-user executables,
so this code is "live". The view that the guest program and
the debugger should see is that the core is always in
user mode (aka unprivileged, Thread mode).

thanks
-- PMM
Re: [PATCH qemu v4 0/2] ARM: Add support for V8M special registers in GDB stub
Posted by Peter Maydell 1 year, 1 month ago
On Tue, 7 Feb 2023 at 17:51, Peter Maydell <peter.maydell@linaro.org> wrote:
>
> On Tue, 7 Feb 2023 at 17:37, ~dreiss-meta <dreiss-meta@git.sr.ht> wrote:
> >
> > New in v4: Moved arm_v7m_mrs_control out of the `#ifdef
> > CONFIG_USER_ONLY` block, unbreaking the user-only build.  The downside
> > is that this function is now taking up space in the user-only binary,
> > but it can (presumably?) never be used because there are no user modes
> > for v8m cores.  Let me know if you'd prefer for me to wrap `#ifdef
> > CONFIG_USER_ONLY` around the v8m registers in the gdb stub.  Also, let
> > me know if you'd prefer a separate commit for renaming and moving
> > v7m_mrs_control.
>
> We do support the M-profile cores in the linux-user executables,
> so this code is "live". The view that the guest program and
> the debugger should see is that the core is always in
> user mode (aka unprivileged, Thread mode).

Having looked more closely at the registers in the
org.gnu.gdb.arm.secext set, I think the right answer here is
that for CONFIG_USER_ONLY we should never expose the second
org.gnu.gdb.arm.secext block. Because the guest is restricted
to non-secure mode only, the registers in the
org.gnu.gdb.arm.m-system block are sufficient to get all the
interesting information, and the debugger shouldn't be confused
by only seeing one of the two, because that's what it sees when
talking to a CPU which doesn't have the security extension.
The register state in the other block is OK.

(I don't think the code as it stands will be broken as such,
but it will let the debugger read a lot of confusing 0s, and
it leaves the door open for us forgetting to prevent writes
if we add system register write suport later, which would let
gdb put QEMU into states it doesn't expect to be in.)

That change is simple (couple of ifdefs in gdbstub.c, and
then it lets us avoid moving the definition of the
arm_v7m_get_sp_ptr() function). But Richard has just told
me he has a rework of some of the gdbstub handling which
he incorporated this patchset into, so I think the best
thing is if he can post that series.

thanks
-- PMM