mbox series

[0/5] target/arm: Implement (or don't) OS Lock and DoubleLock properly

Message ID 20220630194116.3438513-1-peter.maydell@linaro.org
Headers show
Series target/arm: Implement (or don't) OS Lock and DoubleLock properly | expand

Message

Peter Maydell June 30, 2022, 7:41 p.m. UTC 
Continuing in my series of filling in bits of the architecture
that probably nobody much cares about, this series fixes up
Feat_DoubleLock. DoubleLock is a part of the debug architecture
which allows a guest OS to suppress debug exceptions while
it is powering down a CPU so that they don't cause updates to
bits of debug register state that then don't get preserved
across the power-down/up sequence. The reason for looking
at QEMU's support here is that recent versions of the architecture
define that the feature becomes first optional (after v8.2 or
so) and then mustn't be implemented at all at v9.

We have only ever implemented this by NOPing the OSDLR_EL1 register,
which is not correct for either the "implement the feature"
or the "don't implement the feature" case. What is supposed
to happen is that if the feature is implemented then there is
one writable bit which is set to 1 to suppress debug exceptions,
and if the feature is not implemented then the bit is RAZ/WI.
We also don't properly implement the related OS Lock which
does something very similar. There we correctly implemented
the register reading and writing parts but didn't make the
bit do anything.

The series starts with some code movement, while I was messing
with the debug code, shifting 500 lines of debug related code
out of the massive helper.c and into debug_helper.c. Patch 2
is big but almost entirely pure code motion (best reviewed with
git's --color-moved support). I think this helps in our
ongoing quest to make helper.c less of a massive grabbag
of miscellaneous things.

Patch 3 implements the required behaviour of the OS Lock
(which turns out to be very easy).

Patch 4 adds support for some AArch32 debug ID registers we
turn out to be missing. Clearly nobody was trying to read these,
but one of them is where the field for "is FEAT_DoubleLock
present" is kept, so we need the data internally.

Finally, patch 5 fixes the implementation of OSDLR_EL1 to
either be RAZ/WI or to have a bit that has the required
suppress-debug-exceptions behaviour.

thanks
-- PMM

Peter Maydell (5):
  target/arm: Fix code style issues in debug helper functions
  target/arm: Move define_debug_regs() to debug_helper.c
  target/arm: Suppress debug exceptions when OS Lock set
  target/arm: Implement AArch32 DBGDEVID, DBGDEVID1, DBGDEVID2
  target/arm: Correctly implement Feat_DoubleLock

 target/arm/cpregs.h       |   3 +
 target/arm/cpu.h          |  43 +++
 target/arm/internals.h    |   9 +
 target/arm/cpu64.c        |   6 +
 target/arm/cpu_tcg.c      |   6 +
 target/arm/debug_helper.c | 577 ++++++++++++++++++++++++++++++++++++++
 target/arm/helper.c       | 513 +--------------------------------
 7 files changed, 645 insertions(+), 512 deletions(-)