Split out from 

v1: https://lore.kernel.org/qemu-devel/20220906101747.344559-1-richard.henderson@linaro.org/
v4: https://lore.kernel.org/qemu-devel/20230220184052.163465-1-richard.henderson@linaro.org/

after I noticed that the testcase,

https://lore.kernel.org/qemu-devel/20221130174610.434590-2-iii@linux.ibm.com/

does not actually work under KVM.

In fact, support for PER_CR9_EVENT_BRANCH in TCG is broken.
This cleans it up and improves things so that when only one PER bit
is set (what I assume to be the normal state of afairs), we have
less overhead.

I'm not sure that all of the interactions are correct, particularly
with multiple PER triggers or other exceptions.  For that, we'd need
a more complex test case.  But at least branches match hardware now.


r~


Richard Henderson (14):
  target/s390x: Do not use unwind for per_check_exception
  target/s390x: Move cpu_get_tb_cpu_state out of line
  target/s390x: Update CR9 bits
  target/s390x: Record separate PER bits in TB flags
  target/s390x: Disable conditional branch-to-next for PER
  target/s390x: Introduce help_goto_indirect
  target/s390x: Simplify help_branch
  target/s390x: Split per_breaking_event from per_branch_*
  target/s390x: Raise exception from helper_per_branch
  target/s390x: Raise exception from per_store_real
  target/s390x: Fix helper_per_ifetch flags
  target/s390x: Simplify per_ifetch, per_check_exception
  target/s390x: Adjust check of noreturn in translate_one
  tests/tcg/s390x: Add per.S

 target/s390x/cpu.h                      |  83 ++++----
 target/s390x/helper.h                   |   8 +-
 target/s390x/cpu.c                      |  36 ++++
 target/s390x/tcg/excp_helper.c          |   2 +-
 target/s390x/tcg/misc_helper.c          |  68 ++++---
 target/s390x/tcg/translate.c            | 248 +++++++++---------------
 tests/tcg/s390x/Makefile.softmmu-target |   1 +
 tests/tcg/s390x/per.S                   |  82 ++++++++
 8 files changed, 302 insertions(+), 226 deletions(-)
 create mode 100644 tests/tcg/s390x/per.S

-- 
2.34.1