| Message ID | 1621950703-1532-1-git-send-email-atrajeev@linux.vnet.ibm.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | selftests/powerpc: Updates to EBB selftest for ISA v3.1 | expand |
> On 25-May-2021, at 7:21 PM, Athira Rajeev <atrajeev@linux.vnet.ibm.com> wrote: > > The "no_handler_test" in ebb selftests attempts to read the PMU > registers after closing of the event via helper function > "dump_ebb_state". With the MMCR0 control bit (PMCCEXT) in ISA v3.1, > read access to group B registers is restricted when MMCR0 PMCC=0b00. > Hence the call to dump_ebb_state after closing of event will generate > a SIGILL, which is expected. > > Test has below in logs: > <<>> > !! child died by signal 4 > failure: no_handler_test > <<>> > Tested patches on Power9 and Power10 - PowerVM environment. Power9 - no_handler_test - with and without patch - success Power10 - no_handler_test - without patch - failure Power10 - no_handler_test - with patch - success Power9 - regs_access_pmccext_test - with patch - success Power10 - regs_access_pmccext_test - with patch - success Tested-by: Nageswara R Sastry <rnsastry@linux.ibm.com <mailto:rnsastry@linux.ibm.com>> Thank you.
On Tue, 25 May 2021 09:51:41 -0400, Athira Rajeev wrote: > The "no_handler_test" in ebb selftests attempts to read the PMU > registers after closing of the event via helper function > "dump_ebb_state". With the MMCR0 control bit (PMCCEXT) in ISA v3.1, > read access to group B registers is restricted when MMCR0 PMCC=0b00. > Hence the call to dump_ebb_state after closing of event will generate > a SIGILL, which is expected. > > [...] Applied to powerpc/next. [1/2] selftests/powerpc: Fix "no_handler" EBB selftest https://git.kernel.org/powerpc/c/45677c9aebe926192e59475b35a1ff35ff2d4217 [2/2] selftests/powerpc: EBB selftest for MMCR0 control for PMU SPRs in ISA v3.1 https://git.kernel.org/powerpc/c/d81090ed44c0d15abf2b07663d5f0b9e5ba51525 cheers
The "no_handler_test" in ebb selftests attempts to read the PMU registers after closing of the event via helper function "dump_ebb_state". With the MMCR0 control bit (PMCCEXT) in ISA v3.1, read access to group B registers is restricted when MMCR0 PMCC=0b00. Hence the call to dump_ebb_state after closing of event will generate a SIGILL, which is expected. Test has below in logs: <<>> !! child died by signal 4 failure: no_handler_test <<>> In other platforms (like power9), the older behaviour works where group B PMU SPRs are readable. The "dump_ebb_state" is called twice in the test. The second call after closing of event was done inorder to dump state of registers when the counters are frozen. But since the counters should already be frozen by the time first dump is done, patch1 drops the second call to "dump_ebb_state". To address the new sigill behaviour in ISA v3.1, patch2 creates a separate selftest. Changelog: v2 -> v3: Fixed a space issue in patch2. v1 -> v2: Addressed review comments from Michael Ellerman. First version attempted to address the SIGILL behaviour in existing "no_handler_test" test itself. As per mpe's suggestion, moved that to a separate test and removed the second call to "dump_ebb_state" since that is actually not needed. Athira Rajeev (2): selftests/powerpc: Fix "no_handler" EBB selftest selftests/powerpc: EBB selftest for MMCR0 control for PMU SPRs in ISA v3.1 tools/testing/selftests/powerpc/pmu/ebb/Makefile | 2 +- .../selftests/powerpc/pmu/ebb/no_handler_test.c | 2 - .../powerpc/pmu/ebb/regs_access_pmccext_test.c | 63 ++++++++++++++++++++++ 3 files changed, 64 insertions(+), 3 deletions(-) create mode 100644 tools/testing/selftests/powerpc/pmu/ebb/regs_access_pmccext_test.c