Open lsf37 opened 2 years ago
Occasionally the TK1
board also fails the same test (for the same config, i.e. TK1_debug_hyp_MCS_clang_32
).
It looks like something is definitely up there.
Comment from Indan on that thread linked above about what might be wrong:
Running test CACHEFLUSH0001 (Test a cache maintenance on pages) *ptrc == 0xBEEFCAFE at line 104 of file /github/workspace/projects/sel4test/apps/sel4test-tests/src/tests/cache.c
I'm not convinced that test_page_flush is correct: It relies on no cache miss or prefetching happening between the seL4_ARM_Page_Invalidate_Data call and the *ptr write. I don't think that's guaranteed.
In case the invalidation performs an implicit clean, the result is the same as for seL4_ARM_Page_CleanInvalidate_Data above, which is already tested.
This seems a duplicate/extension of #65 btw.
In config
ODROID_XU4_debug_hyp_MCS_clang_32
(and apparently only this config and only on clang), ODROID_XU4 is failing the testCACHEFLUSH0001
. The configuration is disabled for tests for now.It looks like this might be a problem with the test setup, not with MCS.
See also discussion on seL4/seL4#877.